From c6e67f5a9373e916a8d2333585cb5787aa5f7bb7 Mon Sep 17 00:00:00 2001 From: Tim Rowley Date: Tue, 16 Feb 2016 17:28:09 -0600 Subject: [PATCH] gallium/swr: add OpenSWR rasterizer Acked-by: Roland Scheidegger Acked-by: Jose Fonseca --- .../drivers/swr/rasterizer/common/containers.hpp | 208 + .../drivers/swr/rasterizer/common/formats.cpp | 5469 ++++++++++++++++++++ .../drivers/swr/rasterizer/common/formats.h | 251 + src/gallium/drivers/swr/rasterizer/common/isa.hpp | 235 + src/gallium/drivers/swr/rasterizer/common/os.h | 221 + .../swr/rasterizer/common/rdtsc_buckets.cpp | 188 + .../drivers/swr/rasterizer/common/rdtsc_buckets.h | 229 + .../swr/rasterizer/common/rdtsc_buckets_shared.h | 167 + .../drivers/swr/rasterizer/common/simdintrin.h | 787 +++ .../drivers/swr/rasterizer/common/swr_assert.cpp | 238 + .../drivers/swr/rasterizer/common/swr_assert.h | 109 + src/gallium/drivers/swr/rasterizer/core/api.cpp | 1511 ++++++ src/gallium/drivers/swr/rasterizer/core/api.h | 500 ++ src/gallium/drivers/swr/rasterizer/core/arena.cpp | 166 + src/gallium/drivers/swr/rasterizer/core/arena.h | 69 + .../drivers/swr/rasterizer/core/backend.cpp | 1899 +++++++ src/gallium/drivers/swr/rasterizer/core/backend.h | 59 + src/gallium/drivers/swr/rasterizer/core/blend.h | 318 ++ src/gallium/drivers/swr/rasterizer/core/clip.cpp | 201 + src/gallium/drivers/swr/rasterizer/core/clip.h | 868 ++++ src/gallium/drivers/swr/rasterizer/core/context.h | 495 ++ .../drivers/swr/rasterizer/core/depthstencil.h | 245 + src/gallium/drivers/swr/rasterizer/core/fifo.hpp | 136 + .../swr/rasterizer/core/format_conversion.h | 196 + .../drivers/swr/rasterizer/core/format_traits.h | 3548 +++++++++++++ .../drivers/swr/rasterizer/core/format_types.h | 1075 ++++ .../drivers/swr/rasterizer/core/frontend.cpp | 2345 +++++++++ src/gallium/drivers/swr/rasterizer/core/frontend.h | 327 ++ src/gallium/drivers/swr/rasterizer/core/knobs.h | 142 + .../drivers/swr/rasterizer/core/knobs_init.h | 98 + .../drivers/swr/rasterizer/core/multisample.cpp | 51 + .../drivers/swr/rasterizer/core/multisample.h | 620 +++ src/gallium/drivers/swr/rasterizer/core/pa.h | 1208 +++++ src/gallium/drivers/swr/rasterizer/core/pa_avx.cpp | 1177 +++++ .../drivers/swr/rasterizer/core/rasterizer.cpp | 1393 +++++ .../drivers/swr/rasterizer/core/rasterizer.h | 35 + .../drivers/swr/rasterizer/core/rdtsc_core.cpp | 91 + .../drivers/swr/rasterizer/core/rdtsc_core.h | 177 + src/gallium/drivers/swr/rasterizer/core/state.h | 1027 ++++ .../drivers/swr/rasterizer/core/tessellator.h | 88 + .../drivers/swr/rasterizer/core/threads.cpp | 962 ++++ src/gallium/drivers/swr/rasterizer/core/threads.h | 63 + .../drivers/swr/rasterizer/core/tilemgr.cpp | 105 + src/gallium/drivers/swr/rasterizer/core/tilemgr.h | 390 ++ src/gallium/drivers/swr/rasterizer/core/utils.cpp | 148 + src/gallium/drivers/swr/rasterizer/core/utils.h | 831 +++ .../drivers/swr/rasterizer/jitter/JitManager.cpp | 313 ++ .../drivers/swr/rasterizer/jitter/JitManager.h | 186 + .../drivers/swr/rasterizer/jitter/blend_jit.cpp | 772 +++ .../drivers/swr/rasterizer/jitter/blend_jit.h | 93 + .../drivers/swr/rasterizer/jitter/builder.cpp | 71 + .../drivers/swr/rasterizer/jitter/builder.h | 71 + .../drivers/swr/rasterizer/jitter/builder_math.h | 34 + .../drivers/swr/rasterizer/jitter/builder_misc.cpp | 1447 ++++++ .../drivers/swr/rasterizer/jitter/builder_misc.h | 149 + .../drivers/swr/rasterizer/jitter/fetch_jit.cpp | 1431 +++++ .../drivers/swr/rasterizer/jitter/fetch_jit.h | 128 + .../drivers/swr/rasterizer/jitter/jit_api.h | 108 + .../jitter/scripts/gen_llvm_ir_macros.py | 401 ++ .../rasterizer/jitter/scripts/gen_llvm_types.py | 341 ++ .../swr/rasterizer/jitter/streamout_jit.cpp | 357 ++ .../drivers/swr/rasterizer/jitter/streamout_jit.h | 94 + .../drivers/swr/rasterizer/memory/ClearTile.cpp | 287 + .../drivers/swr/rasterizer/memory/Convert.h | 698 +++ .../drivers/swr/rasterizer/memory/LoadTile.cpp | 396 ++ .../drivers/swr/rasterizer/memory/StoreTile.cpp | 1717 ++++++ .../swr/rasterizer/memory/TilingFunctions.h | 581 +++ .../drivers/swr/rasterizer/memory/tilingtraits.h | 263 + .../drivers/swr/rasterizer/scripts/gen_knobs.py | 79 + .../drivers/swr/rasterizer/scripts/knob_defs.py | 226 + .../swr/rasterizer/scripts/mako/__init__.py | 8 + .../swr/rasterizer/scripts/mako/_ast_util.py | 845 +++ .../drivers/swr/rasterizer/scripts/mako/ast.py | 178 + .../drivers/swr/rasterizer/scripts/mako/cache.py | 238 + .../drivers/swr/rasterizer/scripts/mako/cmd.py | 62 + .../drivers/swr/rasterizer/scripts/mako/codegen.py | 1237 +++++ .../drivers/swr/rasterizer/scripts/mako/compat.py | 174 + .../swr/rasterizer/scripts/mako/exceptions.py | 373 ++ .../drivers/swr/rasterizer/scripts/mako/filters.py | 201 + .../drivers/swr/rasterizer/scripts/mako/lexer.py | 441 ++ .../drivers/swr/rasterizer/scripts/mako/lookup.py | 359 ++ .../swr/rasterizer/scripts/mako/parsetree.py | 594 +++ .../drivers/swr/rasterizer/scripts/mako/pygen.py | 299 ++ .../swr/rasterizer/scripts/mako/pyparser.py | 232 + .../drivers/swr/rasterizer/scripts/mako/runtime.py | 878 ++++ .../swr/rasterizer/scripts/mako/template.py | 705 +++ .../drivers/swr/rasterizer/scripts/mako/util.py | 360 ++ .../rasterizer/scripts/templates/knobs.template | 141 + 88 files changed, 48234 insertions(+) create mode 100644 src/gallium/drivers/swr/rasterizer/common/containers.hpp create mode 100644 src/gallium/drivers/swr/rasterizer/common/formats.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/common/formats.h create mode 100644 src/gallium/drivers/swr/rasterizer/common/isa.hpp create mode 100644 src/gallium/drivers/swr/rasterizer/common/os.h create mode 100644 src/gallium/drivers/swr/rasterizer/common/rdtsc_buckets.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/common/rdtsc_buckets.h create mode 100644 src/gallium/drivers/swr/rasterizer/common/rdtsc_buckets_shared.h create mode 100644 src/gallium/drivers/swr/rasterizer/common/simdintrin.h create mode 100644 src/gallium/drivers/swr/rasterizer/common/swr_assert.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/common/swr_assert.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/api.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/core/api.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/arena.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/core/arena.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/backend.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/core/backend.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/blend.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/clip.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/core/clip.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/context.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/depthstencil.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/fifo.hpp create mode 100644 src/gallium/drivers/swr/rasterizer/core/format_conversion.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/format_traits.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/format_types.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/frontend.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/core/frontend.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/knobs.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/knobs_init.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/multisample.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/core/multisample.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/pa.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/pa_avx.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/core/rasterizer.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/core/rasterizer.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/rdtsc_core.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/core/rdtsc_core.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/state.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/tessellator.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/threads.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/core/threads.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/tilemgr.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/core/tilemgr.h create mode 100644 src/gallium/drivers/swr/rasterizer/core/utils.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/core/utils.h create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/JitManager.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/JitManager.h create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/blend_jit.h create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/builder.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/builder.h create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/builder_math.h create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.h create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/jit_api.h create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/scripts/gen_llvm_ir_macros.py create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/scripts/gen_llvm_types.py create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.h create mode 100644 src/gallium/drivers/swr/rasterizer/memory/ClearTile.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/memory/Convert.h create mode 100644 src/gallium/drivers/swr/rasterizer/memory/LoadTile.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/memory/StoreTile.cpp create mode 100644 src/gallium/drivers/swr/rasterizer/memory/TilingFunctions.h create mode 100644 src/gallium/drivers/swr/rasterizer/memory/tilingtraits.h create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/gen_knobs.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/knob_defs.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/__init__.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/_ast_util.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/ast.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/cache.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/cmd.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/codegen.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/compat.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/exceptions.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/filters.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/lexer.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/lookup.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/parsetree.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/pygen.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/pyparser.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/runtime.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/template.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/mako/util.py create mode 100644 src/gallium/drivers/swr/rasterizer/scripts/templates/knobs.template diff --git a/src/gallium/drivers/swr/rasterizer/common/containers.hpp b/src/gallium/drivers/swr/rasterizer/common/containers.hpp new file mode 100644 index 00000000000..bc96c5f62fd --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/common/containers.hpp @@ -0,0 +1,208 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +****************************************************************************/ + +#ifndef SWRLIB_CONTAINERS_HPP__ +#define SWRLIB_CONTAINERS_HPP__ + +#include +#include "common/os.h" + +namespace SWRL +{ + +template +struct UncheckedFixedVector +{ + UncheckedFixedVector() : mSize(0) + { + } + + UncheckedFixedVector(std::size_t size, T const& exemplar) + { + this->mSize = 0; + for (std::size_t i = 0; i < size; ++i) + this->push_back(exemplar); + } + + template + UncheckedFixedVector(Iter fst, Iter lst) + { + this->mSize = 0; + for ( ; fst != lst; ++fst) + this->push_back(*fst); + } + + UncheckedFixedVector(UncheckedFixedVector const& UFV) + { + this->mSize = 0; + for (std::size_t i = 0, N = UFV.size(); i < N; ++i) + (*this)[i] = UFV[i]; + this->mSize = UFV.size(); + } + + UncheckedFixedVector& operator=(UncheckedFixedVector const& UFV) + { + for (std::size_t i = 0, N = UFV.size(); i < N; ++i) + (*this)[i] = UFV[i]; + this->mSize = UFV.size(); + return *this; + } + + T* begin() { return &this->mElements[0]; } + T* end() { return &this->mElements[0] + this->mSize; } + T const* begin() const { return &this->mElements[0]; } + T const* end() const { return &this->mElements[0] + this->mSize; } + + friend bool operator==(UncheckedFixedVector const& L, UncheckedFixedVector const& R) + { + if (L.size() != R.size()) return false; + for (std::size_t i = 0, N = L.size(); i < N; ++i) + { + if (L[i] != R[i]) return false; + } + return true; + } + + friend bool operator!=(UncheckedFixedVector const& L, UncheckedFixedVector const& R) + { + if (L.size() != R.size()) return true; + for (std::size_t i = 0, N = L.size(); i < N; ++i) + { + if (L[i] != R[i]) return true; + } + return false; + } + + T& operator[](std::size_t idx) + { + return this->mElements[idx]; + } + T const& operator[](std::size_t idx) const + { + return this->mElements[idx]; + } + void push_back(T const& t) + { + this->mElements[this->mSize] = t; + ++this->mSize; + } + void pop_back() + { + SWR_ASSERT(this->mSize > 0); + --this->mSize; + } + T& back() + { + return this->mElements[this->mSize-1]; + } + T const& back() const + { + return this->mElements[this->mSize-1]; + } + bool empty() const + { + return this->mSize == 0; + } + std::size_t size() const + { + return this->mSize; + } + void resize(std::size_t sz) + { + this->mSize = sz; + } + void clear() + { + this->resize(0); + } +private: + std::size_t mSize; + T mElements[NUM_ELEMENTS]; +}; + +template +struct FixedStack : UncheckedFixedVector +{ + FixedStack() {} + + void push(T const& t) + { + this->push_back(t); + } + + void pop() + { + this->pop_back(); + } + + T& top() + { + return this->back(); + } + + T const& top() const + { + return this->back(); + } +}; + +template +struct CRCHash +{ + static_assert((sizeof(T) % sizeof(UINT)) == 0, "CRCHash expects templated type size is even multiple of 4B"); + UINT operator()(const T& k) const + { + UINT *pData = (UINT*)&k; + UINT crc = 0; + for (UINT i = 0; i < sizeof(T) / sizeof(UINT); ++i) + { + crc = _mm_crc32_u32(crc, pData[i]); + } + return crc; + } +}; + +}// end SWRL + +namespace std +{ + +template +struct hash> +{ + size_t operator() (SWRL::UncheckedFixedVector const& v) const + { + if (v.size() == 0) return 0; + std::hash H; + size_t x = H(v[0]); + if (v.size() == 1) return x; + for (size_t i = 1; i < v.size(); ++i) + x ^= H(v[i]) + 0x9e3779b9 + (x<<6) + (x>>2); + return x; + } +}; + + +}// end std. + +#endif//SWRLIB_CONTAINERS_HPP__ diff --git a/src/gallium/drivers/swr/rasterizer/common/formats.cpp b/src/gallium/drivers/swr/rasterizer/common/formats.cpp new file mode 100644 index 00000000000..ed8ce7e5b0f --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/common/formats.cpp @@ -0,0 +1,5469 @@ + +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file formats.cpp +* +* @brief auto-generated file +* +* DO NOT EDIT +* +******************************************************************************/ + +#include "formats.h" + +// lookup table for unorm8 srgb -> float conversion +const uint32_t srgb8Table[256] = { + 0x00000000, 0x399f22b4, 0x3a1f22b4, 0x3a6eb40f, 0x3a9f22b4, 0x3ac6eb61, 0x3aeeb40f, 0x3b0b3e5e, 0x3b1f22b4, 0x3b33070b, 0x3b46eb61, 0x3b5b518d, 0x3b70f18d, 0x3b83e1c6, 0x3b8fe616, 0x3b9c87fd, + 0x3ba9c9b5, 0x3bb7ad6f, 0x3bc63549, 0x3bd5635f, 0x3be539c1, 0x3bf5ba70, 0x3c0373b5, 0x3c0c6152, 0x3c15a703, 0x3c1f45be, 0x3c293e6b, 0x3c3391f7, 0x3c3e4149, 0x3c494d43, 0x3c54b6c7, 0x3c607eb1, + 0x3c6ca5dc, 0x3c792d22, 0x3c830aa8, 0x3c89af9f, 0x3c9085db, 0x3c978dc5, 0x3c9ec7c0, 0x3ca63431, 0x3cadd37d, 0x3cb5a601, 0x3cbdac20, 0x3cc5e639, 0x3cce54ab, 0x3cd6f7d3, 0x3cdfd00e, 0x3ce8ddb9, + 0x3cf22131, 0x3cfb9ac6, 0x3d02a56c, 0x3d0798df, 0x3d0ca7e7, 0x3d11d2b0, 0x3d171965, 0x3d1c7c31, 0x3d21fb3c, 0x3d2796b2, 0x3d2d4ebe, 0x3d332384, 0x3d39152e, 0x3d3f23e6, 0x3d454fd4, 0x3d4b991f, + 0x3d51ffef, 0x3d58846a, 0x3d5f26b7, 0x3d65e6fe, 0x3d6cc564, 0x3d73c20f, 0x3d7add25, 0x3d810b66, 0x3d84b795, 0x3d887330, 0x3d8c3e4a, 0x3d9018f6, 0x3d940345, 0x3d97fd4a, 0x3d9c0716, 0x3da020bb, + 0x3da44a4b, 0x3da883d7, 0x3daccd70, 0x3db12728, 0x3db59110, 0x3dba0b38, 0x3dbe95b5, 0x3dc33092, 0x3dc7dbe2, 0x3dcc97b6, 0x3dd1641f, 0x3dd6412c, 0x3ddb2eef, 0x3de02d77, 0x3de53cd5, 0x3dea5d19, + 0x3def8e55, 0x3df4d093, 0x3dfa23e8, 0x3dff8861, 0x3e027f07, 0x3e054282, 0x3e080ea5, 0x3e0ae379, 0x3e0dc107, 0x3e10a755, 0x3e13966c, 0x3e168e53, 0x3e198f11, 0x3e1c98ae, 0x3e1fab32, 0x3e22c6a3, + 0x3e25eb09, 0x3e29186c, 0x3e2c4ed2, 0x3e2f8e45, 0x3e32d6c8, 0x3e362865, 0x3e398322, 0x3e3ce706, 0x3e405419, 0x3e43ca62, 0x3e4749e8, 0x3e4ad2b1, 0x3e4e64c6, 0x3e52002b, 0x3e55a4e9, 0x3e595307, + 0x3e5d0a8b, 0x3e60cb7c, 0x3e6495e0, 0x3e6869bf, 0x3e6c4720, 0x3e702e08, 0x3e741e7f, 0x3e78188c, 0x3e7c1c38, 0x3e8014c2, 0x3e82203c, 0x3e84308d, 0x3e8645ba, 0x3e885fc5, 0x3e8a7eb2, 0x3e8ca283, + 0x3e8ecb3d, 0x3e90f8e1, 0x3e932b74, 0x3e9562f8, 0x3e979f71, 0x3e99e0e2, 0x3e9c274e, 0x3e9e72b7, 0x3ea0c322, 0x3ea31892, 0x3ea57308, 0x3ea7d289, 0x3eaa3718, 0x3eaca0b7, 0x3eaf0f69, 0x3eb18333, + 0x3eb3fc16, 0x3eb67a15, 0x3eb8fd34, 0x3ebb8576, 0x3ebe12e1, 0x3ec0a571, 0x3ec33d2d, 0x3ec5da17, 0x3ec87c33, 0x3ecb2383, 0x3ecdd00b, 0x3ed081cd, 0x3ed338cc, 0x3ed5f50b, 0x3ed8b68d, 0x3edb7d54, + 0x3ede4965, 0x3ee11ac1, 0x3ee3f16b, 0x3ee6cd67, 0x3ee9aeb6, 0x3eec955d, 0x3eef815d, 0x3ef272ba, 0x3ef56976, 0x3ef86594, 0x3efb6717, 0x3efe6e02, 0x3f00bd2b, 0x3f02460c, 0x3f03d1a5, 0x3f055ff8, + 0x3f06f106, 0x3f0884cf, 0x3f0a1b57, 0x3f0bb49d, 0x3f0d50a2, 0x3f0eef69, 0x3f1090f2, 0x3f123540, 0x3f13dc53, 0x3f15862d, 0x3f1732cf, 0x3f18e23b, 0x3f1a9471, 0x3f1c4973, 0x3f1e0143, 0x3f1fbbe1, + 0x3f217950, 0x3f23398f, 0x3f24fca2, 0x3f26c288, 0x3f288b43, 0x3f2a56d5, 0x3f2c253f, 0x3f2df681, 0x3f2fca9e, 0x3f31a197, 0x3f337b6c, 0x3f355820, 0x3f3737b3, 0x3f391a26, 0x3f3aff7e, 0x3f3ce7b7, + 0x3f3ed2d4, 0x3f40c0d6, 0x3f42b1c0, 0x3f44a592, 0x3f469c4d, 0x3f4895f3, 0x3f4a9284, 0x3f4c9203, 0x3f4e9470, 0x3f5099cd, 0x3f52a21a, 0x3f54ad59, 0x3f56bb8c, 0x3f58ccb3, 0x3f5ae0cf, 0x3f5cf7e2, + 0x3f5f11ee, 0x3f612ef2, 0x3f634eef, 0x3f6571ec, 0x3f6797e1, 0x3f69c0d8, 0x3f6beccb, 0x3f6e1bc2, 0x3f704db6, 0x3f7282b1, 0x3f74baae, 0x3f76f5b3, 0x3f7933b9, 0x3f7b74cb, 0x3f7db8e0, 0x3f800000, +}; + +// order must match SWR_FORMAT +const SWR_FORMAT_INFO gFormatInfo[] = { + // R32G32B32A32_FLOAT (0x0) + { + "R32G32B32A32_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_FLOAT }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 32, 32, 32, 32 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32G32B32A32_SINT (0x1) + { + "R32G32B32A32_SINT", + { SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_SINT }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 32, 32, 32, 32 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32G32B32A32_UINT (0x2) + { + "R32G32B32A32_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 32, 32, 32, 32 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x3 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x4 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x5 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R32G32B32X32_FLOAT (0x6) + { + "R32G32B32X32_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNUSED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 32, 32, 32, 32 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32G32B32A32_SSCALED (0x7) + { + "R32G32B32A32_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_SSCALED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 32, 32, 32, 32 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32G32B32A32_USCALED (0x8) + { + "R32G32B32A32_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 32, 32, 32, 32 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x9 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xa (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xb (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xc (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xd (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xe (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xf (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x10 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x11 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x12 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x13 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x14 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x15 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x16 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x17 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x18 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x19 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1a (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x20 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x21 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x22 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x23 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x24 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x25 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x26 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x27 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x28 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x29 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x2a (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x2b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x2c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x2d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x2e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x2f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x30 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x31 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x32 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x33 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x34 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x35 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x36 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x37 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x38 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x39 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x3a (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x3b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x3c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x3d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x3e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x3f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R32G32B32_FLOAT (0x40) + { + "R32G32B32_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 32, 32, 32, 0 }, // Bits per component + 96, // Bits per element + 12, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32G32B32_SINT (0x41) + { + "R32G32B32_SINT", + { SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 32, 32, 32, 0 }, // Bits per component + 96, // Bits per element + 12, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32G32B32_UINT (0x42) + { + "R32G32B32_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 32, 32, 32, 0 }, // Bits per component + 96, // Bits per element + 12, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x43 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x44 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R32G32B32_SSCALED (0x45) + { + "R32G32B32_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 32, 32, 32, 0 }, // Bits per component + 96, // Bits per element + 12, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32G32B32_USCALED (0x46) + { + "R32G32B32_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 32, 32, 32, 0 }, // Bits per component + 96, // Bits per element + 12, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x47 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x48 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x49 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x4a (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x4b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x4c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x4d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x4e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x4f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x50 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x51 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x52 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x53 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x54 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x55 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x56 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x57 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x58 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x59 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x5a (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x5b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x5c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x5d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x5e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x5f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x60 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x61 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x62 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x63 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x64 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x65 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x66 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x67 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x68 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x69 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x6a (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x6b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x6c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x6d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x6e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x6f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x70 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x71 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x72 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x73 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x74 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x75 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x76 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x77 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x78 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x79 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x7a (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x7b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x7c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x7d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x7e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x7f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R16G16B16A16_UNORM (0x80) + { + "R16G16B16A16_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 16, 16, 16, 16 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 65535.0f, 1.0f / 65535.0f, 1.0f / 65535.0f, 1.0f / 65535.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16B16A16_SNORM (0x81) + { + "R16G16B16A16_SNORM", + { SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_SNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 16, 16, 16, 16 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 32767.0f, 1.0f / 32767.0f, 1.0f / 32767.0f, 1.0f / 32767.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16B16A16_SINT (0x82) + { + "R16G16B16A16_SINT", + { SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_SINT }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 16, 16, 16, 16 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16B16A16_UINT (0x83) + { + "R16G16B16A16_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 16, 16, 16, 16 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16B16A16_FLOAT (0x84) + { + "R16G16B16A16_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_FLOAT }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 16, 16, 16, 16 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32G32_FLOAT (0x85) + { + "R32G32_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 32, 32, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32G32_SINT (0x86) + { + "R32G32_SINT", + { SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 32, 32, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32G32_UINT (0x87) + { + "R32G32_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 32, 32, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32_FLOAT_X8X24_TYPELESS (0x88) + { + "R32_FLOAT_X8X24_TYPELESS", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 32, 32, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // X32_TYPELESS_G8X24_UINT (0x89) + { + "X32_TYPELESS_G8X24_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 32, 32, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // L32A32_FLOAT (0x8a) + { + "L32A32_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 3, 0, 0 }, // Swizzle + { 32, 32, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // 0x8b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x8c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x8d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R16G16B16X16_UNORM (0x8e) + { + "R16G16B16X16_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNUSED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 16, 16, 16, 16 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 65535.0f, 1.0f / 65535.0f, 1.0f / 65535.0f, 1.0f / 65535.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16B16X16_FLOAT (0x8f) + { + "R16G16B16X16_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNUSED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 16, 16, 16, 16 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x90 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // L32X32_FLOAT (0x91) + { + "L32X32_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 3, 0, 0 }, // Swizzle + { 32, 32, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // I32X32_FLOAT (0x92) + { + "I32X32_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 3, 0, 0 }, // Swizzle + { 32, 32, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // R16G16B16A16_SSCALED (0x93) + { + "R16G16B16A16_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_SSCALED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 16, 16, 16, 16 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16B16A16_USCALED (0x94) + { + "R16G16B16A16_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 16, 16, 16, 16 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32G32_SSCALED (0x95) + { + "R32G32_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 32, 32, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32G32_USCALED (0x96) + { + "R32G32_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 32, 32, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x97 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R32_FLOAT_X8X24_TYPELESS_LD (0x98) + { + "R32_FLOAT_X8X24_TYPELESS_LD", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 32, 32, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x99 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x9a (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x9b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x9c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x9d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x9e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x9f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xa0 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xa1 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xa2 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xa3 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xa4 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xa5 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xa6 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xa7 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xa8 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xa9 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xaa (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xab (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xac (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xad (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xae (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xaf (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xb0 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xb1 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xb2 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xb3 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xb4 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xb5 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xb6 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xb7 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xb8 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xb9 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xba (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xbb (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xbc (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xbd (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xbe (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xbf (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // B8G8R8A8_UNORM (0xc0) + { + "B8G8R8A8_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B8G8R8A8_UNORM_SRGB (0xc1) + { + "B8G8R8A8_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + true, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R10G10B10A2_UNORM (0xc2) + { + "R10G10B10A2_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 1023.0f, 1.0f / 1023.0f, 1.0f / 1023.0f, 1.0f / 3.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R10G10B10A2_UNORM_SRGB (0xc3) + { + "R10G10B10A2_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + true, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 1023.0f, 1.0f / 1023.0f, 1.0f / 1023.0f, 1.0f / 3.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R10G10B10A2_UINT (0xc4) + { + "R10G10B10A2_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0xc5 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xc6 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R8G8B8A8_UNORM (0xc7) + { + "R8G8B8A8_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8B8A8_UNORM_SRGB (0xc8) + { + "R8G8B8A8_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + true, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8B8A8_SNORM (0xc9) + { + "R8G8B8A8_SNORM", + { SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_SNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 127.0f, 1.0f / 127.0f, 1.0f / 127.0f, 1.0f / 127.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8B8A8_SINT (0xca) + { + "R8G8B8A8_SINT", + { SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_SINT }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8B8A8_UINT (0xcb) + { + "R8G8B8A8_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16_UNORM (0xcc) + { + "R16G16_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 16, 16, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 65535.0f, 1.0f / 65535.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16_SNORM (0xcd) + { + "R16G16_SNORM", + { SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 16, 16, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 32767.0f, 1.0f / 32767.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16_SINT (0xce) + { + "R16G16_SINT", + { SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 16, 16, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16_UINT (0xcf) + { + "R16G16_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 16, 16, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16_FLOAT (0xd0) + { + "R16G16_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 16, 16, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B10G10R10A2_UNORM (0xd1) + { + "B10G10R10A2_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 1023.0f, 1.0f / 1023.0f, 1.0f / 1023.0f, 1.0f / 3.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B10G10R10A2_UNORM_SRGB (0xd2) + { + "B10G10R10A2_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + true, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 1023.0f, 1.0f / 1023.0f, 1.0f / 1023.0f, 1.0f / 3.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R11G11B10_FLOAT (0xd3) + { + "R11G11B10_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 11, 11, 10, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0xd4 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xd5 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R32_SINT (0xd6) + { + "R32_SINT", + { SWR_TYPE_SINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 32, 0, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32_UINT (0xd7) + { + "R32_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 32, 0, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32_FLOAT (0xd8) + { + "R32_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 32, 0, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R24_UNORM_X8_TYPELESS (0xd9) + { + "R24_UNORM_X8_TYPELESS", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 24, 0, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 16777215.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0xda (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xdb (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R24_UNORM_X8_TYPELESS_LD (0xdc) + { + "R24_UNORM_X8_TYPELESS_LD", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 24, 0, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 16777215.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // L32_UNORM (0xdd) + { + "L32_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 32, 0, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 4294967295.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // 0xde (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // L16A16_UNORM (0xdf) + { + "L16A16_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 3, 0, 0 }, // Swizzle + { 16, 16, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 65535.0f, 1.0f / 65535.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // I24X8_UNORM (0xe0) + { + "I24X8_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 3, 0, 0 }, // Swizzle + { 24, 8, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 16777215.0f, 1.0f / 255.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // L24X8_UNORM (0xe1) + { + "L24X8_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 3, 0, 0 }, // Swizzle + { 24, 8, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 16777215.0f, 1.0f / 255.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // 0xe2 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // I32_FLOAT (0xe3) + { + "I32_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 32, 0, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // L32_FLOAT (0xe4) + { + "L32_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 32, 0, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // A32_FLOAT (0xe5) + { + "A32_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 3, 0, 0, 0 }, // Swizzle + { 32, 0, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0xe6 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xe7 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xe8 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // B8G8R8X8_UNORM (0xe9) + { + "B8G8R8X8_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNUSED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B8G8R8X8_UNORM_SRGB (0xea) + { + "B8G8R8X8_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNUSED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + true, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8B8X8_UNORM (0xeb) + { + "R8G8B8X8_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNUSED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8B8X8_UNORM_SRGB (0xec) + { + "R8G8B8X8_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNUSED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + true, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R9G9B9E5_SHAREDEXP (0xed) + { + "R9G9B9E5_SHAREDEXP", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 9, 9, 9, 5 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B10G10R10X2_UNORM (0xee) + { + "B10G10R10X2_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNUSED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 1023.0f, 1.0f / 1023.0f, 1.0f / 1023.0f, 1.0f / 3.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0xef (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // L16A16_FLOAT (0xf0) + { + "L16A16_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 3, 0, 0 }, // Swizzle + { 16, 16, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // 0xf1 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xf2 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R10G10B10X2_USCALED (0xf3) + { + "R10G10B10X2_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_UNUSED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8B8A8_SSCALED (0xf4) + { + "R8G8B8A8_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_SSCALED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8B8A8_USCALED (0xf5) + { + "R8G8B8A8_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16_SSCALED (0xf6) + { + "R16G16_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 16, 16, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16_USCALED (0xf7) + { + "R16G16_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 16, 16, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32_SSCALED (0xf8) + { + "R32_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 32, 0, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R32_USCALED (0xf9) + { + "R32_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 32, 0, 0, 0 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0xfa (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xfb (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xfc (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xfd (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xfe (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0xff (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // B5G6R5_UNORM (0x100) + { + "B5G6R5_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 0 }, // Swizzle + { 5, 6, 5, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 31.0f, 1.0f / 63.0f, 1.0f / 31.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B5G6R5_UNORM_SRGB (0x101) + { + "B5G6R5_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 0 }, // Swizzle + { 5, 6, 5, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 3, // Num components + true, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 31.0f, 1.0f / 63.0f, 1.0f / 31.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B5G5R5A1_UNORM (0x102) + { + "B5G5R5A1_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 5, 5, 5, 1 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 31.0f, 1.0f / 31.0f, 1.0f / 31.0f, 1.0f / 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B5G5R5A1_UNORM_SRGB (0x103) + { + "B5G5R5A1_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 5, 5, 5, 1 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 4, // Num components + true, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 31.0f, 1.0f / 31.0f, 1.0f / 31.0f, 1.0f / 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B4G4R4A4_UNORM (0x104) + { + "B4G4R4A4_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 4, 4, 4, 4 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 15.0f, 1.0f / 15.0f, 1.0f / 15.0f, 1.0f / 15.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B4G4R4A4_UNORM_SRGB (0x105) + { + "B4G4R4A4_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 4, 4, 4, 4 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 4, // Num components + true, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 15.0f, 1.0f / 15.0f, 1.0f / 15.0f, 1.0f / 15.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8_UNORM (0x106) + { + "R8G8_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 8, 8, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 1.0f / 255.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8_SNORM (0x107) + { + "R8G8_SNORM", + { SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 8, 8, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 127.0f, 1.0f / 127.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8_SINT (0x108) + { + "R8G8_SINT", + { SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 8, 8, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8_UINT (0x109) + { + "R8G8_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 8, 8, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16_UNORM (0x10a) + { + "R16_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 16, 0, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 65535.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16_SNORM (0x10b) + { + "R16_SNORM", + { SWR_TYPE_SNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 16, 0, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 32767.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16_SINT (0x10c) + { + "R16_SINT", + { SWR_TYPE_SINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 16, 0, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16_UINT (0x10d) + { + "R16_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 16, 0, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16_FLOAT (0x10e) + { + "R16_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 16, 0, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x10f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x110 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // I16_UNORM (0x111) + { + "I16_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 16, 0, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 65535.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // L16_UNORM (0x112) + { + "L16_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 16, 0, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 65535.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // A16_UNORM (0x113) + { + "A16_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 3, 0, 0, 0 }, // Swizzle + { 16, 0, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 65535.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // L8A8_UNORM (0x114) + { + "L8A8_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 3, 0, 0 }, // Swizzle + { 8, 8, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 1.0f / 255.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // I16_FLOAT (0x115) + { + "I16_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 16, 0, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // L16_FLOAT (0x116) + { + "L16_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 16, 0, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // A16_FLOAT (0x117) + { + "A16_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 3, 0, 0, 0 }, // Swizzle + { 16, 0, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // L8A8_UNORM_SRGB (0x118) + { + "L8A8_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 3, 0, 0 }, // Swizzle + { 8, 8, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 2, // Num components + true, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 1.0f / 255.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // 0x119 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // B5G5R5X1_UNORM (0x11a) + { + "B5G5R5X1_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNUSED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 5, 5, 5, 1 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 31.0f, 1.0f / 31.0f, 1.0f / 31.0f, 1.0f / 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B5G5R5X1_UNORM_SRGB (0x11b) + { + "B5G5R5X1_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNUSED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 5, 5, 5, 1 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 4, // Num components + true, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 31.0f, 1.0f / 31.0f, 1.0f / 31.0f, 1.0f / 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8_SSCALED (0x11c) + { + "R8G8_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 8, 8, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8_USCALED (0x11d) + { + "R8G8_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 0, 0 }, // Swizzle + { 8, 8, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16_SSCALED (0x11e) + { + "R16_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 16, 0, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16_USCALED (0x11f) + { + "R16_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 16, 0, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x120 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x121 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x122 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x123 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x124 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x125 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // L8A8_UINT (0x126) + { + "L8A8_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 3, 0, 0 }, // Swizzle + { 8, 8, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // L8A8_SINT (0x127) + { + "L8A8_SINT", + { SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 3, 0, 0 }, // Swizzle + { 8, 8, 0, 0 }, // Bits per component + 16, // Bits per element + 2, // Bytes per element + 2, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // 0x128 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x129 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x12a (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x12b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x12c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x12d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x12e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x12f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x130 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x131 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x132 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x133 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x134 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x135 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x136 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x137 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x138 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x139 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x13a (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x13b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x13c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x13d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x13e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x13f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R8_UNORM (0x140) + { + "R8_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8_SNORM (0x141) + { + "R8_SNORM", + { SWR_TYPE_SNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 127.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8_SINT (0x142) + { + "R8_SINT", + { SWR_TYPE_SINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8_UINT (0x143) + { + "R8_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // A8_UNORM (0x144) + { + "A8_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 3, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // I8_UNORM (0x145) + { + "I8_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // L8_UNORM (0x146) + { + "L8_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // 0x147 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x148 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R8_SSCALED (0x149) + { + "R8_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8_USCALED (0x14a) + { + "R8_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x14b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // L8_UNORM_SRGB (0x14c) + { + "L8_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + true, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // 0x14d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x14e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x14f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x150 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x151 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // L8_UINT (0x152) + { + "L8_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // L8_SINT (0x153) + { + "L8_SINT", + { SWR_TYPE_SINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // I8_UINT (0x154) + { + "I8_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // I8_SINT (0x155) + { + "I8_SINT", + { SWR_TYPE_SINT, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 8, // Bits per element + 1, // Bytes per element + 1, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 0, 0, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + true, // isLuminance + }, + // 0x156 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x157 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x158 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x159 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x15a (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x15b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x15c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x15d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x15e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x15f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x160 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x161 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x162 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x163 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x164 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x165 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x166 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x167 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x168 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x169 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x16a (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x16b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x16c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x16d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x16e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x16f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x170 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x171 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x172 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x173 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x174 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x175 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x176 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x177 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x178 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x179 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x17a (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x17b (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x17c (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x17d (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x17e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x17f (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x180 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x181 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x182 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // YCRCB_SWAPUVY (0x183) + { + "YCRCB_SWAPUVY", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + true, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 2, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x184 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x185 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // BC1_UNORM (0x186) + { + "BC1_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 1, // Num components + false, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // BC2_UNORM (0x187) + { + "BC2_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 1, // Num components + false, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // BC3_UNORM (0x188) + { + "BC3_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 1, // Num components + false, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // BC4_UNORM (0x189) + { + "BC4_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 1, // Num components + false, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // BC5_UNORM (0x18a) + { + "BC5_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 1, // Num components + false, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // BC1_UNORM_SRGB (0x18b) + { + "BC1_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 1, // Num components + true, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // BC2_UNORM_SRGB (0x18c) + { + "BC2_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 1, // Num components + true, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // BC3_UNORM_SRGB (0x18d) + { + "BC3_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 1, // Num components + true, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // 0x18e (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // YCRCB_SWAPUV (0x18f) + { + "YCRCB_SWAPUV", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 8, 8, 8, 8 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + true, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 2, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x190 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x191 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x192 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R8G8B8_UNORM (0x193) + { + "R8G8B8_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 8, 8, 8, 0 }, // Bits per component + 24, // Bits per element + 3, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8B8_SNORM (0x194) + { + "R8G8B8_SNORM", + { SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 8, 8, 8, 0 }, // Bits per component + 24, // Bits per element + 3, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 127.0f, 1.0f / 127.0f, 1.0f / 127.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8B8_SSCALED (0x195) + { + "R8G8B8_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 8, 8, 8, 0 }, // Bits per component + 24, // Bits per element + 3, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8B8_USCALED (0x196) + { + "R8G8B8_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 8, 8, 8, 0 }, // Bits per component + 24, // Bits per element + 3, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x197 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x198 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // BC4_SNORM (0x199) + { + "BC4_SNORM", + { SWR_TYPE_SNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 64, // Bits per element + 8, // Bytes per element + 1, // Num components + false, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 127.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // BC5_SNORM (0x19a) + { + "BC5_SNORM", + { SWR_TYPE_SNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 1, // Num components + false, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 127.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // R16G16B16_FLOAT (0x19b) + { + "R16G16B16_FLOAT", + { SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_FLOAT, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 16, 16, 16, 0 }, // Bits per component + 48, // Bits per element + 6, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16B16_UNORM (0x19c) + { + "R16G16B16_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 16, 16, 16, 0 }, // Bits per component + 48, // Bits per element + 6, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 65535.0f, 1.0f / 65535.0f, 1.0f / 65535.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16B16_SNORM (0x19d) + { + "R16G16B16_SNORM", + { SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 16, 16, 16, 0 }, // Bits per component + 48, // Bits per element + 6, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 32767.0f, 1.0f / 32767.0f, 1.0f / 32767.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16B16_SSCALED (0x19e) + { + "R16G16B16_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 16, 16, 16, 0 }, // Bits per component + 48, // Bits per element + 6, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16B16_USCALED (0x19f) + { + "R16G16B16_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 16, 16, 16, 0 }, // Bits per component + 48, // Bits per element + 6, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x1a0 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // BC6H_SF16 (0x1a1) + { + "BC6H_SF16", + { SWR_TYPE_SNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 1, // Num components + false, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 127.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // BC7_UNORM (0x1a2) + { + "BC7_UNORM", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 1, // Num components + false, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // BC7_UNORM_SRGB (0x1a3) + { + "BC7_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 1, // Num components + true, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // BC6H_UF16 (0x1a4) + { + "BC6H_UF16", + { SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 0, 0, 0 }, // Swizzle + { 8, 0, 0, 0 }, // Bits per component + 128, // Bits per element + 16, // Bytes per element + 1, // Num components + false, // isSRGB + true, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 0, 0, 0 }, // To float scale factor + 4, // bcWidth + 4, // bcHeight + false, // isLuminance + }, + // 0x1a5 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1a6 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1a7 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R8G8B8_UNORM_SRGB (0x1a8) + { + "R8G8B8_UNORM_SRGB", + { SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNORM, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 8, 8, 8, 0 }, // Bits per component + 24, // Bits per element + 3, // Bytes per element + 3, // Num components + true, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 255.0f, 1.0f / 255.0f, 1.0f / 255.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x1a9 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1aa (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1ab (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1ac (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1ad (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1ae (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1af (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R16G16B16_UINT (0x1b0) + { + "R16G16B16_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 16, 16, 16, 0 }, // Bits per component + 48, // Bits per element + 6, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R16G16B16_SINT (0x1b1) + { + "R16G16B16_SINT", + { SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 16, 16, 16, 0 }, // Bits per component + 48, // Bits per element + 6, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x1b2 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R10G10B10A2_SNORM (0x1b3) + { + "R10G10B10A2_SNORM", + { SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_SNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 511.0f, 1.0f / 511.0f, 1.0f / 511.0f, 1.0f / 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R10G10B10A2_USCALED (0x1b4) + { + "R10G10B10A2_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R10G10B10A2_SSCALED (0x1b5) + { + "R10G10B10A2_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_SSCALED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R10G10B10A2_SINT (0x1b6) + { + "R10G10B10A2_SINT", + { SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_SINT }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B10G10R10A2_SNORM (0x1b7) + { + "B10G10R10A2_SNORM", + { SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_SNORM, SWR_TYPE_SNORM }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { true, true, true, true }, // Is normalized? + { 1.0f / 511.0f, 1.0f / 511.0f, 1.0f / 511.0f, 1.0f / 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B10G10R10A2_USCALED (0x1b8) + { + "B10G10R10A2_USCALED", + { SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED, SWR_TYPE_USCALED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B10G10R10A2_SSCALED (0x1b9) + { + "B10G10R10A2_SSCALED", + { SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_SSCALED, SWR_TYPE_SSCALED }, + { 0, 0, 0, 0x3f800000 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B10G10R10A2_UINT (0x1ba) + { + "B10G10R10A2_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // B10G10R10A2_SINT (0x1bb) + { + "B10G10R10A2_SINT", + { SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_SINT }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 2, 1, 0, 3 }, // Swizzle + { 10, 10, 10, 2 }, // Bits per component + 32, // Bits per element + 4, // Bytes per element + 4, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 1.0f }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // 0x1bc (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1bd (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1be (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1bf (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1c0 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1c1 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1c2 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1c3 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1c4 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1c5 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1c6 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // 0x1c7 (Padding) + { + "UNKNOWN", + { SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, { 0, 0, 0, 0 }, 0, 0, 0, false, false, false, + { false, false, false, false }, + { 0.0f, 0.0f, 0.0f, 0.0f }, + 1, 1, false }, + // R8G8B8_UINT (0x1c8) + { + "R8G8B8_UINT", + { SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UINT, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 8, 8, 8, 0 }, // Bits per component + 24, // Bits per element + 3, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, + // R8G8B8_SINT (0x1c9) + { + "R8G8B8_SINT", + { SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_SINT, SWR_TYPE_UNKNOWN }, + { 0, 0, 0, 0x1 }, // Defaults for missing components + { 0, 1, 2, 0 }, // Swizzle + { 8, 8, 8, 0 }, // Bits per component + 24, // Bits per element + 3, // Bytes per element + 3, // Num components + false, // isSRGB + false, // isBC + false, // isSubsampled + { false, false, false, false }, // Is normalized? + { 1.0f, 1.0f, 1.0f, 0 }, // To float scale factor + 1, // bcWidth + 1, // bcHeight + false, // isLuminance + }, +}; diff --git a/src/gallium/drivers/swr/rasterizer/common/formats.h b/src/gallium/drivers/swr/rasterizer/common/formats.h new file mode 100644 index 00000000000..b9dd53ebaa4 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/common/formats.h @@ -0,0 +1,251 @@ + +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file formats.h +* +* @brief auto-generated file +* +* DO NOT EDIT +* +******************************************************************************/ + +#pragma once + +#include "common/os.h" + +////////////////////////////////////////////////////////////////////////// +/// SWR_TYPE - Format component type +////////////////////////////////////////////////////////////////////////// +enum SWR_TYPE +{ + SWR_TYPE_UNKNOWN, + SWR_TYPE_UNUSED, + SWR_TYPE_UNORM, + SWR_TYPE_SNORM, + SWR_TYPE_UINT, + SWR_TYPE_SINT, + SWR_TYPE_FLOAT, + SWR_TYPE_SSCALED, + SWR_TYPE_USCALED, +}; +////////////////////////////////////////////////////////////////////////// +/// SWR_FORMAT +////////////////////////////////////////////////////////////////////////// +enum SWR_FORMAT +{ + R32G32B32A32_FLOAT = 0x0, + R32G32B32A32_SINT = 0x1, + R32G32B32A32_UINT = 0x2, + R32G32B32X32_FLOAT = 0x6, + R32G32B32A32_SSCALED = 0x7, + R32G32B32A32_USCALED = 0x8, + R32G32B32_FLOAT = 0x40, + R32G32B32_SINT = 0x41, + R32G32B32_UINT = 0x42, + R32G32B32_SSCALED = 0x45, + R32G32B32_USCALED = 0x46, + R16G16B16A16_UNORM = 0x80, + R16G16B16A16_SNORM = 0x81, + R16G16B16A16_SINT = 0x82, + R16G16B16A16_UINT = 0x83, + R16G16B16A16_FLOAT = 0x84, + R32G32_FLOAT = 0x85, + R32G32_SINT = 0x86, + R32G32_UINT = 0x87, + R32_FLOAT_X8X24_TYPELESS = 0x88, + X32_TYPELESS_G8X24_UINT = 0x89, + L32A32_FLOAT = 0x8A, + R16G16B16X16_UNORM = 0x8E, + R16G16B16X16_FLOAT = 0x8F, + L32X32_FLOAT = 0x91, + I32X32_FLOAT = 0x92, + R16G16B16A16_SSCALED = 0x93, + R16G16B16A16_USCALED = 0x94, + R32G32_SSCALED = 0x95, + R32G32_USCALED = 0x96, + R32_FLOAT_X8X24_TYPELESS_LD = 0x98, + B8G8R8A8_UNORM = 0xC0, + B8G8R8A8_UNORM_SRGB = 0xC1, + R10G10B10A2_UNORM = 0xC2, + R10G10B10A2_UNORM_SRGB = 0xC3, + R10G10B10A2_UINT = 0xC4, + R8G8B8A8_UNORM = 0xC7, + R8G8B8A8_UNORM_SRGB = 0xC8, + R8G8B8A8_SNORM = 0xC9, + R8G8B8A8_SINT = 0xCA, + R8G8B8A8_UINT = 0xCB, + R16G16_UNORM = 0xCC, + R16G16_SNORM = 0xCD, + R16G16_SINT = 0xCE, + R16G16_UINT = 0xCF, + R16G16_FLOAT = 0xD0, + B10G10R10A2_UNORM = 0xD1, + B10G10R10A2_UNORM_SRGB = 0xD2, + R11G11B10_FLOAT = 0xD3, + R32_SINT = 0xD6, + R32_UINT = 0xD7, + R32_FLOAT = 0xD8, + R24_UNORM_X8_TYPELESS = 0xD9, + R24_UNORM_X8_TYPELESS_LD = 0xDC, + L32_UNORM = 0xDD, + L16A16_UNORM = 0xDF, + I24X8_UNORM = 0xE0, + L24X8_UNORM = 0xE1, + I32_FLOAT = 0xE3, + L32_FLOAT = 0xE4, + A32_FLOAT = 0xE5, + B8G8R8X8_UNORM = 0xE9, + B8G8R8X8_UNORM_SRGB = 0xEA, + R8G8B8X8_UNORM = 0xEB, + R8G8B8X8_UNORM_SRGB = 0xEC, + R9G9B9E5_SHAREDEXP = 0xED, + B10G10R10X2_UNORM = 0xEE, + L16A16_FLOAT = 0xF0, + R10G10B10X2_USCALED = 0xF3, + R8G8B8A8_SSCALED = 0xF4, + R8G8B8A8_USCALED = 0xF5, + R16G16_SSCALED = 0xF6, + R16G16_USCALED = 0xF7, + R32_SSCALED = 0xF8, + R32_USCALED = 0xF9, + B5G6R5_UNORM = 0x100, + B5G6R5_UNORM_SRGB = 0x101, + B5G5R5A1_UNORM = 0x102, + B5G5R5A1_UNORM_SRGB = 0x103, + B4G4R4A4_UNORM = 0x104, + B4G4R4A4_UNORM_SRGB = 0x105, + R8G8_UNORM = 0x106, + R8G8_SNORM = 0x107, + R8G8_SINT = 0x108, + R8G8_UINT = 0x109, + R16_UNORM = 0x10A, + R16_SNORM = 0x10B, + R16_SINT = 0x10C, + R16_UINT = 0x10D, + R16_FLOAT = 0x10E, + I16_UNORM = 0x111, + L16_UNORM = 0x112, + A16_UNORM = 0x113, + L8A8_UNORM = 0x114, + I16_FLOAT = 0x115, + L16_FLOAT = 0x116, + A16_FLOAT = 0x117, + L8A8_UNORM_SRGB = 0x118, + B5G5R5X1_UNORM = 0x11A, + B5G5R5X1_UNORM_SRGB = 0x11B, + R8G8_SSCALED = 0x11C, + R8G8_USCALED = 0x11D, + R16_SSCALED = 0x11E, + R16_USCALED = 0x11F, + L8A8_UINT = 0x126, + L8A8_SINT = 0x127, + R8_UNORM = 0x140, + R8_SNORM = 0x141, + R8_SINT = 0x142, + R8_UINT = 0x143, + A8_UNORM = 0x144, + I8_UNORM = 0x145, + L8_UNORM = 0x146, + R8_SSCALED = 0x149, + R8_USCALED = 0x14A, + L8_UNORM_SRGB = 0x14C, + L8_UINT = 0x152, + L8_SINT = 0x153, + I8_UINT = 0x154, + I8_SINT = 0x155, + YCRCB_SWAPUVY = 0x183, + BC1_UNORM = 0x186, + BC2_UNORM = 0x187, + BC3_UNORM = 0x188, + BC4_UNORM = 0x189, + BC5_UNORM = 0x18A, + BC1_UNORM_SRGB = 0x18B, + BC2_UNORM_SRGB = 0x18C, + BC3_UNORM_SRGB = 0x18D, + YCRCB_SWAPUV = 0x18F, + R8G8B8_UNORM = 0x193, + R8G8B8_SNORM = 0x194, + R8G8B8_SSCALED = 0x195, + R8G8B8_USCALED = 0x196, + BC4_SNORM = 0x199, + BC5_SNORM = 0x19A, + R16G16B16_FLOAT = 0x19B, + R16G16B16_UNORM = 0x19C, + R16G16B16_SNORM = 0x19D, + R16G16B16_SSCALED = 0x19E, + R16G16B16_USCALED = 0x19F, + BC6H_SF16 = 0x1A1, + BC7_UNORM = 0x1A2, + BC7_UNORM_SRGB = 0x1A3, + BC6H_UF16 = 0x1A4, + R8G8B8_UNORM_SRGB = 0x1A8, + R16G16B16_UINT = 0x1B0, + R16G16B16_SINT = 0x1B1, + R10G10B10A2_SNORM = 0x1B3, + R10G10B10A2_USCALED = 0x1B4, + R10G10B10A2_SSCALED = 0x1B5, + R10G10B10A2_SINT = 0x1B6, + B10G10R10A2_SNORM = 0x1B7, + B10G10R10A2_USCALED = 0x1B8, + B10G10R10A2_SSCALED = 0x1B9, + B10G10R10A2_UINT = 0x1BA, + B10G10R10A2_SINT = 0x1BB, + R8G8B8_UINT = 0x1C8, + R8G8B8_SINT = 0x1C9, + NUM_SWR_FORMATS = 0x1CA, +}; +////////////////////////////////////////////////////////////////////////// +/// SWR_FORMAT_INFO - Format information +////////////////////////////////////////////////////////////////////////// +struct SWR_FORMAT_INFO +{ + const char* name; + SWR_TYPE type[4]; + uint32_t defaults[4]; + uint32_t swizzle[4]; ///< swizzle per component + uint32_t bpc[4]; ///< bits per component + uint32_t bpp; ///< bits per pixel + uint32_t Bpp; ///< bytes per pixel + uint32_t numComps; ///< number of components + bool isSRGB; + bool isBC; + bool isSubsampled; + bool isNormalized[4]; + float toFloat[4]; + uint32_t bcWidth; + uint32_t bcHeight; + bool isLuminance; +}; + +extern const SWR_FORMAT_INFO gFormatInfo[]; + +////////////////////////////////////////////////////////////////////////// +/// @brief Retrieves format info struct for given format. +/// @param format - SWR format +INLINE const SWR_FORMAT_INFO& GetFormatInfo(SWR_FORMAT format) +{ + return gFormatInfo[format]; +} + +// lookup table for unorm8 srgb -> float conversion +extern const uint32_t srgb8Table[256]; diff --git a/src/gallium/drivers/swr/rasterizer/common/isa.hpp b/src/gallium/drivers/swr/rasterizer/common/isa.hpp new file mode 100644 index 00000000000..ef381799bc3 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/common/isa.hpp @@ -0,0 +1,235 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +****************************************************************************/ + +#pragma once + +#include +#include +#include +#include +#include +#include + +#if defined(_WIN32) +#include +#else +#include +#include +#endif + +class InstructionSet +{ +public: + InstructionSet() : CPU_Rep() {}; + + // getters + std::string Vendor(void) { return CPU_Rep.vendor_; } + std::string Brand(void) { return CPU_Rep.brand_; } + + bool SSE3(void) { return CPU_Rep.f_1_ECX_[0]; } + bool PCLMULQDQ(void) { return CPU_Rep.f_1_ECX_[1]; } + bool MONITOR(void) { return CPU_Rep.f_1_ECX_[3]; } + bool SSSE3(void) { return CPU_Rep.f_1_ECX_[9]; } + bool FMA(void) { return CPU_Rep.f_1_ECX_[12]; } + bool CMPXCHG16B(void) { return CPU_Rep.f_1_ECX_[13]; } + bool SSE41(void) { return CPU_Rep.f_1_ECX_[19]; } + bool SSE42(void) { return CPU_Rep.f_1_ECX_[20]; } + bool MOVBE(void) { return CPU_Rep.f_1_ECX_[22]; } + bool POPCNT(void) { return CPU_Rep.f_1_ECX_[23]; } + bool AES(void) { return CPU_Rep.f_1_ECX_[25]; } + bool XSAVE(void) { return CPU_Rep.f_1_ECX_[26]; } + bool OSXSAVE(void) { return CPU_Rep.f_1_ECX_[27]; } + bool RDRAND(void) { return CPU_Rep.f_1_ECX_[30]; } + + bool MSR(void) { return CPU_Rep.f_1_EDX_[5]; } + bool CX8(void) { return CPU_Rep.f_1_EDX_[8]; } + bool SEP(void) { return CPU_Rep.f_1_EDX_[11]; } + bool CMOV(void) { return CPU_Rep.f_1_EDX_[15]; } + bool CLFSH(void) { return CPU_Rep.f_1_EDX_[19]; } + bool MMX(void) { return CPU_Rep.f_1_EDX_[23]; } + bool FXSR(void) { return CPU_Rep.f_1_EDX_[24]; } + bool SSE(void) { return CPU_Rep.f_1_EDX_[25]; } + bool SSE2(void) { return CPU_Rep.f_1_EDX_[26]; } + + bool FSGSBASE(void) { return CPU_Rep.f_7_EBX_[0]; } + bool BMI1(void) { return CPU_Rep.f_7_EBX_[3]; } + bool HLE(void) { return CPU_Rep.isIntel_ && CPU_Rep.f_7_EBX_[4]; } + bool BMI2(void) { return CPU_Rep.f_7_EBX_[8]; } + bool ERMS(void) { return CPU_Rep.f_7_EBX_[9]; } + bool INVPCID(void) { return CPU_Rep.f_7_EBX_[10]; } + bool RTM(void) { return CPU_Rep.isIntel_ && CPU_Rep.f_7_EBX_[11]; } + bool RDSEED(void) { return CPU_Rep.f_7_EBX_[18]; } + bool ADX(void) { return CPU_Rep.f_7_EBX_[19]; } + bool SHA(void) { return CPU_Rep.f_7_EBX_[29]; } + + bool PREFETCHWT1(void) { return CPU_Rep.f_7_ECX_[0]; } + + bool LAHF(void) { return CPU_Rep.f_81_ECX_[0]; } + bool LZCNT(void) { return CPU_Rep.isIntel_ && CPU_Rep.f_81_ECX_[5]; } + bool ABM(void) { return CPU_Rep.isAMD_ && CPU_Rep.f_81_ECX_[5]; } + bool SSE4a(void) { return CPU_Rep.isAMD_ && CPU_Rep.f_81_ECX_[6]; } + bool XOP(void) { return CPU_Rep.isAMD_ && CPU_Rep.f_81_ECX_[11]; } + bool TBM(void) { return CPU_Rep.isAMD_ && CPU_Rep.f_81_ECX_[21]; } + + bool SYSCALL(void) { return CPU_Rep.isIntel_ && CPU_Rep.f_81_EDX_[11]; } + bool MMXEXT(void) { return CPU_Rep.isAMD_ && CPU_Rep.f_81_EDX_[22]; } + bool RDTSCP(void) { return CPU_Rep.isIntel_ && CPU_Rep.f_81_EDX_[27]; } + bool _3DNOWEXT(void) { return CPU_Rep.isAMD_ && CPU_Rep.f_81_EDX_[30]; } + bool _3DNOW(void) { return CPU_Rep.isAMD_ && CPU_Rep.f_81_EDX_[31]; } + + bool AVX(void) { return CPU_Rep.f_1_ECX_[28]; } + bool F16C(void) { return CPU_Rep.f_1_ECX_[29]; } + bool AVX2(void) { return CPU_Rep.f_7_EBX_[5]; } + bool AVX512F(void) { return CPU_Rep.f_7_EBX_[16]; } + bool AVX512PF(void) { return CPU_Rep.f_7_EBX_[26]; } + bool AVX512ER(void) { return CPU_Rep.f_7_EBX_[27]; } + bool AVX512CD(void) { return CPU_Rep.f_7_EBX_[28]; } + +private: + class InstructionSet_Internal + { + public: + InstructionSet_Internal() + : nIds_{ 0 }, + nExIds_{ 0 }, + isIntel_{ false }, + isAMD_{ false }, + f_1_ECX_{ 0 }, + f_1_EDX_{ 0 }, + f_7_EBX_{ 0 }, + f_7_ECX_{ 0 }, + f_81_ECX_{ 0 }, + f_81_EDX_{ 0 }, + data_{}, + extdata_{} + { + //int cpuInfo[4] = {-1}; + std::array cpui; + + // Calling __cpuid with 0x0 as the function_id argument + // gets the number of the highest valid function ID. +#if defined(_WIN32) + __cpuid(cpui.data(), 0); + nIds_ = cpui[0]; +#else + nIds_ = __get_cpuid_max(0, NULL); +#endif + + for (int i = 0; i <= nIds_; ++i) + { +#if defined(_WIN32) + __cpuidex(cpui.data(), i, 0); +#else + int *data = cpui.data(); + __cpuid_count(i, 0, data[0], data[1], data[2], data[3]); +#endif + data_.push_back(cpui); + } + + // Capture vendor string + char vendor[0x20]; + memset(vendor, 0, sizeof(vendor)); + *reinterpret_cast(vendor) = data_[0][1]; + *reinterpret_cast(vendor + 4) = data_[0][3]; + *reinterpret_cast(vendor + 8) = data_[0][2]; + vendor_ = vendor; + if (vendor_ == "GenuineIntel") + { + isIntel_ = true; + } + else if (vendor_ == "AuthenticAMD") + { + isAMD_ = true; + } + + // load bitset with flags for function 0x00000001 + if (nIds_ >= 1) + { + f_1_ECX_ = data_[1][2]; + f_1_EDX_ = data_[1][3]; + } + + // load bitset with flags for function 0x00000007 + if (nIds_ >= 7) + { + f_7_EBX_ = data_[7][1]; + f_7_ECX_ = data_[7][2]; + } + + // Calling __cpuid with 0x80000000 as the function_id argument + // gets the number of the highest valid extended ID. +#if defined(_WIN32) + __cpuid(cpui.data(), 0x80000000); + nExIds_ = cpui[0]; +#else + nExIds_ = __get_cpuid_max(0x80000000, NULL); +#endif + + char brand[0x40]; + memset(brand, 0, sizeof(brand)); + + for (unsigned i = 0x80000000; i <= nExIds_; ++i) + { +#if defined(_WIN32) + __cpuidex(cpui.data(), i, 0); +#else + int *data = cpui.data(); + __cpuid_count(i, 0, data[0], data[1], data[2], data[3]); +#endif + extdata_.push_back(cpui); + } + + // load bitset with flags for function 0x80000001 + if (nExIds_ >= 0x80000001) + { + f_81_ECX_ = extdata_[1][2]; + f_81_EDX_ = extdata_[1][3]; + } + + // Interpret CPU brand string if reported + if (nExIds_ >= 0x80000004) + { + memcpy(brand, extdata_[2].data(), sizeof(cpui)); + memcpy(brand + 16, extdata_[3].data(), sizeof(cpui)); + memcpy(brand + 32, extdata_[4].data(), sizeof(cpui)); + brand_ = brand; + } + }; + + int nIds_; + unsigned nExIds_; + std::string vendor_; + std::string brand_; + bool isIntel_; + bool isAMD_; + std::bitset<32> f_1_ECX_; + std::bitset<32> f_1_EDX_; + std::bitset<32> f_7_EBX_; + std::bitset<32> f_7_ECX_; + std::bitset<32> f_81_ECX_; + std::bitset<32> f_81_EDX_; + std::vector> data_; + std::vector> extdata_; + }; + const InstructionSet_Internal CPU_Rep; +}; diff --git a/src/gallium/drivers/swr/rasterizer/common/os.h b/src/gallium/drivers/swr/rasterizer/common/os.h new file mode 100644 index 00000000000..736d29856a6 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/common/os.h @@ -0,0 +1,221 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +****************************************************************************/ + +#ifndef __SWR_OS_H__ +#define __SWR_OS_H__ + +#include "core/knobs.h" + +#if (defined(FORCE_WINDOWS) || defined(_WIN32)) && !defined(FORCE_LINUX) + +#define SWR_API __cdecl + +#ifndef _CRT_SECURE_NO_WARNINGS +#define _CRT_SECURE_NO_WARNINGS +#endif + +#ifndef NOMINMAX +#define NOMINMAX +#endif +#include "Windows.h" +#include +#include + +#define OSALIGN(RWORD, WIDTH) __declspec(align(WIDTH)) RWORD +#define THREAD __declspec(thread) +#define INLINE __forceinline +#define DEBUGBREAK __debugbreak() + +#define PRAGMA_WARNING_PUSH_DISABLE(...) \ + __pragma(warning(push));\ + __pragma(warning(disable:__VA_ARGS__)); + +#define PRAGMA_WARNING_POP() __pragma(warning(pop)) + +#if defined(_WIN32) +#if defined(_WIN64) +#define BitScanForwardSizeT BitScanForward64 +#define _mm_popcount_sizeT _mm_popcnt_u64 +#else +#define BitScanForwardSizeT BitScanForward +#define _mm_popcount_sizeT _mm_popcnt_u32 +#endif +#endif + +#elif defined(FORCE_LINUX) || defined(__linux__) || defined(__gnu_linux__) + +#define SWR_API + +#include +#include +#include +#include +#include +#include +#include +#include + +typedef void VOID; +typedef void* LPVOID; +typedef CARD8 BOOL; +typedef wchar_t WCHAR; +typedef uint16_t UINT16; +typedef int INT; +typedef int INT32; +typedef unsigned int UINT; +typedef uint32_t UINT32; +typedef uint64_t UINT64; +typedef int64_t INT64; +typedef void* HANDLE; +typedef float FLOAT; +typedef int LONG; +typedef CARD8 BYTE; +typedef unsigned char UCHAR; +typedef unsigned int DWORD; + +#undef FALSE +#define FALSE 0 + +#undef TRUE +#define TRUE 1 + +#define OSALIGN(RWORD, WIDTH) RWORD __attribute__((aligned(WIDTH))) +#define THREAD __thread +#ifndef INLINE +#define INLINE __inline +#endif +#define DEBUGBREAK asm ("int $3") +#define __cdecl +#define __declspec(X) + +#define GCC_VERSION (__GNUC__ * 10000 \ + + __GNUC_MINOR__ * 100 \ + + __GNUC_PATCHLEVEL__) + +#if !defined(__clang__) && (__GNUC__) && (GCC_VERSION < 40500) +inline +uint64_t __rdtsc() +{ + long low, high; + asm volatile("rdtsc" : "=a"(low), "=d"(high)); + return (low | ((uint64_t)high << 32)); +} +#endif + +#ifndef __clang__ +// Intrinsic not defined in gcc +static INLINE +void _mm256_storeu2_m128i(__m128i *hi, __m128i *lo, __m256i a) +{ + _mm_storeu_si128((__m128i*)lo, _mm256_castsi256_si128(a)); + _mm_storeu_si128((__m128i*)hi, _mm256_extractf128_si256(a, 0x1)); +} +#endif + +inline +unsigned char _BitScanForward(unsigned long *Index, unsigned long Mask) +{ + *Index = __builtin_ctz(Mask); + return (Mask != 0); +} + +inline +unsigned char _BitScanForward(unsigned int *Index, unsigned int Mask) +{ + *Index = __builtin_ctz(Mask); + return (Mask != 0); +} + +inline +unsigned char _BitScanReverse(unsigned long *Index, unsigned long Mask) +{ + *Index = __builtin_clz(Mask); + return (Mask != 0); +} + +inline +unsigned char _BitScanReverse(unsigned int *Index, unsigned int Mask) +{ + *Index = __builtin_clz(Mask); + return (Mask != 0); +} + +inline +void *_aligned_malloc(unsigned int size, unsigned int alignment) +{ + void *ret; + if (posix_memalign(&ret, alignment, size)) + { + return NULL; + } + return ret; +} + +inline +unsigned char _bittest(const LONG *a, LONG b) +{ + return ((*(unsigned *)(a) & (1 << b)) != 0); +} + +#define GetCurrentProcessId getpid + +#define CreateDirectory(name, pSecurity) mkdir(name, 0777) + +#if defined(_WIN32) +static inline +unsigned int _mm_popcnt_u32(unsigned int v) +{ + return __builtin_popcount(v); +} +#endif + +#define _aligned_free free +#define InterlockedCompareExchange(Dest, Exchange, Comparand) __sync_val_compare_and_swap(Dest, Comparand, Exchange) +#define InterlockedExchangeAdd(Addend, Value) __sync_fetch_and_add(Addend, Value) +#define InterlockedDecrement(Append) __sync_sub_and_fetch(Append, 1) +#define InterlockedIncrement(Append) __sync_add_and_fetch(Append, 1) +#define _ReadWriteBarrier() asm volatile("" ::: "memory") +#define __stdcall + +#define PRAGMA_WARNING_PUSH_DISABLE(...) +#define PRAGMA_WARNING_POP() + +#else + +#error Unsupported OS/system. + +#endif + +// Universal types +typedef BYTE KILOBYTE[1024]; +typedef KILOBYTE MEGABYTE[1024]; +typedef MEGABYTE GIGABYTE[1024]; + +#define OSALIGNLINE(RWORD) OSALIGN(RWORD, 64) +#if KNOB_SIMD_WIDTH == 8 +#define OSALIGNSIMD(RWORD) OSALIGN(RWORD, 32) +#endif + +#include "common/swr_assert.h" + +#endif//__SWR_OS_H__ diff --git a/src/gallium/drivers/swr/rasterizer/common/rdtsc_buckets.cpp b/src/gallium/drivers/swr/rasterizer/common/rdtsc_buckets.cpp new file mode 100644 index 00000000000..454641b2751 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/common/rdtsc_buckets.cpp @@ -0,0 +1,188 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file rdtsc_buckets.cpp +* +* @brief implementation of rdtsc buckets. +* +* Notes: +* +******************************************************************************/ +#include "rdtsc_buckets.h" +#include + +THREAD UINT tlsThreadId = 0; + +void BucketManager::RegisterThread(const std::string& name) +{ + BUCKET_THREAD newThread; + newThread.name = name; + newThread.root.children.reserve(mBuckets.size()); + newThread.root.id = 0; + newThread.root.pParent = nullptr; + newThread.pCurrent = &newThread.root; + + mThreadMutex.lock(); + + // assign unique thread id for this thread + size_t id = mThreads.size(); + newThread.id = (UINT)id; + tlsThreadId = (UINT)id; + + // open threadviz file if enabled + if (mThreadViz) + { + std::stringstream ss; + ss << mThreadVizDir << "\\threadviz_thread." << newThread.id << ".dat"; + newThread.vizFile = fopen(ss.str().c_str(), "wb"); + } + + // store new thread + mThreads.push_back(newThread); + + mThreadMutex.unlock(); +} + +UINT BucketManager::RegisterBucket(const BUCKET_DESC& desc) +{ + size_t id = mBuckets.size(); + mBuckets.push_back(desc); + return (UINT)id; +} + +void BucketManager::PrintBucket(FILE* f, UINT level, UINT64 threadCycles, UINT64 parentCycles, const BUCKET& bucket) +{ + const char *arrows[] = { + "", + "|-> ", + " |-> ", + " |-> ", + " |-> ", + " |-> ", + " |-> " + }; + + // compute percent of total cycles used by this bucket + float percentTotal = (float)((double)bucket.elapsed / (double)threadCycles * 100.0); + + // compute percent of parent cycles used by this bucket + float percentParent = (float)((double)bucket.elapsed / (double)parentCycles * 100.0); + + // compute average cycle count per invocation + UINT64 CPE = bucket.elapsed / bucket.count; + + BUCKET_DESC &desc = mBuckets[bucket.id]; + + // construct hierarchy visualization + char hier[80]; + strcpy(hier, arrows[level]); + strcat(hier, desc.name.c_str()); + + // print out + fprintf(f, "%6.2f %6.2f %-10" PRIu64 " %-10" PRIu64 " %-10u %-10lu %-10u %s\n", + percentTotal, + percentParent, + bucket.elapsed, + CPE, + bucket.count, + (unsigned long)0, + (uint32_t)0, + hier + ); + + // dump all children of this bucket + for (const BUCKET& child : bucket.children) + { + if (child.count) + { + PrintBucket(f, level + 1, threadCycles, bucket.elapsed, child); + } + } +} + +void BucketManager::PrintThread(FILE* f, const BUCKET_THREAD& thread) +{ + // print header + fprintf(f, "\nThread %u (%s)\n", thread.id, thread.name.c_str()); + fprintf(f, " %%Tot %%Par Cycles CPE NumEvent CPE2 NumEvent2 Bucket\n"); + + // compute thread level total cycle counts across all buckets from root + const BUCKET& root = thread.root; + UINT64 totalCycles = 0; + for (const BUCKET& child : root.children) + { + totalCycles += child.elapsed; + } + + for (const BUCKET& child : root.children) + { + if (child.count) + { + PrintBucket(f, 0, totalCycles, totalCycles, child); + } + } +} + +void BucketManager::DumpThreadViz() +{ + // ensure all thread data is flushed + mThreadMutex.lock(); + for (auto& thread : mThreads) + { + fflush(thread.vizFile); + fclose(thread.vizFile); + } + mThreadMutex.unlock(); + + // dump bucket descriptions + std::stringstream ss; + ss << mThreadVizDir << "\\threadviz_buckets.dat"; + + FILE* f = fopen(ss.str().c_str(), "wb"); + for (auto& bucket : mBuckets) + { + Serialize(f, bucket); + } + fclose(f); +} + +void BucketManager::PrintReport(const std::string& filename) +{ + if (mThreadViz) + { + DumpThreadViz(); + } + else + { + FILE* f = fopen(filename.c_str(), "w"); + + mThreadMutex.lock(); + for (const BUCKET_THREAD& thread : mThreads) + { + PrintThread(f, thread); + fprintf(f, "\n"); + } + mThreadMutex.unlock(); + + fclose(f); + } +} diff --git a/src/gallium/drivers/swr/rasterizer/common/rdtsc_buckets.h b/src/gallium/drivers/swr/rasterizer/common/rdtsc_buckets.h new file mode 100644 index 00000000000..99cb10ec6e8 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/common/rdtsc_buckets.h @@ -0,0 +1,229 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file rdtsc_buckets.h +* +* @brief declaration for rdtsc buckets. +* +* Notes: +* +******************************************************************************/ +#pragma once + +#include "os.h" +#include +#include +#include + +#include "rdtsc_buckets_shared.h" + +// unique thread id stored in thread local storage +extern THREAD UINT tlsThreadId; + +////////////////////////////////////////////////////////////////////////// +/// @brief BucketManager encapsulates a single instance of the buckets +/// functionality. There can be one or many bucket managers active +/// at any time. The manager owns all the threads and +/// bucket information that have been registered to it. +class BucketManager +{ +public: + BucketManager(bool enableThreadViz) : mThreadViz(enableThreadViz) + { + if (mThreadViz) + { + uint32_t pid = GetCurrentProcessId(); + std::stringstream str; + str << "threadviz." << pid; + mThreadVizDir = str.str(); + CreateDirectory(mThreadVizDir.c_str(), NULL); + } + } + + // removes all registered thread data + void ClearThreads() + { + mThreadMutex.lock(); + mThreads.clear(); + mThreadMutex.unlock(); + } + + // removes all registered buckets + void ClearBuckets() + { + mBuckets.clear(); + } + + /// Registers a new thread with the manager. + /// @param name - name of thread, used for labels in reports and threadviz + void RegisterThread(const std::string& name); + + /// Registers a new bucket type with the manager. Returns a unique + /// id which should be used in subsequent calls to start/stop the bucket + /// @param desc - description of the bucket + /// @return unique id + UINT RegisterBucket(const BUCKET_DESC& desc); + + // dump threadviz data + void DumpThreadViz(); + + // print report + void PrintReport(const std::string& filename); + + // start capturing + INLINE void StartCapture() + { + mCapturing = true; + } + + // stop capturing + INLINE void StopCapture() + { + mCapturing = false; + + // wait for all threads to pop back to root bucket + bool stillCapturing = true; + while (stillCapturing) + { + stillCapturing = false; + for (const BUCKET_THREAD& t : mThreads) + { + if (t.pCurrent != &t.root) + { + stillCapturing = true; + continue; + } + } + } + } + + // start a bucket + // @param id generated by RegisterBucket + INLINE void StartBucket(UINT id) + { + if (!mCapturing) return; + + SWR_ASSERT(tlsThreadId < mThreads.size()); + + BUCKET_THREAD& bt = mThreads[tlsThreadId]; + + // if threadviz is enabled, only need to dump start info to threads viz file + if (mThreadViz) + { + SWR_ASSERT(bt.vizFile != nullptr); + if (mBuckets[id].enableThreadViz) + { + VIZ_START_DATA data{ VIZ_START, id, __rdtsc() }; + Serialize(bt.vizFile, data); + } + } + else + { + if (bt.pCurrent->children.size() < mBuckets.size()) + { + bt.pCurrent->children.resize(mBuckets.size()); + } + BUCKET &child = bt.pCurrent->children[id]; + child.pParent = bt.pCurrent; + child.id = id; + child.start = __rdtsc(); + + // update thread's currently executing bucket + bt.pCurrent = &child; + } + + bt.level++; + } + + // stop the currently executing bucket + INLINE void StopBucket(UINT id) + { + SWR_ASSERT(tlsThreadId < mThreads.size()); + BUCKET_THREAD &bt = mThreads[tlsThreadId]; + + if (bt.level == 0) return; + + if (mThreadViz) + { + SWR_ASSERT(bt.vizFile != nullptr); + if (mBuckets[id].enableThreadViz) + { + VIZ_STOP_DATA data{ VIZ_STOP, __rdtsc() }; + Serialize(bt.vizFile, data); + } + } + else + { + if (bt.pCurrent->start == 0) return; + SWR_ASSERT(bt.pCurrent->id == id, "Mismatched buckets detected"); + + bt.pCurrent->elapsed += (__rdtsc() - bt.pCurrent->start); + bt.pCurrent->count++; + + // pop to parent + bt.pCurrent = bt.pCurrent->pParent; + } + + bt.level--; + } + + INLINE void AddEvent(uint32_t id, uint32_t count) + { + if (!mCapturing) return; + + SWR_ASSERT(tlsThreadId < mThreads.size()); + + BUCKET_THREAD& bt = mThreads[tlsThreadId]; + + // don't record events for threadviz + if (!mThreadViz) + { + if (bt.pCurrent->children.size() < mBuckets.size()) + { + bt.pCurrent->children.resize(mBuckets.size()); + } + BUCKET &child = bt.pCurrent->children[id]; + child.pParent = bt.pCurrent; + child.id = id; + child.count += count; + } + } + +private: + void PrintBucket(FILE* f, UINT level, UINT64 threadCycles, UINT64 parentCycles, const BUCKET& bucket); + void PrintThread(FILE* f, const BUCKET_THREAD& thread); + + // list of active threads that have registered with this manager + std::vector mThreads; + + // list of buckets registered with this manager + std::vector mBuckets; + + // is capturing currently enabled + volatile bool mCapturing{ false }; + + std::mutex mThreadMutex; + + // enable threadviz + bool mThreadViz{ false }; + std::string mThreadVizDir; +}; diff --git a/src/gallium/drivers/swr/rasterizer/common/rdtsc_buckets_shared.h b/src/gallium/drivers/swr/rasterizer/common/rdtsc_buckets_shared.h new file mode 100644 index 00000000000..41c6d5dec79 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/common/rdtsc_buckets_shared.h @@ -0,0 +1,167 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file rdtsc_buckets.h +* +* @brief declaration for rdtsc buckets. +* +* Notes: +* +******************************************************************************/ +#pragma once + +#include +#include + +struct BUCKET +{ + uint32_t id{ 0 }; + uint64_t start{ 0 }; + uint64_t elapsed{ 0 }; + uint32_t count{ 0 }; + + BUCKET* pParent{ nullptr }; + std::vector children; +}; + +struct BUCKET_DESC +{ + // name of bucket, used in reports + std::string name; + + // description of bucket, used in threadviz + std::string description; + + // enable for threadviz dumping + bool enableThreadViz; + + // threadviz color of bucket, in RGBA8_UNORM format + uint32_t color; +}; + +struct BUCKET_THREAD +{ + // name of thread, used in reports + std::string name; + + // id for this thread, assigned by the thread manager + uint32_t id; + + // root of the bucket hierarchy for this thread + BUCKET root; + + // currently executing bucket somewhere in the hierarchy + BUCKET* pCurrent; + + // currently executing hierarchy level + uint32_t level{ 0 }; + + // threadviz file object + FILE* vizFile{ nullptr }; + + BUCKET_THREAD() {} + BUCKET_THREAD(const BUCKET_THREAD& that) + { + name = that.name; + id = that.id; + root = that.root; + pCurrent = &root; + vizFile = that.vizFile; + } +}; + +enum VIZ_TYPE +{ + VIZ_START = 0, + VIZ_STOP = 1, + VIZ_DATA = 2 +}; + +struct VIZ_START_DATA +{ + uint8_t type; + uint32_t bucketId; + uint64_t timestamp; +}; + +struct VIZ_STOP_DATA +{ + uint8_t type; + uint64_t timestamp; +}; + +inline void Serialize(FILE* f, const VIZ_START_DATA& data) +{ + fwrite(&data, sizeof(VIZ_START_DATA), 1, f); +} + +inline void Deserialize(FILE* f, VIZ_START_DATA& data) +{ + fread(&data, sizeof(VIZ_START_DATA), 1, f); + assert(data.type == VIZ_START); +} + +inline void Serialize(FILE* f, const VIZ_STOP_DATA& data) +{ + fwrite(&data, sizeof(VIZ_STOP_DATA), 1, f); +} + +inline void Deserialize(FILE* f, VIZ_STOP_DATA& data) +{ + fread(&data, sizeof(VIZ_STOP_DATA), 1, f); + assert(data.type == VIZ_STOP); +} + +inline void Serialize(FILE* f, const std::string& string) +{ + assert(string.size() <= 256); + + uint8_t length = (uint8_t)string.size(); + fwrite(&length, sizeof(length), 1, f); + fwrite(string.c_str(), string.size(), 1, f); +} + +inline void Deserialize(FILE* f, std::string& string) +{ + char cstr[256]; + uint8_t length; + fread(&length, sizeof(length), 1, f); + fread(cstr, length, 1, f); + cstr[length] = 0; + string.assign(cstr); +} + +inline void Serialize(FILE* f, const BUCKET_DESC& desc) +{ + Serialize(f, desc.name); + Serialize(f, desc.description); + fwrite(&desc.enableThreadViz, sizeof(desc.enableThreadViz), 1, f); + fwrite(&desc.color, sizeof(desc.color), 1, f); +} + +inline void Deserialize(FILE* f, BUCKET_DESC& desc) +{ + Deserialize(f, desc.name); + Deserialize(f, desc.description); + fread(&desc.enableThreadViz, sizeof(desc.enableThreadViz), 1, f); + fread(&desc.color, sizeof(desc.color), 1, f); +} diff --git a/src/gallium/drivers/swr/rasterizer/common/simdintrin.h b/src/gallium/drivers/swr/rasterizer/common/simdintrin.h new file mode 100644 index 00000000000..8fa6d9ef408 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/common/simdintrin.h @@ -0,0 +1,787 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +****************************************************************************/ + +#ifndef __SWR_SIMDINTRIN_H__ +#define __SWR_SIMDINTRIN_H__ + +#include "os.h" + +#include + +#include +#include +#include + +#if KNOB_SIMD_WIDTH == 8 +typedef __m256 simdscalar; +typedef __m256i simdscalari; +typedef uint8_t simdmask; +#else +#error Unsupported vector width +#endif + +// simd vector +OSALIGNSIMD(union) simdvector +{ + simdscalar v[4]; + struct + { + simdscalar x, y, z, w; + }; + + simdscalar& operator[] (const int i) { return v[i]; } + const simdscalar& operator[] (const int i) const { return v[i]; } +}; + +#if KNOB_SIMD_WIDTH == 8 +#define _simd128_maskstore_ps _mm_maskstore_ps +#define _simd_load_ps _mm256_load_ps +#define _simd_load1_ps _mm256_broadcast_ss +#define _simd_loadu_ps _mm256_loadu_ps +#define _simd_setzero_ps _mm256_setzero_ps +#define _simd_set1_ps _mm256_set1_ps +#define _simd_blend_ps _mm256_blend_ps +#define _simd_blendv_ps _mm256_blendv_ps +#define _simd_store_ps _mm256_store_ps +#define _simd_mul_ps _mm256_mul_ps +#define _simd_add_ps _mm256_add_ps +#define _simd_sub_ps _mm256_sub_ps +#define _simd_rsqrt_ps _mm256_rsqrt_ps +#define _simd_min_ps _mm256_min_ps +#define _simd_max_ps _mm256_max_ps +#define _simd_movemask_ps _mm256_movemask_ps +#define _simd_cvtps_epi32 _mm256_cvtps_epi32 +#define _simd_cvttps_epi32 _mm256_cvttps_epi32 +#define _simd_cvtepi32_ps _mm256_cvtepi32_ps +#define _simd_cmplt_ps(a, b) _mm256_cmp_ps(a, b, _CMP_LT_OQ) +#define _simd_cmpgt_ps(a, b) _mm256_cmp_ps(a, b, _CMP_GT_OQ) +#define _simd_cmpneq_ps(a, b) _mm256_cmp_ps(a, b, _CMP_NEQ_OQ) +#define _simd_cmpeq_ps(a, b) _mm256_cmp_ps(a, b, _CMP_EQ_OQ) +#define _simd_cmpge_ps(a, b) _mm256_cmp_ps(a, b, _CMP_GE_OQ) +#define _simd_cmple_ps(a, b) _mm256_cmp_ps(a, b, _CMP_LE_OQ) +#define _simd_cmp_ps(a, b, imm) _mm256_cmp_ps(a, b, imm) +#define _simd_and_ps _mm256_and_ps +#define _simd_or_ps _mm256_or_ps + +#define _simd_rcp_ps _mm256_rcp_ps +#define _simd_div_ps _mm256_div_ps +#define _simd_castsi_ps _mm256_castsi256_ps +#define _simd_andnot_ps _mm256_andnot_ps +#define _simd_round_ps _mm256_round_ps +#define _simd_castpd_ps _mm256_castpd_ps +#define _simd_broadcast_ps(a) _mm256_broadcast_ps((const __m128*)(a)) + +#define _simd_load_sd _mm256_load_sd +#define _simd_movemask_pd _mm256_movemask_pd +#define _simd_castsi_pd _mm256_castsi256_pd + +// emulated integer simd +#define SIMD_EMU_EPI(func, intrin) \ +INLINE \ +__m256i func(__m256i a, __m256i b)\ +{\ + __m128i aHi = _mm256_extractf128_si256(a, 1);\ + __m128i bHi = _mm256_extractf128_si256(b, 1);\ + __m128i aLo = _mm256_castsi256_si128(a);\ + __m128i bLo = _mm256_castsi256_si128(b);\ +\ + __m128i subLo = intrin(aLo, bLo);\ + __m128i subHi = intrin(aHi, bHi);\ +\ + __m256i result = _mm256_castsi128_si256(subLo);\ + result = _mm256_insertf128_si256(result, subHi, 1);\ +\ + return result;\ +} + +#if (KNOB_ARCH == KNOB_ARCH_AVX) +#define _simd_mul_epi32 _simdemu_mul_epi32 +#define _simd_mullo_epi32 _simdemu_mullo_epi32 +#define _simd_sub_epi32 _simdemu_sub_epi32 +#define _simd_sub_epi64 _simdemu_sub_epi64 +#define _simd_min_epi32 _simdemu_min_epi32 +#define _simd_min_epu32 _simdemu_min_epu32 +#define _simd_max_epi32 _simdemu_max_epi32 +#define _simd_max_epu32 _simdemu_max_epu32 +#define _simd_add_epi32 _simdemu_add_epi32 +#define _simd_and_si _simdemu_and_si +#define _simd_andnot_si _simdemu_andnot_si +#define _simd_cmpeq_epi32 _simdemu_cmpeq_epi32 +#define _simd_cmplt_epi32 _simdemu_cmplt_epi32 +#define _simd_cmpgt_epi32 _simdemu_cmpgt_epi32 +#define _simd_or_si _simdemu_or_si +#define _simd_castps_si _mm256_castps_si256 +#define _simd_adds_epu8 _simdemu_adds_epu8 +#define _simd_subs_epu8 _simdemu_subs_epu8 +#define _simd_add_epi8 _simdemu_add_epi8 +#define _simd_cmpeq_epi64 _simdemu_cmpeq_epi64 +#define _simd_cmpgt_epi64 _simdemu_cmpgt_epi64 +#define _simd_movemask_epi8 _simdemu_movemask_epi8 + +SIMD_EMU_EPI(_simdemu_mul_epi32, _mm_mul_epi32) +SIMD_EMU_EPI(_simdemu_mullo_epi32, _mm_mullo_epi32) +SIMD_EMU_EPI(_simdemu_sub_epi32, _mm_sub_epi32) +SIMD_EMU_EPI(_simdemu_sub_epi64, _mm_sub_epi64) +SIMD_EMU_EPI(_simdemu_min_epi32, _mm_min_epi32) +SIMD_EMU_EPI(_simdemu_min_epu32, _mm_min_epu32) +SIMD_EMU_EPI(_simdemu_max_epi32, _mm_max_epi32) +SIMD_EMU_EPI(_simdemu_max_epu32, _mm_max_epu32) +SIMD_EMU_EPI(_simdemu_add_epi32, _mm_add_epi32) +SIMD_EMU_EPI(_simdemu_and_si, _mm_and_si128) +SIMD_EMU_EPI(_simdemu_andnot_si, _mm_andnot_si128) +SIMD_EMU_EPI(_simdemu_cmpeq_epi32, _mm_cmpeq_epi32) +SIMD_EMU_EPI(_simdemu_cmplt_epi32, _mm_cmplt_epi32) +SIMD_EMU_EPI(_simdemu_cmpgt_epi32, _mm_cmpgt_epi32) +SIMD_EMU_EPI(_simdemu_or_si, _mm_or_si128) +SIMD_EMU_EPI(_simdemu_adds_epu8, _mm_adds_epu8) +SIMD_EMU_EPI(_simdemu_subs_epu8, _mm_subs_epu8) +SIMD_EMU_EPI(_simdemu_add_epi8, _mm_add_epi8) +SIMD_EMU_EPI(_simdemu_cmpeq_epi64, _mm_cmpeq_epi64) +SIMD_EMU_EPI(_simdemu_cmpgt_epi64, _mm_cmpgt_epi64) + +#define _simd_unpacklo_epi32(a, b) _mm256_castps_si256(_mm256_unpacklo_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b))) +#define _simd_unpackhi_epi32(a, b) _mm256_castps_si256(_mm256_unpackhi_ps(_mm256_castsi256_ps(a), _mm256_castsi256_ps(b))) + +#define _simd_slli_epi32(a,i) _simdemu_slli_epi32(a,i) +#define _simd_srai_epi32(a,i) _simdemu_srai_epi32(a,i) +#define _simd_srli_epi32(a,i) _simdemu_srli_epi32(a,i) +#define _simd_srlisi_ps(a,i) _mm256_castsi256_ps(_simdemu_srli_si128(_mm256_castps_si256(a))) + +#define _simd128_fmadd_ps _mm_fmaddemu_ps +#define _simd_fmadd_ps _mm_fmaddemu256_ps +#define _simd_fmsub_ps _mm_fmsubemu256_ps +#define _simd_shuffle_epi8 _simdemu_shuffle_epi8 +SIMD_EMU_EPI(_simdemu_shuffle_epi8, _mm_shuffle_epi8) + +INLINE +__m128 _mm_fmaddemu_ps(__m128 a, __m128 b, __m128 c) +{ + __m128 res = _mm_mul_ps(a, b); + res = _mm_add_ps(res, c); + return res; +} + +INLINE +__m256 _mm_fmaddemu256_ps(__m256 a, __m256 b, __m256 c) +{ + __m256 res = _mm256_mul_ps(a, b); + res = _mm256_add_ps(res, c); + return res; +} + +INLINE +__m256 _mm_fmsubemu256_ps(__m256 a, __m256 b, __m256 c) +{ + __m256 res = _mm256_mul_ps(a, b); + res = _mm256_sub_ps(res, c); + return res; +} + +INLINE +__m256 _simd_i32gather_ps(const float* pBase, __m256i vOffsets, const int scale) +{ + uint32_t *pOffsets = (uint32_t*)&vOffsets; + simdscalar vResult; + float* pResult = (float*)&vResult; + for (uint32_t i = 0; i < KNOB_SIMD_WIDTH; ++i) + { + uint32_t offset = pOffsets[i]; + offset = offset * scale; + pResult[i] = *(float*)(((const uint8_t*)pBase + offset)); + } + + return vResult; +} + +INLINE +__m256 _simd_mask_i32gather_ps(__m256 vSrc, const float* pBase, __m256i vOffsets, __m256 vMask, const int scale) +{ + uint32_t *pOffsets = (uint32_t*)&vOffsets; + simdscalar vResult = vSrc; + float* pResult = (float*)&vResult; + DWORD index; + uint32_t mask = _simd_movemask_ps(vMask); + while (_BitScanForward(&index, mask)) + { + mask &= ~(1 << index); + uint32_t offset = pOffsets[index]; + offset = offset * scale; + pResult[index] = *(float*)(((const uint8_t*)pBase + offset)); + } + + return vResult; +} + +INLINE +__m256i _simd_abs_epi32(__m256i a) +{ + __m128i aHi = _mm256_extractf128_si256(a, 1); + __m128i aLo = _mm256_castsi256_si128(a); + __m128i absLo = _mm_abs_epi32(aLo); + __m128i absHi = _mm_abs_epi32(aHi); + __m256i result = _mm256_castsi128_si256(absLo); + result = _mm256_insertf128_si256(result, absHi, 1); + return result; +} + +INLINE +int _simdemu_movemask_epi8(__m256i a) +{ + __m128i aHi = _mm256_extractf128_si256(a, 1); + __m128i aLo = _mm256_castsi256_si128(a); + + int resHi = _mm_movemask_epi8(aHi); + int resLo = _mm_movemask_epi8(aLo); + + return (resHi << 16) | resLo; +} +#else + +#define _simd_mul_epi32 _mm256_mul_epi32 +#define _simd_mullo_epi32 _mm256_mullo_epi32 +#define _simd_sub_epi32 _mm256_sub_epi32 +#define _simd_sub_epi64 _mm256_sub_epi64 +#define _simd_min_epi32 _mm256_min_epi32 +#define _simd_max_epi32 _mm256_max_epi32 +#define _simd_min_epu32 _mm256_min_epu32 +#define _simd_max_epu32 _mm256_max_epu32 +#define _simd_add_epi32 _mm256_add_epi32 +#define _simd_and_si _mm256_and_si256 +#define _simd_andnot_si _mm256_andnot_si256 +#define _simd_cmpeq_epi32 _mm256_cmpeq_epi32 +#define _simd_cmplt_epi32(a,b) _mm256_cmpgt_epi32(b,a) +#define _simd_cmpgt_epi32(a,b) _mm256_cmpgt_epi32(a,b) +#define _simd_or_si _mm256_or_si256 +#define _simd_castps_si _mm256_castps_si256 + +#define _simd_unpacklo_epi32 _mm256_unpacklo_epi32 +#define _simd_unpackhi_epi32 _mm256_unpackhi_epi32 + +#define _simd_srli_si(a,i) _simdemu_srli_si128(a) +#define _simd_slli_epi32 _mm256_slli_epi32 +#define _simd_srai_epi32 _mm256_srai_epi32 +#define _simd_srli_epi32 _mm256_srli_epi32 +#define _simd_srlisi_ps(a,i) _mm256_castsi256_ps(_simdemu_srli_si128(_mm256_castps_si256(a))) +#define _simd128_fmadd_ps _mm_fmadd_ps +#define _simd_fmadd_ps _mm256_fmadd_ps +#define _simd_fmsub_ps _mm256_fmsub_ps +#define _simd_shuffle_epi8 _mm256_shuffle_epi8 +#define _simd_adds_epu8 _mm256_adds_epu8 +#define _simd_subs_epu8 _mm256_subs_epu8 +#define _simd_add_epi8 _mm256_add_epi8 +#define _simd_i32gather_ps _mm256_i32gather_ps +#define _simd_mask_i32gather_ps _mm256_mask_i32gather_ps +#define _simd_abs_epi32 _mm256_abs_epi32 + +#define _simd_cmpeq_epi64 _mm256_cmpeq_epi64 +#define _simd_cmpgt_epi64 _mm256_cmpgt_epi64 +#define _simd_movemask_epi8 _mm256_movemask_epi8 +#endif + +#define _simd_shuffleps_epi32(vA, vB, imm) _mm256_castps_si256(_mm256_shuffle_ps(_mm256_castsi256_ps(vA), _mm256_castsi256_ps(vB), imm)) +#define _simd_shuffle_ps _mm256_shuffle_ps +#define _simd_set1_epi32 _mm256_set1_epi32 +#define _simd_set1_epi8 _mm256_set1_epi8 +#define _simd_setzero_si _mm256_setzero_si256 +#define _simd_cvttps_epi32 _mm256_cvttps_epi32 +#define _simd_store_si _mm256_store_si256 +#define _simd_broadcast_ss _mm256_broadcast_ss +#define _simd_maskstore_ps _mm256_maskstore_ps +#define _simd_load_si _mm256_load_si256 +#define _simd_loadu_si _mm256_loadu_si256 +#define _simd_sub_ps _mm256_sub_ps +#define _simd_testz_ps _mm256_testz_ps +#define _simd_xor_ps _mm256_xor_ps + + +INLINE +simdscalari _simd_blendv_epi32(simdscalari a, simdscalari b, simdscalar mask) +{ + return _simd_castps_si(_simd_blendv_ps(_simd_castsi_ps(a), _simd_castsi_ps(b), mask)); +} + +// convert bitmask to vector mask +INLINE +simdscalar vMask(int32_t mask) +{ + __m256i vec = _mm256_set1_epi32(mask); + const __m256i bit = _mm256_set_epi32(0x80, 0x40, 0x20, 0x10, 0x08, 0x04, 0x02, 0x01); + vec = _simd_and_si(vec, bit); + vec = _simd_cmplt_epi32(_mm256_setzero_si256(), vec); + return _simd_castsi_ps(vec); +} + +INLINE +void _simd_mov(simdscalar &r, unsigned int rlane, simdscalar& s, unsigned int slane) +{ + OSALIGNSIMD(float) rArray[KNOB_SIMD_WIDTH], sArray[KNOB_SIMD_WIDTH]; + _mm256_store_ps(rArray, r); + _mm256_store_ps(sArray, s); + rArray[rlane] = sArray[slane]; + r = _mm256_load_ps(rArray); +} + +INLINE __m256i _simdemu_slli_epi32(__m256i a, uint32_t i) +{ + __m128i aHi = _mm256_extractf128_si256(a, 1); + __m128i aLo = _mm256_castsi256_si128(a); + + __m128i resHi = _mm_slli_epi32(aHi, i); + __m128i resLo = _mm_slli_epi32(aLo, i); + + __m256i result = _mm256_castsi128_si256(resLo); + result = _mm256_insertf128_si256(result, resHi, 1); + + return result; +} + +INLINE __m256i _simdemu_srai_epi32(__m256i a, uint32_t i) +{ + __m128i aHi = _mm256_extractf128_si256(a, 1); + __m128i aLo = _mm256_castsi256_si128(a); + + __m128i resHi = _mm_srai_epi32(aHi, i); + __m128i resLo = _mm_srai_epi32(aLo, i); + + __m256i result = _mm256_castsi128_si256(resLo); + result = _mm256_insertf128_si256(result, resHi, 1); + + return result; +} + +INLINE __m256i _simdemu_srli_epi32(__m256i a, uint32_t i) +{ + __m128i aHi = _mm256_extractf128_si256(a, 1); + __m128i aLo = _mm256_castsi256_si128(a); + + __m128i resHi = _mm_srli_epi32(aHi, i); + __m128i resLo = _mm_srli_epi32(aLo, i); + + __m256i result = _mm256_castsi128_si256(resLo); + result = _mm256_insertf128_si256(result, resHi, 1); + + return result; +} + +INLINE +void _simdvec_transpose(simdvector &v) +{ + SWR_ASSERT(false, "Need to implement 8 wide version"); +} + +#else +#error Unsupported vector width +#endif + +// Populates a simdvector from a vector. So p = xyzw becomes xxxx yyyy zzzz wwww. +INLINE +void _simdvec_load_ps(simdvector& r, const float *p) +{ + r[0] = _simd_set1_ps(p[0]); + r[1] = _simd_set1_ps(p[1]); + r[2] = _simd_set1_ps(p[2]); + r[3] = _simd_set1_ps(p[3]); +} + +INLINE +void _simdvec_mov(simdvector& r, const simdscalar& s) +{ + r[0] = s; + r[1] = s; + r[2] = s; + r[3] = s; +} + +INLINE +void _simdvec_mov(simdvector& r, const simdvector& v) +{ + r[0] = v[0]; + r[1] = v[1]; + r[2] = v[2]; + r[3] = v[3]; +} + +// just move a lane from the source simdvector to dest simdvector +INLINE +void _simdvec_mov(simdvector &r, unsigned int rlane, simdvector& s, unsigned int slane) +{ + _simd_mov(r[0], rlane, s[0], slane); + _simd_mov(r[1], rlane, s[1], slane); + _simd_mov(r[2], rlane, s[2], slane); + _simd_mov(r[3], rlane, s[3], slane); +} + +INLINE +void _simdvec_dp3_ps(simdscalar& r, const simdvector& v0, const simdvector& v1) +{ + simdscalar tmp; + r = _simd_mul_ps(v0[0], v1[0]); // (v0.x*v1.x) + + tmp = _simd_mul_ps(v0[1], v1[1]); // (v0.y*v1.y) + r = _simd_add_ps(r, tmp); // (v0.x*v1.x) + (v0.y*v1.y) + + tmp = _simd_mul_ps(v0[2], v1[2]); // (v0.z*v1.z) + r = _simd_add_ps(r, tmp); // (v0.x*v1.x) + (v0.y*v1.y) + (v0.z*v1.z) +} + +INLINE +void _simdvec_dp4_ps(simdscalar& r, const simdvector& v0, const simdvector& v1) +{ + simdscalar tmp; + r = _simd_mul_ps(v0[0], v1[0]); // (v0.x*v1.x) + + tmp = _simd_mul_ps(v0[1], v1[1]); // (v0.y*v1.y) + r = _simd_add_ps(r, tmp); // (v0.x*v1.x) + (v0.y*v1.y) + + tmp = _simd_mul_ps(v0[2], v1[2]); // (v0.z*v1.z) + r = _simd_add_ps(r, tmp); // (v0.x*v1.x) + (v0.y*v1.y) + (v0.z*v1.z) + + tmp = _simd_mul_ps(v0[3], v1[3]); // (v0.w*v1.w) + r = _simd_add_ps(r, tmp); // (v0.x*v1.x) + (v0.y*v1.y) + (v0.z*v1.z) +} + +INLINE +simdscalar _simdvec_rcp_length_ps(const simdvector& v) +{ + simdscalar length; + _simdvec_dp4_ps(length, v, v); + return _simd_rsqrt_ps(length); +} + +INLINE +void _simdvec_normalize_ps(simdvector& r, const simdvector& v) +{ + simdscalar vecLength; + vecLength = _simdvec_rcp_length_ps(v); + + r[0] = _simd_mul_ps(v[0], vecLength); + r[1] = _simd_mul_ps(v[1], vecLength); + r[2] = _simd_mul_ps(v[2], vecLength); + r[3] = _simd_mul_ps(v[3], vecLength); +} + +INLINE +void _simdvec_mul_ps(simdvector& r, const simdvector& v, const simdscalar& s) +{ + r[0] = _simd_mul_ps(v[0], s); + r[1] = _simd_mul_ps(v[1], s); + r[2] = _simd_mul_ps(v[2], s); + r[3] = _simd_mul_ps(v[3], s); +} + +INLINE +void _simdvec_mul_ps(simdvector& r, const simdvector& v0, const simdvector& v1) +{ + r[0] = _simd_mul_ps(v0[0], v1[0]); + r[1] = _simd_mul_ps(v0[1], v1[1]); + r[2] = _simd_mul_ps(v0[2], v1[2]); + r[3] = _simd_mul_ps(v0[3], v1[3]); +} + +INLINE +void _simdvec_add_ps(simdvector& r, const simdvector& v0, const simdvector& v1) +{ + r[0] = _simd_add_ps(v0[0], v1[0]); + r[1] = _simd_add_ps(v0[1], v1[1]); + r[2] = _simd_add_ps(v0[2], v1[2]); + r[3] = _simd_add_ps(v0[3], v1[3]); +} + +INLINE +void _simdvec_min_ps(simdvector& r, const simdvector& v0, const simdscalar& s) +{ + r[0] = _simd_min_ps(v0[0], s); + r[1] = _simd_min_ps(v0[1], s); + r[2] = _simd_min_ps(v0[2], s); + r[3] = _simd_min_ps(v0[3], s); +} + +INLINE +void _simdvec_max_ps(simdvector& r, const simdvector& v0, const simdscalar& s) +{ + r[0] = _simd_max_ps(v0[0], s); + r[1] = _simd_max_ps(v0[1], s); + r[2] = _simd_max_ps(v0[2], s); + r[3] = _simd_max_ps(v0[3], s); +} + +// Matrix4x4 * Vector4 +// outVec.x = (m00 * v.x) + (m01 * v.y) + (m02 * v.z) + (m03 * v.w) +// outVec.y = (m10 * v.x) + (m11 * v.y) + (m12 * v.z) + (m13 * v.w) +// outVec.z = (m20 * v.x) + (m21 * v.y) + (m22 * v.z) + (m23 * v.w) +// outVec.w = (m30 * v.x) + (m31 * v.y) + (m32 * v.z) + (m33 * v.w) +INLINE +void _simd_mat4x4_vec4_multiply( + simdvector& result, + const float *pMatrix, + const simdvector& v) +{ + simdscalar m; + simdscalar r0; + simdscalar r1; + + m = _simd_load1_ps(pMatrix + 0*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 0*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 0*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + m = _simd_load1_ps(pMatrix + 0*4 + 3); // m[row][3] + r1 = _simd_mul_ps(m, v[3]); // (m3 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * v.w) + result[0] = r0; + + m = _simd_load1_ps(pMatrix + 1*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 1*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 1*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + m = _simd_load1_ps(pMatrix + 1*4 + 3); // m[row][3] + r1 = _simd_mul_ps(m, v[3]); // (m3 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * v.w) + result[1] = r0; + + m = _simd_load1_ps(pMatrix + 2*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 2*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 2*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + m = _simd_load1_ps(pMatrix + 2*4 + 3); // m[row][3] + r1 = _simd_mul_ps(m, v[3]); // (m3 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * v.w) + result[2] = r0; + + m = _simd_load1_ps(pMatrix + 3*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 3*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 3*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + m = _simd_load1_ps(pMatrix + 3*4 + 3); // m[row][3] + r1 = _simd_mul_ps(m, v[3]); // (m3 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * v.w) + result[3] = r0; +} + +// Matrix4x4 * Vector3 - Direction Vector where w = 0. +// outVec.x = (m00 * v.x) + (m01 * v.y) + (m02 * v.z) + (m03 * 0) +// outVec.y = (m10 * v.x) + (m11 * v.y) + (m12 * v.z) + (m13 * 0) +// outVec.z = (m20 * v.x) + (m21 * v.y) + (m22 * v.z) + (m23 * 0) +// outVec.w = (m30 * v.x) + (m31 * v.y) + (m32 * v.z) + (m33 * 0) +INLINE +void _simd_mat3x3_vec3_w0_multiply( + simdvector& result, + const float *pMatrix, + const simdvector& v) +{ + simdscalar m; + simdscalar r0; + simdscalar r1; + + m = _simd_load1_ps(pMatrix + 0*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 0*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 0*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + result[0] = r0; + + m = _simd_load1_ps(pMatrix + 1*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 1*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 1*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + result[1] = r0; + + m = _simd_load1_ps(pMatrix + 2*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 2*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 2*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + result[2] = r0; + + result[3] = _simd_setzero_ps(); +} + +// Matrix4x4 * Vector3 - Position vector where w = 1. +// outVec.x = (m00 * v.x) + (m01 * v.y) + (m02 * v.z) + (m03 * 1) +// outVec.y = (m10 * v.x) + (m11 * v.y) + (m12 * v.z) + (m13 * 1) +// outVec.z = (m20 * v.x) + (m21 * v.y) + (m22 * v.z) + (m23 * 1) +// outVec.w = (m30 * v.x) + (m31 * v.y) + (m32 * v.z) + (m33 * 1) +INLINE +void _simd_mat4x4_vec3_w1_multiply( + simdvector& result, + const float *pMatrix, + const simdvector& v) +{ + simdscalar m; + simdscalar r0; + simdscalar r1; + + m = _simd_load1_ps(pMatrix + 0*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 0*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 0*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + m = _simd_load1_ps(pMatrix + 0*4 + 3); // m[row][3] + r0 = _simd_add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1) + result[0] = r0; + + m = _simd_load1_ps(pMatrix + 1*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 1*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 1*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + m = _simd_load1_ps(pMatrix + 1*4 + 3); // m[row][3] + r0 = _simd_add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1) + result[1] = r0; + + m = _simd_load1_ps(pMatrix + 2*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 2*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 2*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + m = _simd_load1_ps(pMatrix + 2*4 + 3); // m[row][3] + r0 = _simd_add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1) + result[2] = r0; + + m = _simd_load1_ps(pMatrix + 3*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 3*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 3*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + m = _simd_load1_ps(pMatrix + 3*4 + 3); // m[row][3] + result[3] = _simd_add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1) +} + +INLINE +void _simd_mat4x3_vec3_w1_multiply( + simdvector& result, + const float *pMatrix, + const simdvector& v) +{ + simdscalar m; + simdscalar r0; + simdscalar r1; + + m = _simd_load1_ps(pMatrix + 0*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 0*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 0*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + m = _simd_load1_ps(pMatrix + 0*4 + 3); // m[row][3] + r0 = _simd_add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1) + result[0] = r0; + + m = _simd_load1_ps(pMatrix + 1*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 1*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 1*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + m = _simd_load1_ps(pMatrix + 1*4 + 3); // m[row][3] + r0 = _simd_add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1) + result[1] = r0; + + m = _simd_load1_ps(pMatrix + 2*4 + 0); // m[row][0] + r0 = _simd_mul_ps(m, v[0]); // (m00 * v.x) + m = _simd_load1_ps(pMatrix + 2*4 + 1); // m[row][1] + r1 = _simd_mul_ps(m, v[1]); // (m1 * v.y) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + m = _simd_load1_ps(pMatrix + 2*4 + 2); // m[row][2] + r1 = _simd_mul_ps(m, v[2]); // (m2 * v.z) + r0 = _simd_add_ps(r0, r1); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + m = _simd_load1_ps(pMatrix + 2*4 + 3); // m[row][3] + r0 = _simd_add_ps(r0, m); // (m0 * v.x) + (m1 * v.y) + (m2 * v.z) + (m2 * 1) + result[2] = r0; + result[3] = _simd_set1_ps(1.0f); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Compute plane equation vA * vX + vB * vY + vC +INLINE simdscalar vplaneps(simdscalar vA, simdscalar vB, simdscalar vC, simdscalar &vX, simdscalar &vY) +{ + simdscalar vOut = _simd_fmadd_ps(vA, vX, vC); + vOut = _simd_fmadd_ps(vB, vY, vOut); + return vOut; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Interpolates a single component. +/// @param vI - barycentric I +/// @param vJ - barycentric J +/// @param pInterpBuffer - pointer to attribute barycentric coeffs +template +static INLINE simdscalar InterpolateComponent(simdscalar vI, simdscalar vJ, const float *pInterpBuffer) +{ + const float *pInterpA = &pInterpBuffer[Attrib * 12 + 0 + Comp]; + const float *pInterpB = &pInterpBuffer[Attrib * 12 + 4 + Comp]; + const float *pInterpC = &pInterpBuffer[Attrib * 12 + 8 + Comp]; + + simdscalar vA = _simd_broadcast_ss(pInterpA); + simdscalar vB = _simd_broadcast_ss(pInterpB); + simdscalar vC = _simd_broadcast_ss(pInterpC); + + simdscalar vk = _simd_sub_ps(_simd_sub_ps(_simd_set1_ps(1.0f), vI), vJ); + vC = _simd_mul_ps(vk, vC); + + return vplaneps(vA, vB, vC, vI, vJ); +} + + +#endif//__SWR_SIMDINTRIN_H__ diff --git a/src/gallium/drivers/swr/rasterizer/common/swr_assert.cpp b/src/gallium/drivers/swr/rasterizer/common/swr_assert.cpp new file mode 100644 index 00000000000..0bffd2c8000 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/common/swr_assert.cpp @@ -0,0 +1,238 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +****************************************************************************/ + +#include "common/os.h" +#include +#include +#include + +#if SWR_ENABLE_ASSERTS || SWR_ENABLE_REL_ASSERTS + +#if defined(_WIN32) +#pragma comment(lib, "user32.lib") +#endif // _WIN32 + +enum TextColor +{ + TEXT_BLACK = 0, + TEXT_RED = 1, + TEXT_GREEN = 2, + TEXT_BLUE = 4, + TEXT_PURPLE = TEXT_RED | TEXT_BLUE, + TEXT_CYAN = TEXT_GREEN | TEXT_BLUE, + TEXT_YELLOW = TEXT_RED | TEXT_GREEN, + TEXT_WHITE = TEXT_RED | TEXT_GREEN | TEXT_BLUE, +}; + +enum TextStyle +{ + TEXT_NORMAL = 0, + TEXT_INTENSITY = 1, +}; + +void SetTextColor(FILE* stream, TextColor color = TEXT_WHITE, TextStyle style = TEXT_NORMAL) +{ +#if defined(_WIN32) + + HANDLE hConsoleHandle = nullptr; + if (stream == stderr) + { + hConsoleHandle = GetStdHandle(STD_ERROR_HANDLE); + } + else if (stream == stdout) + { + hConsoleHandle = GetStdHandle(STD_OUTPUT_HANDLE); + } + else + { + // Not a console stream, do nothing + return; + } + + WORD textAttributes = 0; + if (color & TEXT_RED) + { + textAttributes |= FOREGROUND_RED; + } + if (color & TEXT_GREEN) + { + textAttributes |= FOREGROUND_GREEN; + } + if (color & TEXT_BLUE) + { + textAttributes |= FOREGROUND_BLUE; + } + if (style & TEXT_INTENSITY) + { + textAttributes |= FOREGROUND_INTENSITY; + } + SetConsoleTextAttribute(hConsoleHandle, textAttributes); + +#else // !_WIN32 + + // Print ANSI codes + uint32_t cc = 30 + (style ? 60 : 0) + color; + fprintf(stream, "\033[0m\033[%d;%dm", style, cc); + +#endif +} + +void ResetTextColor(FILE* stream) +{ +#if defined(_WIN32) + + SetTextColor(stream); + +#else // !_WIN32 + + // Print ANSI codes + fprintf(stream, "\033[0m"); + +#endif +} + +bool SwrAssert( + bool chkDebugger, + bool& enabled, + const char* pExpression, + const char* pFileName, + uint32_t lineNum, + const char* pFunction, + const char* pFmtString /* = nullptr */, + ...) +{ + if (!enabled) return false; + + SetTextColor(stderr, TEXT_CYAN, TEXT_NORMAL); + + fprintf(stderr, "%s(%d): ", pFileName, lineNum); + + SetTextColor(stderr, TEXT_RED, TEXT_INTENSITY); + + fprintf(stderr, "ASSERT: %s\n", pExpression); + + SetTextColor(stderr, TEXT_CYAN, TEXT_INTENSITY); + fprintf(stderr, "\t%s\n", pFunction); + + if (pFmtString) + { + SetTextColor(stderr, TEXT_YELLOW, TEXT_INTENSITY); + fprintf(stderr, "\t"); + va_list args; + va_start(args, pFmtString); + vfprintf(stderr, pFmtString, args); + va_end(args); + fprintf(stderr, "\n"); + } + ResetTextColor(stderr); + fflush(stderr); + +#if defined(_WIN32) + static const int MAX_MESSAGE_LEN = 2048; + char msgBuf[MAX_MESSAGE_LEN]; + + sprintf_s(msgBuf, "%s(%d): ASSERT: %s\n", pFileName, lineNum, pExpression); + msgBuf[MAX_MESSAGE_LEN - 2] = '\n'; + msgBuf[MAX_MESSAGE_LEN - 1] = 0; + OutputDebugStringA(msgBuf); + + sprintf_s(msgBuf, "\t%s\n", pFunction); + msgBuf[MAX_MESSAGE_LEN - 2] = '\n'; + msgBuf[MAX_MESSAGE_LEN - 1] = 0; + OutputDebugStringA(msgBuf); + + int offset = 0; + + if (pFmtString) + { + va_list args; + va_start(args, pFmtString); + offset = _vsnprintf_s( + msgBuf, + sizeof(msgBuf), + sizeof(msgBuf), + pFmtString, + args); + va_end(args); + + if (offset < 0) { return true; } + + OutputDebugStringA("\t"); + OutputDebugStringA(msgBuf); + OutputDebugStringA("\n"); + } + + if (KNOB_ENABLE_ASSERT_DIALOGS) + { + int retval = sprintf_s( + &msgBuf[offset], + MAX_MESSAGE_LEN - offset, + "\n\n" + "File: %s\n" + "Line: %d\n" + "\n" + "Expression: %s\n\n" + "Cancel: Disable this assert for the remainder of the process\n" + "Try Again: Break into the debugger\n" + "Continue: Continue execution (but leave assert enabled)", + pFileName, + lineNum, + pExpression); + + if (retval < 0) { return true; } + + offset += retval; + + if (!IsDebuggerPresent()) + { + sprintf_s( + &msgBuf[offset], + MAX_MESSAGE_LEN - offset, + "\n\n*** NO DEBUGGER DETECTED ***\n\nPressing \"Try Again\" will cause a program crash!"); + } + + retval = MessageBoxA(nullptr, msgBuf, "Assert Failed", MB_CANCELTRYCONTINUE | MB_ICONEXCLAMATION); + + switch (retval) + { + case IDCANCEL: + enabled = false; + return false; + + case IDTRYAGAIN: + return true; + + case IDCONTINUE: + return false; + } + } + else + { + return IsDebuggerPresent() || !chkDebugger; + } +#endif // _WIN32 + + return true; +} + +#endif // SWR_ENABLE_ASSERTS diff --git a/src/gallium/drivers/swr/rasterizer/common/swr_assert.h b/src/gallium/drivers/swr/rasterizer/common/swr_assert.h new file mode 100644 index 00000000000..fecadb3d499 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/common/swr_assert.h @@ -0,0 +1,109 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +****************************************************************************/ + +#ifndef __SWR_ASSERT_H__ +#define __SWR_ASSERT_H__ + +#if !defined(__SWR_OS_H__) +#error swr_assert.h should not be included directly, please include "common/os.h" instead. +#endif + +#if !defined(SWR_ENABLE_ASSERTS) + +#if !defined(NDEBUG) +#define SWR_ENABLE_ASSERTS 1 +#else +#define SWR_ENABLE_ASSERTS 0 +#endif // _DEBUG + +#endif // SWR_ENABLE_ASSERTS + +#if !defined(SWR_ENABLE_REL_ASSERTS) +#define SWR_ENABLE_REL_ASSERTS 1 +#endif + +#if SWR_ENABLE_ASSERTS || SWR_ENABLE_REL_ASSERTS +#include "assert.h" + +#if !defined(__cplusplus) + +#pragma message("C++ is required for SWR Asserts, falling back to assert.h") + +#if SWR_ENABLE_ASSERTS +#define SWR_ASSERT(e, ...) assert(e) +#endif + +#if SWR_ENABLE_REL_ASSERTS +#define SWR_REL_ASSERT(e, ...) assert(e) +#endif + +#else + +#if SWR_ENABLE_ASSERTS +#if defined(assert) +#undef assert +#endif +#define assert(exp) SWR_ASSERT(exp) +#endif + +bool SwrAssert( + bool chkDebugger, + bool& enabled, + const char* pExpression, + const char* pFileName, + uint32_t lineNum, + const char* function, + const char* pFmtString = nullptr, + ...); + +#define _SWR_ASSERT(chkDebugger, e, ...) {\ + bool expFailed = !(e);\ + if (expFailed) {\ + static bool swrAssertEnabled = true;\ + expFailed = SwrAssert(chkDebugger, swrAssertEnabled, #e, __FILE__, __LINE__, __FUNCTION__, ##__VA_ARGS__);\ + if (expFailed) { DEBUGBREAK; }\ + }\ +} + +#if SWR_ENABLE_ASSERTS +#define SWR_ASSERT(e, ...) _SWR_ASSERT(true, e, ##__VA_ARGS__) +#endif + +#if SWR_ENABLE_REL_ASSERTS +#define SWR_REL_ASSERT(e, ...) _SWR_ASSERT(false, e, ##__VA_ARGS__) +#endif +#endif // C++ + +#endif // SWR_ENABLE_ASSERTS || SWR_ENABLE_REL_ASSERTS + +#if !SWR_ENABLE_ASSERTS +#define SWR_ASSERT(e, ...) +#endif + +#if !SWR_ENABLE_REL_ASSERTS +#define SWR_REL_ASSERT(e, ...) +#endif + +#define SWR_NOT_IMPL SWR_ASSERT(0, "%s not implemented", __FUNCTION__) + +#endif//__SWR_ASSERT_H__ diff --git a/src/gallium/drivers/swr/rasterizer/core/api.cpp b/src/gallium/drivers/swr/rasterizer/core/api.cpp new file mode 100644 index 00000000000..fccccab503c --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/api.cpp @@ -0,0 +1,1511 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file api.cpp +* +* @brief API implementation +* +******************************************************************************/ + +#include +#include +#include + +#include "core/api.h" +#include "core/backend.h" +#include "core/context.h" +#include "core/frontend.h" +#include "core/rasterizer.h" +#include "core/rdtsc_core.h" +#include "core/threads.h" +#include "core/tilemgr.h" +#include "core/clip.h" + +#include "common/simdintrin.h" +#include "common/os.h" + +void SetupDefaultState(SWR_CONTEXT *pContext); + +////////////////////////////////////////////////////////////////////////// +/// @brief Create SWR Context. +/// @param pCreateInfo - pointer to creation info. +HANDLE SwrCreateContext( + const SWR_CREATECONTEXT_INFO* pCreateInfo) +{ + RDTSC_RESET(); + RDTSC_INIT(0); + + void* pContextMem = _aligned_malloc(sizeof(SWR_CONTEXT), KNOB_SIMD_WIDTH * 4); + memset(pContextMem, 0, sizeof(SWR_CONTEXT)); + SWR_CONTEXT *pContext = new (pContextMem) SWR_CONTEXT(); + + pContext->driverType = pCreateInfo->driver; + pContext->privateStateSize = pCreateInfo->privateStateSize; + + pContext->dcRing = (DRAW_CONTEXT*)_aligned_malloc(sizeof(DRAW_CONTEXT)*KNOB_MAX_DRAWS_IN_FLIGHT, 64); + memset(pContext->dcRing, 0, sizeof(DRAW_CONTEXT)*KNOB_MAX_DRAWS_IN_FLIGHT); + + pContext->dsRing = (DRAW_STATE*)_aligned_malloc(sizeof(DRAW_STATE)*KNOB_MAX_DRAWS_IN_FLIGHT, 64); + memset(pContext->dsRing, 0, sizeof(DRAW_STATE)*KNOB_MAX_DRAWS_IN_FLIGHT); + + pContext->numSubContexts = pCreateInfo->maxSubContexts; + if (pContext->numSubContexts > 1) + { + pContext->subCtxSave = (DRAW_STATE*)_aligned_malloc(sizeof(DRAW_STATE) * pContext->numSubContexts, 64); + memset(pContext->subCtxSave, 0, sizeof(DRAW_STATE) * pContext->numSubContexts); + } + + for (uint32_t dc = 0; dc < KNOB_MAX_DRAWS_IN_FLIGHT; ++dc) + { + pContext->dcRing[dc].pArena = new Arena(); + pContext->dcRing[dc].inUse = false; + pContext->dcRing[dc].pTileMgr = new MacroTileMgr(*(pContext->dcRing[dc].pArena)); + pContext->dcRing[dc].pDispatch = new DispatchQueue(); /// @todo Could lazily allocate this if Dispatch seen. + + pContext->dsRing[dc].pArena = new Arena(); + } + + if (!KNOB_SINGLE_THREADED) + { + memset(&pContext->WaitLock, 0, sizeof(pContext->WaitLock)); + memset(&pContext->FifosNotEmpty, 0, sizeof(pContext->FifosNotEmpty)); + new (&pContext->WaitLock) std::mutex(); + new (&pContext->FifosNotEmpty) std::condition_variable(); + + CreateThreadPool(pContext, &pContext->threadPool); + } + + // Calling createThreadPool() above can set SINGLE_THREADED + if (KNOB_SINGLE_THREADED) + { + pContext->NumWorkerThreads = 1; + } + + // Allocate scratch space for workers. + ///@note We could lazily allocate this but its rather small amount of memory. + for (uint32_t i = 0; i < pContext->NumWorkerThreads; ++i) + { + ///@todo Use numa API for allocations using numa information from thread data (if exists). + pContext->pScratch[i] = (uint8_t*)_aligned_malloc((32 * 1024), KNOB_SIMD_WIDTH * 4); + } + + pContext->nextDrawId = 1; + pContext->DrawEnqueued = 1; + + // State setup AFTER context is fully initialized + SetupDefaultState(pContext); + + // initialize hot tile manager + pContext->pHotTileMgr = new HotTileMgr(); + + // initialize function pointer tables + InitClearTilesTable(); + + // initialize store tiles function + pContext->pfnLoadTile = pCreateInfo->pfnLoadTile; + pContext->pfnStoreTile = pCreateInfo->pfnStoreTile; + pContext->pfnClearTile = pCreateInfo->pfnClearTile; + + return (HANDLE)pContext; +} + +void SwrDestroyContext(HANDLE hContext) +{ + SWR_CONTEXT *pContext = (SWR_CONTEXT*)hContext; + DestroyThreadPool(pContext, &pContext->threadPool); + + // free the fifos + for (uint32_t i = 0; i < KNOB_MAX_DRAWS_IN_FLIGHT; ++i) + { + delete pContext->dcRing[i].pArena; + delete pContext->dsRing[i].pArena; + delete(pContext->dcRing[i].pTileMgr); + delete(pContext->dcRing[i].pDispatch); + } + + // Free scratch space. + for (uint32_t i = 0; i < pContext->NumWorkerThreads; ++i) + { + _aligned_free(pContext->pScratch[i]); + } + + _aligned_free(pContext->dcRing); + _aligned_free(pContext->dsRing); + _aligned_free(pContext->subCtxSave); + + delete(pContext->pHotTileMgr); + + pContext->~SWR_CONTEXT(); + _aligned_free((SWR_CONTEXT*)hContext); +} + +void CopyState(DRAW_STATE& dst, const DRAW_STATE& src) +{ + memcpy(&dst.state, &src.state, sizeof(API_STATE)); +} + +void WakeAllThreads(SWR_CONTEXT *pContext) +{ + pContext->FifosNotEmpty.notify_all(); +} + +bool StillDrawing(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC) +{ + // For single thread nothing should still be drawing. + if (KNOB_SINGLE_THREADED) { return false; } + + if (pDC->isCompute) + { + if (pDC->doneCompute) + { + pDC->inUse = false; + return false; + } + } + + // Check if backend work is done. First make sure all triangles have been binned. + if (pDC->doneFE == true) + { + // ensure workers have all moved passed this draw + if (pDC->threadsDoneFE != pContext->NumWorkerThreads) + { + return true; + } + + if (pDC->threadsDoneBE != pContext->NumWorkerThreads) + { + return true; + } + + pDC->inUse = false; // all work is done. + } + + return pDC->inUse; +} + +void QueueDraw(SWR_CONTEXT *pContext) +{ + SWR_ASSERT(pContext->pCurDrawContext->inUse == false); + pContext->pCurDrawContext->inUse = true; + + _ReadWriteBarrier(); + { + std::unique_lock lock(pContext->WaitLock); + pContext->DrawEnqueued++; + } + + if (KNOB_SINGLE_THREADED) + { + // flush denormals to 0 + uint32_t mxcsr = _mm_getcsr(); + _mm_setcsr(mxcsr | _MM_FLUSH_ZERO_ON | _MM_DENORMALS_ZERO_ON); + + std::unordered_set lockedTiles; + uint64_t curDraw[2] = { pContext->pCurDrawContext->drawId, pContext->pCurDrawContext->drawId }; + WorkOnFifoFE(pContext, 0, curDraw[0], 0); + WorkOnFifoBE(pContext, 0, curDraw[1], lockedTiles); + + // restore csr + _mm_setcsr(mxcsr); + } + else + { + RDTSC_START(APIDrawWakeAllThreads); + WakeAllThreads(pContext); + RDTSC_STOP(APIDrawWakeAllThreads, 1, 0); + } + + // Set current draw context to NULL so that next state call forces a new draw context to be created and populated. + pContext->pPrevDrawContext = pContext->pCurDrawContext; + pContext->pCurDrawContext = nullptr; +} + +///@todo Combine this with QueueDraw +void QueueDispatch(SWR_CONTEXT *pContext) +{ + SWR_ASSERT(pContext->pCurDrawContext->inUse == false); + pContext->pCurDrawContext->inUse = true; + + _ReadWriteBarrier(); + { + std::unique_lock lock(pContext->WaitLock); + pContext->DrawEnqueued++; + } + + if (KNOB_SINGLE_THREADED) + { + // flush denormals to 0 + uint32_t mxcsr = _mm_getcsr(); + _mm_setcsr(mxcsr | _MM_FLUSH_ZERO_ON | _MM_DENORMALS_ZERO_ON); + + uint64_t curDispatch = pContext->pCurDrawContext->drawId; + WorkOnCompute(pContext, 0, curDispatch); + + // restore csr + _mm_setcsr(mxcsr); + } + else + { + RDTSC_START(APIDrawWakeAllThreads); + WakeAllThreads(pContext); + RDTSC_STOP(APIDrawWakeAllThreads, 1, 0); + } + + // Set current draw context to NULL so that next state call forces a new draw context to be created and populated. + pContext->pPrevDrawContext = pContext->pCurDrawContext; + pContext->pCurDrawContext = nullptr; +} + +DRAW_CONTEXT* GetDrawContext(SWR_CONTEXT *pContext, bool isSplitDraw = false) +{ + RDTSC_START(APIGetDrawContext); + // If current draw context is null then need to obtain a new draw context to use from ring. + if (pContext->pCurDrawContext == nullptr) + { + uint32_t dcIndex = pContext->nextDrawId % KNOB_MAX_DRAWS_IN_FLIGHT; + + DRAW_CONTEXT* pCurDrawContext = &pContext->dcRing[dcIndex]; + pContext->pCurDrawContext = pCurDrawContext; + + // Need to wait until this draw context is available to use. + while (StillDrawing(pContext, pCurDrawContext)) + { + _mm_pause(); + } + + // Assign next available entry in DS ring to this DC. + uint32_t dsIndex = pContext->curStateId % KNOB_MAX_DRAWS_IN_FLIGHT; + pCurDrawContext->pState = &pContext->dsRing[dsIndex]; + + Arena& stateArena = *(pCurDrawContext->pState->pArena); + + // Copy previous state to current state. + if (pContext->pPrevDrawContext) + { + DRAW_CONTEXT* pPrevDrawContext = pContext->pPrevDrawContext; + + // If we're splitting our draw then we can just use the same state from the previous + // draw. In this case, we won't increment the DS ring index so the next non-split + // draw can receive the state. + if (isSplitDraw == false) + { + CopyState(*pCurDrawContext->pState, *pPrevDrawContext->pState); + + stateArena.Reset(true); // Reset memory. + pCurDrawContext->pState->pPrivateState = nullptr; + + pContext->curStateId++; // Progress state ring index forward. + } + else + { + // If its a split draw then just copy the state pointer over + // since its the same draw. + pCurDrawContext->pState = pPrevDrawContext->pState; + } + } + else + { + stateArena.Reset(); // Reset memory. + pContext->curStateId++; // Progress state ring index forward. + } + + pCurDrawContext->dependency = 0; + pCurDrawContext->pArena->Reset(); + pCurDrawContext->pContext = pContext; + pCurDrawContext->isCompute = false; // Dispatch has to set this to true. + pCurDrawContext->inUse = false; + + pCurDrawContext->doneCompute = false; + pCurDrawContext->doneFE = false; + pCurDrawContext->FeLock = 0; + pCurDrawContext->threadsDoneFE = 0; + pCurDrawContext->threadsDoneBE = 0; + + pCurDrawContext->pTileMgr->initialize(); + + // Assign unique drawId for this DC + pCurDrawContext->drawId = pContext->nextDrawId++; + } + else + { + SWR_ASSERT(isSplitDraw == false, "Split draw should only be used when obtaining a new DC"); + } + + RDTSC_STOP(APIGetDrawContext, 0, 0); + return pContext->pCurDrawContext; +} + +void SWR_API SwrSetActiveSubContext( + HANDLE hContext, + uint32_t subContextIndex) +{ + SWR_CONTEXT *pContext = (SWR_CONTEXT*)hContext; + if (subContextIndex >= pContext->numSubContexts) + { + return; + } + + if (subContextIndex != pContext->curSubCtxId) + { + // Save and restore draw state + DRAW_CONTEXT* pDC = GetDrawContext(pContext); + CopyState( + pContext->subCtxSave[pContext->curSubCtxId], + *(pDC->pState)); + + CopyState( + *(pDC->pState), + pContext->subCtxSave[subContextIndex]); + + pContext->curSubCtxId = subContextIndex; + } +} + +API_STATE* GetDrawState(SWR_CONTEXT *pContext) +{ + DRAW_CONTEXT* pDC = GetDrawContext(pContext); + SWR_ASSERT(pDC->pState != nullptr); + + return &pDC->pState->state; +} + +void SetupDefaultState(SWR_CONTEXT *pContext) +{ + API_STATE* pState = GetDrawState(pContext); + + pState->rastState.cullMode = SWR_CULLMODE_NONE; + pState->rastState.frontWinding = SWR_FRONTWINDING_CCW; +} + +static INLINE SWR_CONTEXT* GetContext(HANDLE hContext) +{ + return (SWR_CONTEXT*)hContext; +} + +void SwrSync(HANDLE hContext, PFN_CALLBACK_FUNC pfnFunc, uint64_t userData, uint64_t userData2, uint64_t userData3) +{ + RDTSC_START(APISync); + + SWR_ASSERT(pfnFunc != nullptr); + + SWR_CONTEXT *pContext = GetContext(hContext); + DRAW_CONTEXT* pDC = GetDrawContext(pContext); + + pDC->FeWork.type = SYNC; + pDC->FeWork.pfnWork = ProcessSync; + pDC->FeWork.desc.sync.pfnCallbackFunc = pfnFunc; + pDC->FeWork.desc.sync.userData = userData; + pDC->FeWork.desc.sync.userData2 = userData2; + pDC->FeWork.desc.sync.userData3 = userData3; + + // cannot execute until all previous draws have completed + pDC->dependency = pDC->drawId - 1; + + //enqueue + QueueDraw(pContext); + + RDTSC_STOP(APISync, 1, 0); +} + +void SwrWaitForIdle(HANDLE hContext) +{ + SWR_CONTEXT *pContext = GetContext(hContext); + + RDTSC_START(APIWaitForIdle); + // Wait for all work to complete. + for (uint32_t dc = 0; dc < KNOB_MAX_DRAWS_IN_FLIGHT; ++dc) + { + DRAW_CONTEXT *pDC = &pContext->dcRing[dc]; + + while (StillDrawing(pContext, pDC)) + { + _mm_pause(); + } + } + RDTSC_STOP(APIWaitForIdle, 1, 0); +} + +void SwrSetVertexBuffers( + HANDLE hContext, + uint32_t numBuffers, + const SWR_VERTEX_BUFFER_STATE* pVertexBuffers) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + + for (uint32_t i = 0; i < numBuffers; ++i) + { + const SWR_VERTEX_BUFFER_STATE *pVB = &pVertexBuffers[i]; + pState->vertexBuffers[pVB->index] = *pVB; + } +} + +void SwrSetIndexBuffer( + HANDLE hContext, + const SWR_INDEX_BUFFER_STATE* pIndexBuffer) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + + pState->indexBuffer = *pIndexBuffer; +} + +void SwrSetFetchFunc( + HANDLE hContext, + PFN_FETCH_FUNC pfnFetchFunc) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + + pState->pfnFetchFunc = pfnFetchFunc; +} + +void SwrSetSoFunc( + HANDLE hContext, + PFN_SO_FUNC pfnSoFunc, + uint32_t streamIndex) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + + SWR_ASSERT(streamIndex < MAX_SO_STREAMS); + + pState->pfnSoFunc[streamIndex] = pfnSoFunc; +} + +void SwrSetSoState( + HANDLE hContext, + SWR_STREAMOUT_STATE* pSoState) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + + pState->soState = *pSoState; +} + +void SwrSetSoBuffers( + HANDLE hContext, + SWR_STREAMOUT_BUFFER* pSoBuffer, + uint32_t slot) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + + SWR_ASSERT((slot < 4), "There are only 4 SO buffer slots [0, 3]\nSlot requested: %d", slot); + + pState->soBuffer[slot] = *pSoBuffer; +} + +void SwrSetVertexFunc( + HANDLE hContext, + PFN_VERTEX_FUNC pfnVertexFunc) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + + pState->pfnVertexFunc = pfnVertexFunc; +} + +void SwrSetFrontendState( + HANDLE hContext, + SWR_FRONTEND_STATE *pFEState) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + pState->frontendState = *pFEState; +} + +void SwrSetGsState( + HANDLE hContext, + SWR_GS_STATE *pGSState) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + pState->gsState = *pGSState; +} + +void SwrSetGsFunc( + HANDLE hContext, + PFN_GS_FUNC pfnGsFunc) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + pState->pfnGsFunc = pfnGsFunc; +} + +void SwrSetCsFunc( + HANDLE hContext, + PFN_CS_FUNC pfnCsFunc, + uint32_t totalThreadsInGroup) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + pState->pfnCsFunc = pfnCsFunc; + pState->totalThreadsInGroup = totalThreadsInGroup; +} + +void SwrSetTsState( + HANDLE hContext, + SWR_TS_STATE *pState) +{ + API_STATE* pApiState = GetDrawState(GetContext(hContext)); + pApiState->tsState = *pState; +} + +void SwrSetHsFunc( + HANDLE hContext, + PFN_HS_FUNC pfnFunc) +{ + API_STATE* pApiState = GetDrawState(GetContext(hContext)); + pApiState->pfnHsFunc = pfnFunc; +} + +void SwrSetDsFunc( + HANDLE hContext, + PFN_DS_FUNC pfnFunc) +{ + API_STATE* pApiState = GetDrawState(GetContext(hContext)); + pApiState->pfnDsFunc = pfnFunc; +} + +void SwrSetDepthStencilState( + HANDLE hContext, + SWR_DEPTH_STENCIL_STATE *pDSState) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + + pState->depthStencilState = *pDSState; +} + +void SwrSetBackendState( + HANDLE hContext, + SWR_BACKEND_STATE *pBEState) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + + pState->backendState = *pBEState; +} + +void SwrSetPixelShaderState( + HANDLE hContext, + SWR_PS_STATE *pPSState) +{ + API_STATE *pState = GetDrawState(GetContext(hContext)); + pState->psState = *pPSState; +} + +void SwrSetBlendState( + HANDLE hContext, + SWR_BLEND_STATE *pBlendState) +{ + API_STATE *pState = GetDrawState(GetContext(hContext)); + memcpy(&pState->blendState, pBlendState, sizeof(SWR_BLEND_STATE)); +} + +void SwrSetBlendFunc( + HANDLE hContext, + uint32_t renderTarget, + PFN_BLEND_JIT_FUNC pfnBlendFunc) +{ + SWR_ASSERT(renderTarget < SWR_NUM_RENDERTARGETS); + API_STATE *pState = GetDrawState(GetContext(hContext)); + pState->pfnBlendFunc[renderTarget] = pfnBlendFunc; +} + +void SwrSetLinkage( + HANDLE hContext, + uint32_t mask, + const uint8_t* pMap) +{ + API_STATE* pState = GetDrawState(GetContext(hContext)); + + static const uint8_t IDENTITY_MAP[] = + { + 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, + 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, + }; + static_assert(sizeof(IDENTITY_MAP) == sizeof(pState->linkageMap), + "Update for new value of MAX_ATTRIBUTES"); + + pState->linkageMask = mask; + pState->linkageCount = _mm_popcnt_u32(mask); + + if (!pMap) + { + pMap = IDENTITY_MAP; + } + memcpy(pState->linkageMap, pMap, pState->linkageCount); +} + +// update guardband multipliers for the viewport +void updateGuardband(API_STATE *pState) +{ + // guardband center is viewport center + pState->gbState.left = KNOB_GUARDBAND_WIDTH / pState->vp[0].width; + pState->gbState.right = KNOB_GUARDBAND_WIDTH / pState->vp[0].width; + pState->gbState.top = KNOB_GUARDBAND_HEIGHT / pState->vp[0].height; + pState->gbState.bottom = KNOB_GUARDBAND_HEIGHT / pState->vp[0].height; +} + +void SwrSetRastState( + HANDLE hContext, + const SWR_RASTSTATE *pRastState) +{ + SWR_CONTEXT *pContext = GetContext(hContext); + API_STATE* pState = GetDrawState(pContext); + + memcpy(&pState->rastState, pRastState, sizeof(SWR_RASTSTATE)); +} + +void SwrSetViewports( + HANDLE hContext, + uint32_t numViewports, + const SWR_VIEWPORT* pViewports, + const SWR_VIEWPORT_MATRIX* pMatrices) +{ + SWR_ASSERT(numViewports <= KNOB_NUM_VIEWPORTS_SCISSORS, + "Invalid number of viewports."); + + SWR_CONTEXT *pContext = GetContext(hContext); + API_STATE* pState = GetDrawState(pContext); + + memcpy(&pState->vp[0], pViewports, sizeof(SWR_VIEWPORT) * numViewports); + + if (pMatrices != nullptr) + { + memcpy(&pState->vpMatrix[0], pMatrices, sizeof(SWR_VIEWPORT_MATRIX) * numViewports); + } + else + { + // Compute default viewport transform. + for (uint32_t i = 0; i < numViewports; ++i) + { + if (pContext->driverType == DX) + { + pState->vpMatrix[i].m00 = pState->vp[i].width / 2.0f; + pState->vpMatrix[i].m11 = -pState->vp[i].height / 2.0f; + pState->vpMatrix[i].m22 = pState->vp[i].maxZ - pState->vp[i].minZ; + pState->vpMatrix[i].m30 = pState->vp[i].x + pState->vpMatrix[i].m00; + pState->vpMatrix[i].m31 = pState->vp[i].y - pState->vpMatrix[i].m11; + pState->vpMatrix[i].m32 = pState->vp[i].minZ; + } + else + { + // Standard, with the exception that Y is inverted. + pState->vpMatrix[i].m00 = (pState->vp[i].width - pState->vp[i].x) / 2.0f; + pState->vpMatrix[i].m11 = (pState->vp[i].y - pState->vp[i].height) / 2.0f; + pState->vpMatrix[i].m22 = (pState->vp[i].maxZ - pState->vp[i].minZ) / 2.0f; + pState->vpMatrix[i].m30 = pState->vp[i].x + pState->vpMatrix[i].m00; + pState->vpMatrix[i].m31 = pState->vp[i].height + pState->vpMatrix[i].m11; + pState->vpMatrix[i].m32 = pState->vp[i].minZ + pState->vpMatrix[i].m22; + + // Now that the matrix is calculated, clip the view coords to screen size. + // OpenGL allows for -ve x,y in the viewport. + pState->vp[i].x = std::max(pState->vp[i].x, 0.0f); + pState->vp[i].y = std::max(pState->vp[i].y, 0.0f); + } + } + } + + updateGuardband(pState); +} + +void SwrSetScissorRects( + HANDLE hContext, + uint32_t numScissors, + const BBOX* pScissors) +{ + SWR_ASSERT(numScissors <= KNOB_NUM_VIEWPORTS_SCISSORS, + "Invalid number of scissor rects."); + + API_STATE* pState = GetDrawState(GetContext(hContext)); + memcpy(&pState->scissorRects[0], pScissors, numScissors * sizeof(BBOX)); +}; + +void SetupMacroTileScissors(DRAW_CONTEXT *pDC) +{ + API_STATE *pState = &pDC->pState->state; + uint32_t left, right, top, bottom; + + // Set up scissor dimensions based on scissor or viewport + if (pState->rastState.scissorEnable) + { + // scissor rect right/bottom edge are exclusive, core expects scissor dimensions to be inclusive, so subtract one pixel from right/bottom edges + left = pState->scissorRects[0].left; + right = pState->scissorRects[0].right; + top = pState->scissorRects[0].top; + bottom = pState->scissorRects[0].bottom; + } + else + { + left = (int32_t)pState->vp[0].x; + right = (int32_t)pState->vp[0].x + (int32_t)pState->vp[0].width; + top = (int32_t)pState->vp[0].y; + bottom = (int32_t)pState->vp[0].y + (int32_t)pState->vp[0].height; + } + + right = std::min(right, KNOB_MAX_SCISSOR_X); + bottom = std::min(bottom, KNOB_MAX_SCISSOR_Y); + + if (left > KNOB_MAX_SCISSOR_X || top > KNOB_MAX_SCISSOR_Y) + { + pState->scissorInFixedPoint.left = 0; + pState->scissorInFixedPoint.right = 0; + pState->scissorInFixedPoint.top = 0; + pState->scissorInFixedPoint.bottom = 0; + } + else + { + pState->scissorInFixedPoint.left = left * FIXED_POINT_SCALE; + pState->scissorInFixedPoint.right = right * FIXED_POINT_SCALE - 1; + pState->scissorInFixedPoint.top = top * FIXED_POINT_SCALE; + pState->scissorInFixedPoint.bottom = bottom * FIXED_POINT_SCALE - 1; + } +} + +void SetupPipeline(DRAW_CONTEXT *pDC) +{ + DRAW_STATE* pState = pDC->pState; + const SWR_RASTSTATE &rastState = pState->state.rastState; + BACKEND_FUNCS& backendFuncs = pState->backendFuncs; + const uint32_t forcedSampleCount = (rastState.bForcedSampleCount) ? 1 : 0; + + // setup backend + if (pState->state.psState.pfnPixelShader == nullptr) + { + backendFuncs.pfnBackend = gBackendNullPs[pState->state.rastState.sampleCount]; + // always need to generate I & J per sample for Z interpolation + backendFuncs.pfnCalcSampleBarycentrics = gSampleBarycentricTable[1]; + } + else + { + const bool bMultisampleEnable = ((rastState.sampleCount > SWR_MULTISAMPLE_1X) || rastState.bForcedSampleCount) ? 1 : 0; + const uint32_t centroid = ((pState->state.psState.barycentricsMask & SWR_BARYCENTRIC_CENTROID_MASK) > 0) ? 1 : 0; + + // currently only support 'normal' input coverage + SWR_ASSERT(pState->state.psState.inputCoverage == SWR_INPUT_COVERAGE_NORMAL || + pState->state.psState.inputCoverage == SWR_INPUT_COVERAGE_NONE); + + SWR_BARYCENTRICS_MASK barycentricsMask = (SWR_BARYCENTRICS_MASK)pState->state.psState.barycentricsMask; + + // select backend function + switch(pState->state.psState.shadingRate) + { + case SWR_SHADING_RATE_PIXEL: + if(bMultisampleEnable) + { + // always need to generate I & J per sample for Z interpolation + barycentricsMask = (SWR_BARYCENTRICS_MASK)(barycentricsMask | SWR_BARYCENTRIC_PER_SAMPLE_MASK); + backendFuncs.pfnBackend = gBackendPixelRateTable[rastState.sampleCount][rastState.samplePattern][pState->state.psState.inputCoverage][centroid][forcedSampleCount]; + backendFuncs.pfnOutputMerger = gBackendOutputMergerTable[pState->state.psState.numRenderTargets][pState->state.blendState.sampleCount]; + } + else + { + // always need to generate I & J per pixel for Z interpolation + barycentricsMask = (SWR_BARYCENTRICS_MASK)(barycentricsMask | SWR_BARYCENTRIC_PER_PIXEL_MASK); + backendFuncs.pfnBackend = gBackendSingleSample[pState->state.psState.inputCoverage][centroid]; + backendFuncs.pfnOutputMerger = gBackendOutputMergerTable[pState->state.psState.numRenderTargets][SWR_MULTISAMPLE_1X]; + } + break; + case SWR_SHADING_RATE_SAMPLE: + SWR_ASSERT(rastState.samplePattern == SWR_MSAA_STANDARD_PATTERN); + // always need to generate I & J per sample for Z interpolation + barycentricsMask = (SWR_BARYCENTRICS_MASK)(barycentricsMask | SWR_BARYCENTRIC_PER_SAMPLE_MASK); + backendFuncs.pfnBackend = gBackendSampleRateTable[rastState.sampleCount][pState->state.psState.inputCoverage][centroid]; + backendFuncs.pfnOutputMerger = gBackendOutputMergerTable[pState->state.psState.numRenderTargets][pState->state.blendState.sampleCount]; + break; + case SWR_SHADING_RATE_COARSE: + default: + SWR_ASSERT(0 && "Invalid shading rate"); + break; + } + + // setup pointer to function that generates necessary barycentrics required by the PS + bool bBarycentrics = (barycentricsMask & SWR_BARYCENTRIC_PER_PIXEL_MASK) > 0 ? 1 : 0; + backendFuncs.pfnCalcPixelBarycentrics = gPixelBarycentricTable[bBarycentrics]; + + bBarycentrics = (barycentricsMask & SWR_BARYCENTRIC_PER_SAMPLE_MASK) > 0 ? 1 : 0; + backendFuncs.pfnCalcSampleBarycentrics = gSampleBarycentricTable[bBarycentrics]; + + bBarycentrics = (barycentricsMask & SWR_BARYCENTRIC_CENTROID_MASK) > 0 ? 1 : 0; + backendFuncs.pfnCalcCentroidBarycentrics = gCentroidBarycentricTable[rastState.sampleCount][bBarycentrics][rastState.samplePattern][forcedSampleCount]; + } + + PFN_PROCESS_PRIMS pfnBinner; + switch (pState->state.topology) + { + case TOP_POINT_LIST: + pState->pfnProcessPrims = ClipPoints; + pfnBinner = BinPoints; + break; + case TOP_LINE_LIST: + case TOP_LINE_STRIP: + case TOP_LINE_LOOP: + case TOP_LINE_LIST_ADJ: + case TOP_LISTSTRIP_ADJ: + pState->pfnProcessPrims = ClipLines; + pfnBinner = BinLines; + break; + default: + pState->pfnProcessPrims = ClipTriangles; + pfnBinner = BinTriangles; + break; + }; + + // disable clipper if viewport transform is disabled + if (pState->state.frontendState.vpTransformDisable) + { + pState->pfnProcessPrims = pfnBinner; + } + + if ((pState->state.psState.pfnPixelShader == nullptr) && + (pState->state.depthStencilState.depthTestEnable == FALSE) && + (pState->state.depthStencilState.depthWriteEnable == FALSE) && + (pState->state.depthStencilState.stencilTestEnable == FALSE) && + (pState->state.depthStencilState.stencilWriteEnable == FALSE) && + (pState->state.linkageCount == 0)) + { + pState->pfnProcessPrims = nullptr; + pState->state.linkageMask = 0; + } + + if (pState->state.soState.rasterizerDisable == true) + { + pState->pfnProcessPrims = nullptr; + pState->state.linkageMask = 0; + } + + // set up the frontend attrib mask + pState->state.feAttribMask = pState->state.linkageMask; + if (pState->state.soState.soEnable) + { + for (uint32_t i = 0; i < 4; ++i) + { + pState->state.feAttribMask |= pState->state.soState.streamMasks[i]; + } + } + + // complicated logic to test for cases where we don't need backing hottile memory for a draw + // have to check for the special case where depth/stencil test is enabled but depthwrite is disabled. + pState->state.depthHottileEnable = ((!(pState->state.depthStencilState.depthTestEnable && + !pState->state.depthStencilState.depthWriteEnable && + pState->state.depthStencilState.depthTestFunc == ZFUNC_ALWAYS)) && + (pState->state.depthStencilState.depthTestEnable || + pState->state.depthStencilState.depthWriteEnable)) ? true : false; + + pState->state.stencilHottileEnable = (((!(pState->state.depthStencilState.stencilTestEnable && + !pState->state.depthStencilState.stencilWriteEnable && + pState->state.depthStencilState.stencilTestFunc == ZFUNC_ALWAYS)) || + // for stencil we have to check the double sided state as well + (!(pState->state.depthStencilState.doubleSidedStencilTestEnable && + !pState->state.depthStencilState.stencilWriteEnable && + pState->state.depthStencilState.backfaceStencilTestFunc == ZFUNC_ALWAYS))) && + (pState->state.depthStencilState.stencilTestEnable || + pState->state.depthStencilState.stencilWriteEnable)) ? true : false; + + uint32_t numRTs = pState->state.psState.numRenderTargets; + pState->state.colorHottileEnable = 0; + if(pState->state.psState.pfnPixelShader != nullptr) + { + for (uint32_t rt = 0; rt < numRTs; ++rt) + { + pState->state.colorHottileEnable |= + (!pState->state.blendState.renderTarget[rt].writeDisableAlpha || + !pState->state.blendState.renderTarget[rt].writeDisableRed || + !pState->state.blendState.renderTarget[rt].writeDisableGreen || + !pState->state.blendState.renderTarget[rt].writeDisableBlue) ? (1 << rt) : 0; + } + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief InitDraw +/// @param pDC - Draw context to initialize for this draw. +void InitDraw( + DRAW_CONTEXT *pDC, + bool isSplitDraw) +{ + // We don't need to re-setup the scissors/pipeline state again for split draw. + if (isSplitDraw == false) + { + SetupMacroTileScissors(pDC); + SetupPipeline(pDC); + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief We can split the draw for certain topologies for better performance. +/// @param totalVerts - Total vertices for draw +/// @param topology - Topology used for draw +uint32_t MaxVertsPerDraw( + DRAW_CONTEXT* pDC, + uint32_t totalVerts, + PRIMITIVE_TOPOLOGY topology) +{ + API_STATE& state = pDC->pState->state; + + uint32_t vertsPerDraw = totalVerts; + + if (state.soState.soEnable) + { + return totalVerts; + } + + switch (topology) + { + case TOP_POINT_LIST: + case TOP_TRIANGLE_LIST: + vertsPerDraw = KNOB_MAX_PRIMS_PER_DRAW; + break; + + case TOP_PATCHLIST_1: + case TOP_PATCHLIST_2: + case TOP_PATCHLIST_3: + case TOP_PATCHLIST_4: + case TOP_PATCHLIST_5: + case TOP_PATCHLIST_6: + case TOP_PATCHLIST_7: + case TOP_PATCHLIST_8: + case TOP_PATCHLIST_9: + case TOP_PATCHLIST_10: + case TOP_PATCHLIST_11: + case TOP_PATCHLIST_12: + case TOP_PATCHLIST_13: + case TOP_PATCHLIST_14: + case TOP_PATCHLIST_15: + case TOP_PATCHLIST_16: + case TOP_PATCHLIST_17: + case TOP_PATCHLIST_18: + case TOP_PATCHLIST_19: + case TOP_PATCHLIST_20: + case TOP_PATCHLIST_21: + case TOP_PATCHLIST_22: + case TOP_PATCHLIST_23: + case TOP_PATCHLIST_24: + case TOP_PATCHLIST_25: + case TOP_PATCHLIST_26: + case TOP_PATCHLIST_27: + case TOP_PATCHLIST_28: + case TOP_PATCHLIST_29: + case TOP_PATCHLIST_30: + case TOP_PATCHLIST_31: + case TOP_PATCHLIST_32: + if (pDC->pState->state.tsState.tsEnable) + { + uint32_t vertsPerPrim = topology - TOP_PATCHLIST_BASE; + vertsPerDraw = vertsPerPrim * KNOB_MAX_TESS_PRIMS_PER_DRAW; + } + break; + + default: + // We are not splitting up draws for other topologies. + break; + } + + return vertsPerDraw; +} + +// Recursive template used to auto-nest conditionals. Converts dynamic boolean function +// arguments to static template arguments. +template +struct FEDrawChooser +{ + // Last Arg Terminator + static PFN_FE_WORK_FUNC GetFunc(bool bArg) + { + if (bArg) + { + return ProcessDraw; + } + + return ProcessDraw; + } + + // Recursively parse args + template + static PFN_FE_WORK_FUNC GetFunc(bool bArg, TArgsT... remainingArgs) + { + if (bArg) + { + return FEDrawChooser::GetFunc(remainingArgs...); + } + + return FEDrawChooser::GetFunc(remainingArgs...); + } +}; + +// Selector for correct templated Draw front-end function +INLINE +static PFN_FE_WORK_FUNC GetFEDrawFunc(bool IsIndexed, bool HasTessellation, bool HasGeometryShader, bool HasStreamOut, bool RasterizerEnabled) +{ + return FEDrawChooser<>::GetFunc(IsIndexed, HasTessellation, HasGeometryShader, HasStreamOut, RasterizerEnabled); +} + + +////////////////////////////////////////////////////////////////////////// +/// @brief DrawInstanced +/// @param hContext - Handle passed back from SwrCreateContext +/// @param topology - Specifies topology for draw. +/// @param numVerts - How many vertices to read sequentially from vertex data (per instance). +/// @param startVertex - Specifies start vertex for draw. (vertex data) +/// @param numInstances - How many instances to render. +/// @param startInstance - Which instance to start sequentially fetching from in each buffer (instanced data) +void DrawInstanced( + HANDLE hContext, + PRIMITIVE_TOPOLOGY topology, + uint32_t numVertices, + uint32_t startVertex, + uint32_t numInstances = 1, + uint32_t startInstance = 0) +{ + if (KNOB_TOSS_DRAW) + { + return; + } + + RDTSC_START(APIDraw); + + SWR_CONTEXT *pContext = GetContext(hContext); + DRAW_CONTEXT* pDC = GetDrawContext(pContext); + + int32_t maxVertsPerDraw = MaxVertsPerDraw(pDC, numVertices, topology); + uint32_t primsPerDraw = GetNumPrims(topology, maxVertsPerDraw); + int32_t remainingVerts = numVertices; + + API_STATE *pState = &pDC->pState->state; + pState->topology = topology; + pState->forceFront = false; + + // disable culling for points/lines + uint32_t oldCullMode = pState->rastState.cullMode; + if (topology == TOP_POINT_LIST) + { + pState->rastState.cullMode = SWR_CULLMODE_NONE; + pState->forceFront = true; + } + + int draw = 0; + while (remainingVerts) + { + uint32_t numVertsForDraw = (remainingVerts < maxVertsPerDraw) ? + remainingVerts : maxVertsPerDraw; + + bool isSplitDraw = (draw > 0) ? true : false; + DRAW_CONTEXT* pDC = GetDrawContext(pContext, isSplitDraw); + InitDraw(pDC, isSplitDraw); + + pDC->FeWork.type = DRAW; + pDC->FeWork.pfnWork = GetFEDrawFunc( + false, // IsIndexed + pState->tsState.tsEnable, + pState->gsState.gsEnable, + pState->soState.soEnable, + pDC->pState->pfnProcessPrims != nullptr); + pDC->FeWork.desc.draw.numVerts = numVertsForDraw; + pDC->FeWork.desc.draw.startVertex = startVertex; + pDC->FeWork.desc.draw.numInstances = numInstances; + pDC->FeWork.desc.draw.startInstance = startInstance; + pDC->FeWork.desc.draw.startPrimID = draw * primsPerDraw; + pDC->FeWork.desc.draw.startVertexID = draw * maxVertsPerDraw; + + //enqueue DC + QueueDraw(pContext); + + remainingVerts -= numVertsForDraw; + draw++; + } + + // restore culling state + pDC = GetDrawContext(pContext); + pDC->pState->state.rastState.cullMode = oldCullMode; + + RDTSC_STOP(APIDraw, numVertices * numInstances, 0); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief SwrDraw +/// @param hContext - Handle passed back from SwrCreateContext +/// @param topology - Specifies topology for draw. +/// @param startVertex - Specifies start vertex in vertex buffer for draw. +/// @param primCount - Number of vertices. +void SwrDraw( + HANDLE hContext, + PRIMITIVE_TOPOLOGY topology, + uint32_t startVertex, + uint32_t numVertices) +{ + DrawInstanced(hContext, topology, numVertices, startVertex); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief SwrDrawInstanced +/// @param hContext - Handle passed back from SwrCreateContext +/// @param topology - Specifies topology for draw. +/// @param numVertsPerInstance - How many vertices to read sequentially from vertex data. +/// @param numInstances - How many instances to render. +/// @param startVertex - Specifies start vertex for draw. (vertex data) +/// @param startInstance - Which instance to start sequentially fetching from in each buffer (instanced data) +void SwrDrawInstanced( + HANDLE hContext, + PRIMITIVE_TOPOLOGY topology, + uint32_t numVertsPerInstance, + uint32_t numInstances, + uint32_t startVertex, + uint32_t startInstance + ) +{ + DrawInstanced(hContext, topology, numVertsPerInstance, startVertex, numInstances, startInstance); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief DrawIndexedInstanced +/// @param hContext - Handle passed back from SwrCreateContext +/// @param topology - Specifies topology for draw. +/// @param numIndices - Number of indices to read sequentially from index buffer. +/// @param indexOffset - Starting index into index buffer. +/// @param baseVertex - Vertex in vertex buffer to consider as index "0". Note value is signed. +/// @param numInstances - Number of instances to render. +/// @param startInstance - Which instance to start sequentially fetching from in each buffer (instanced data) +void DrawIndexedInstance( + HANDLE hContext, + PRIMITIVE_TOPOLOGY topology, + uint32_t numIndices, + uint32_t indexOffset, + int32_t baseVertex, + uint32_t numInstances = 1, + uint32_t startInstance = 0) +{ + if (KNOB_TOSS_DRAW) + { + return; + } + + RDTSC_START(APIDrawIndexed); + + SWR_CONTEXT *pContext = GetContext(hContext); + DRAW_CONTEXT* pDC = GetDrawContext(pContext); + API_STATE* pState = &pDC->pState->state; + + int32_t maxIndicesPerDraw = MaxVertsPerDraw(pDC, numIndices, topology); + uint32_t primsPerDraw = GetNumPrims(topology, maxIndicesPerDraw); + int32_t remainingIndices = numIndices; + + uint32_t indexSize = 0; + switch (pState->indexBuffer.format) + { + case R32_UINT: indexSize = sizeof(uint32_t); break; + case R16_UINT: indexSize = sizeof(uint16_t); break; + case R8_UINT: indexSize = sizeof(uint8_t); break; + default: + SWR_ASSERT(0); + } + + int draw = 0; + uint8_t *pIB = (uint8_t*)pState->indexBuffer.pIndices; + pIB += (uint64_t)indexOffset * (uint64_t)indexSize; + + pState->topology = topology; + pState->forceFront = false; + + // disable culling for points/lines + uint32_t oldCullMode = pState->rastState.cullMode; + if (topology == TOP_POINT_LIST) + { + pState->rastState.cullMode = SWR_CULLMODE_NONE; + pState->forceFront = true; + } + + while (remainingIndices) + { + uint32_t numIndicesForDraw = (remainingIndices < maxIndicesPerDraw) ? + remainingIndices : maxIndicesPerDraw; + + // When breaking up draw, we need to obtain new draw context for each iteration. + bool isSplitDraw = (draw > 0) ? true : false; + pDC = GetDrawContext(pContext, isSplitDraw); + InitDraw(pDC, isSplitDraw); + + pDC->FeWork.type = DRAW; + pDC->FeWork.pfnWork = GetFEDrawFunc( + true, // IsIndexed + pState->tsState.tsEnable, + pState->gsState.gsEnable, + pState->soState.soEnable, + pDC->pState->pfnProcessPrims != nullptr); + pDC->FeWork.desc.draw.pDC = pDC; + pDC->FeWork.desc.draw.numIndices = numIndicesForDraw; + pDC->FeWork.desc.draw.pIB = (int*)pIB; + pDC->FeWork.desc.draw.type = pDC->pState->state.indexBuffer.format; + + pDC->FeWork.desc.draw.numInstances = numInstances; + pDC->FeWork.desc.draw.startInstance = startInstance; + pDC->FeWork.desc.draw.baseVertex = baseVertex; + pDC->FeWork.desc.draw.startPrimID = draw * primsPerDraw; + + //enqueue DC + QueueDraw(pContext); + + pIB += maxIndicesPerDraw * indexSize; + remainingIndices -= numIndicesForDraw; + draw++; + } + + // restore culling state + pDC = GetDrawContext(pContext); + pDC->pState->state.rastState.cullMode = oldCullMode; + + RDTSC_STOP(APIDrawIndexed, numIndices * numInstances, 0); +} + + +////////////////////////////////////////////////////////////////////////// +/// @brief DrawIndexed +/// @param hContext - Handle passed back from SwrCreateContext +/// @param topology - Specifies topology for draw. +/// @param numIndices - Number of indices to read sequentially from index buffer. +/// @param indexOffset - Starting index into index buffer. +/// @param baseVertex - Vertex in vertex buffer to consider as index "0". Note value is signed. +void SwrDrawIndexed( + HANDLE hContext, + PRIMITIVE_TOPOLOGY topology, + uint32_t numIndices, + uint32_t indexOffset, + int32_t baseVertex + ) +{ + DrawIndexedInstance(hContext, topology, numIndices, indexOffset, baseVertex); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief SwrDrawIndexedInstanced +/// @param hContext - Handle passed back from SwrCreateContext +/// @param topology - Specifies topology for draw. +/// @param numIndices - Number of indices to read sequentially from index buffer. +/// @param numInstances - Number of instances to render. +/// @param indexOffset - Starting index into index buffer. +/// @param baseVertex - Vertex in vertex buffer to consider as index "0". Note value is signed. +/// @param startInstance - Which instance to start sequentially fetching from in each buffer (instanced data) +void SwrDrawIndexedInstanced( + HANDLE hContext, + PRIMITIVE_TOPOLOGY topology, + uint32_t numIndices, + uint32_t numInstances, + uint32_t indexOffset, + int32_t baseVertex, + uint32_t startInstance) +{ + DrawIndexedInstance(hContext, topology, numIndices, indexOffset, baseVertex, numInstances, startInstance); +} + +// Attach surfaces to pipeline +void SwrInvalidateTiles( + HANDLE hContext, + uint32_t attachmentMask) +{ + SWR_CONTEXT *pContext = (SWR_CONTEXT*)hContext; + DRAW_CONTEXT* pDC = GetDrawContext(pContext); + + // Queue a load to the hottile + pDC->FeWork.type = INVALIDATETILES; + pDC->FeWork.pfnWork = ProcessInvalidateTiles; + pDC->FeWork.desc.invalidateTiles.attachmentMask = attachmentMask; + + //enqueue + QueueDraw(pContext); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief SwrDispatch +/// @param hContext - Handle passed back from SwrCreateContext +/// @param threadGroupCountX - Number of thread groups dispatched in X direction +/// @param threadGroupCountY - Number of thread groups dispatched in Y direction +/// @param threadGroupCountZ - Number of thread groups dispatched in Z direction +void SwrDispatch( + HANDLE hContext, + uint32_t threadGroupCountX, + uint32_t threadGroupCountY, + uint32_t threadGroupCountZ) +{ + if (KNOB_TOSS_DRAW) + { + return; + } + + RDTSC_START(APIDispatch); + SWR_CONTEXT *pContext = (SWR_CONTEXT*)hContext; + DRAW_CONTEXT* pDC = GetDrawContext(pContext); + + pDC->isCompute = true; // This is a compute context. + + // Ensure spill fill pointers are initialized to nullptr. + memset(pDC->pSpillFill, 0, sizeof(pDC->pSpillFill)); + + COMPUTE_DESC* pTaskData = (COMPUTE_DESC*)pDC->pArena->AllocAligned(sizeof(COMPUTE_DESC), 64); + + pTaskData->threadGroupCountX = threadGroupCountX; + pTaskData->threadGroupCountY = threadGroupCountY; + pTaskData->threadGroupCountZ = threadGroupCountZ; + + uint32_t totalThreadGroups = threadGroupCountX * threadGroupCountY * threadGroupCountZ; + pDC->pDispatch->initialize(totalThreadGroups, pTaskData); + + QueueDispatch(pContext); + RDTSC_STOP(APIDispatch, threadGroupCountX * threadGroupCountY * threadGroupCountZ, 0); +} + +// Deswizzles, converts and stores current contents of the hot tiles to surface +// described by pState +void SwrStoreTiles( + HANDLE hContext, + SWR_RENDERTARGET_ATTACHMENT attachment, + SWR_TILE_STATE postStoreTileState) +{ + RDTSC_START(APIStoreTiles); + + SWR_CONTEXT *pContext = (SWR_CONTEXT*)hContext; + DRAW_CONTEXT* pDC = GetDrawContext(pContext); + + SetupMacroTileScissors(pDC); + + pDC->FeWork.type = STORETILES; + pDC->FeWork.pfnWork = ProcessStoreTiles; + pDC->FeWork.desc.storeTiles.attachment = attachment; + pDC->FeWork.desc.storeTiles.postStoreTileState = postStoreTileState; + + //enqueue + QueueDraw(pContext); + + RDTSC_STOP(APIStoreTiles, 0, 0); +} + +void SwrClearRenderTarget( + HANDLE hContext, + uint32_t clearMask, + const float clearColor[4], + float z, + BYTE stencil) +{ + RDTSC_START(APIClearRenderTarget); + + SWR_CONTEXT *pContext = (SWR_CONTEXT*)hContext; + + DRAW_CONTEXT* pDC = GetDrawContext(pContext); + + SetupMacroTileScissors(pDC); + + CLEAR_FLAGS flags; + flags.mask = clearMask; + + pDC->FeWork.type = CLEAR; + pDC->FeWork.pfnWork = ProcessClear; + pDC->FeWork.desc.clear.flags = flags; + pDC->FeWork.desc.clear.clearDepth = z; + pDC->FeWork.desc.clear.clearRTColor[0] = clearColor[0]; + pDC->FeWork.desc.clear.clearRTColor[1] = clearColor[1]; + pDC->FeWork.desc.clear.clearRTColor[2] = clearColor[2]; + pDC->FeWork.desc.clear.clearRTColor[3] = clearColor[3]; + pDC->FeWork.desc.clear.clearStencil = stencil; + + // enqueue draw + QueueDraw(pContext); + + RDTSC_STOP(APIClearRenderTarget, 0, pDC->drawId); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Returns a pointer to the private context state for the current +/// draw operation. This is used for external componets such as the +/// sampler. +/// SWR is responsible for the allocation of the private context state. +/// @param hContext - Handle passed back from SwrCreateContext +VOID* SwrGetPrivateContextState( + HANDLE hContext) +{ + SWR_CONTEXT* pContext = GetContext(hContext); + DRAW_CONTEXT* pDC = GetDrawContext(pContext); + DRAW_STATE* pState = pDC->pState; + + if (pState->pPrivateState == nullptr) + { + pState->pPrivateState = pState->pArena->AllocAligned(pContext->privateStateSize, KNOB_SIMD_WIDTH*sizeof(float)); + } + + return pState->pPrivateState; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Clients can use this to allocate memory for draw/dispatch +/// operations. The memory will automatically be freed once operation +/// has completed. Client can use this to allocate binding tables, +/// etc. needed for shader execution. +/// @param hContext - Handle passed back from SwrCreateContext +/// @param size - Size of allocation +/// @param align - Alignment needed for allocation. +VOID* SwrAllocDrawContextMemory( + HANDLE hContext, + uint32_t size, + uint32_t align) +{ + SWR_CONTEXT* pContext = GetContext(hContext); + DRAW_CONTEXT* pDC = GetDrawContext(pContext); + + return pDC->pState->pArena->AllocAligned(size, align); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Returns pointer to SWR stats. +/// @note The counters are atomically incremented by multiple threads. +/// When calling this, you need to ensure all previous operations +/// have completed. +/// @todo If necessary, add a callback to avoid stalling the pipe to +/// sample the counters. +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pStats - SWR will fill this out for caller. +void SwrGetStats( + HANDLE hContext, + SWR_STATS* pStats) +{ + SWR_CONTEXT *pContext = GetContext(hContext); + DRAW_CONTEXT* pDC = GetDrawContext(pContext); + + pDC->FeWork.type = QUERYSTATS; + pDC->FeWork.pfnWork = ProcessQueryStats; + pDC->FeWork.desc.queryStats.pStats = pStats; + + // cannot execute until all previous draws have completed + pDC->dependency = pDC->drawId - 1; + + //enqueue + QueueDraw(pContext); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Enables stats counting +/// @param hContext - Handle passed back from SwrCreateContext +/// @param enable - If true then counts are incremented. +void SwrEnableStats( + HANDLE hContext, + bool enable) +{ + SWR_CONTEXT *pContext = GetContext(hContext); + DRAW_CONTEXT* pDC = GetDrawContext(pContext); + + pDC->pState->state.enableStats = enable; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Mark end of frame - used for performance profiling +/// @param hContext - Handle passed back from SwrCreateContext +void SWR_API SwrEndFrame( + HANDLE hContext) +{ + RDTSC_ENDFRAME(); +} diff --git a/src/gallium/drivers/swr/rasterizer/core/api.h b/src/gallium/drivers/swr/rasterizer/core/api.h new file mode 100644 index 00000000000..72fae8b2c21 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/api.h @@ -0,0 +1,500 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file api.h +* +* @brief API definitions +* +******************************************************************************/ + +#ifndef __SWR_API_H__ +#define __SWR_API_H__ + +#include "common/os.h" + +#include +#include + +#include "common/simdintrin.h" +#include "common/formats.h" +#include "core/utils.h" +#include "core/state.h" + +///@todo place all the API functions into the 'swr' namespace. + +typedef void(SWR_API *PFN_CALLBACK_FUNC)(uint64_t data, uint64_t data2, uint64_t data3); + +////////////////////////////////////////////////////////////////////////// +/// @brief Function signature for load hot tiles +/// @param hPrivateContext - handle to private data +/// @param dstFormat - format of the hot tile +/// @param renderTargetIndex - render target to store, can be color, depth or stencil +/// @param x - destination x coordinate +/// @param y - destination y coordinate +/// @param pDstHotTile - pointer to the hot tile surface +typedef void(SWR_API *PFN_LOAD_TILE)(HANDLE hPrivateContext, SWR_FORMAT dstFormat, + SWR_RENDERTARGET_ATTACHMENT renderTargetIndex, + uint32_t x, uint32_t y, uint32_t renderTargetArrayIndex, BYTE *pDstHotTile); + +////////////////////////////////////////////////////////////////////////// +/// @brief Function signature for store hot tiles +/// @param hPrivateContext - handle to private data +/// @param srcFormat - format of the hot tile +/// @param renderTargetIndex - render target to store, can be color, depth or stencil +/// @param x - destination x coordinate +/// @param y - destination y coordinate +/// @param pSrcHotTile - pointer to the hot tile surface +typedef void(SWR_API *PFN_STORE_TILE)(HANDLE hPrivateContext, SWR_FORMAT srcFormat, + SWR_RENDERTARGET_ATTACHMENT renderTargetIndex, + uint32_t x, uint32_t y, uint32_t renderTargetArrayIndex, BYTE *pSrcHotTile); + +/// @brief Function signature for clearing from the hot tiles clear value +/// @param hPrivateContext - handle to private data +/// @param renderTargetIndex - render target to store, can be color, depth or stencil +/// @param x - destination x coordinate +/// @param y - destination y coordinate +/// @param pClearColor - pointer to the hot tile's clear value +typedef void(SWR_API *PFN_CLEAR_TILE)(HANDLE hPrivateContext, + SWR_RENDERTARGET_ATTACHMENT rtIndex, + uint32_t x, uint32_t y, const float* pClearColor); + +////////////////////////////////////////////////////////////////////////// +/// SWR_CREATECONTEXT_INFO +///////////////////////////////////////////////////////////////////////// +struct SWR_CREATECONTEXT_INFO +{ + DRIVER_TYPE driver; + + // External functions (e.g. sampler) need per draw context state. + // Use SwrGetPrivateContextState() to access private state. + uint32_t privateStateSize; + + // Each SWR context can have multiple sets of active state + uint32_t maxSubContexts; + + // tile manipulation functions + PFN_LOAD_TILE pfnLoadTile; + PFN_STORE_TILE pfnStoreTile; + PFN_CLEAR_TILE pfnClearTile; +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_RECT +///////////////////////////////////////////////////////////////////////// +struct SWR_RECT +{ + uint32_t left; + uint32_t right; + uint32_t top; + uint32_t bottom; +}; + +////////////////////////////////////////////////////////////////////////// +/// @brief Create SWR Context. +/// @param pCreateInfo - pointer to creation info. +HANDLE SWR_API SwrCreateContext( + const SWR_CREATECONTEXT_INFO* pCreateInfo); + +////////////////////////////////////////////////////////////////////////// +/// @brief Destroys SWR Context. +/// @param hContext - Handle passed back from SwrCreateContext +void SWR_API SwrDestroyContext( + HANDLE hContext); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set currently active state context +/// @param subContextIndex - value from 0 to +/// SWR_CREATECONTEXT_INFO.maxSubContexts. Defaults to 0. +void SWR_API SwrSetActiveSubContext( + HANDLE hContext, + uint32_t subContextIndex); + +////////////////////////////////////////////////////////////////////////// +/// @brief Sync cmd. Executes the callback func when all rendering up to this sync +/// has been completed +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pfnFunc - pointer to callback function, +/// @param userData - user data to pass back +void SWR_API SwrSync( + HANDLE hContext, + PFN_CALLBACK_FUNC pfnFunc, + uint64_t userData, + uint64_t userData2, + uint64_t userData3 = 0); + +////////////////////////////////////////////////////////////////////////// +/// @brief Blocks until all rendering has been completed. +/// @param hContext - Handle passed back from SwrCreateContext +void SWR_API SwrWaitForIdle( + HANDLE hContext); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set vertex buffer state. +/// @param hContext - Handle passed back from SwrCreateContext +/// @param numBuffers - Number of vertex buffer state descriptors. +/// @param pVertexBuffers - Array of vertex buffer state descriptors. +void SWR_API SwrSetVertexBuffers( + HANDLE hContext, + uint32_t numBuffers, + const SWR_VERTEX_BUFFER_STATE* pVertexBuffers); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set index buffer +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pIndexBuffer - Index buffer. +void SWR_API SwrSetIndexBuffer( + HANDLE hContext, + const SWR_INDEX_BUFFER_STATE* pIndexBuffer); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set fetch shader pointer. +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pfnFetchFunc - Pointer to shader. +void SWR_API SwrSetFetchFunc( + HANDLE hContext, + PFN_FETCH_FUNC pfnFetchFunc); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set streamout shader pointer. +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pfnSoFunc - Pointer to shader. +/// @param streamIndex - specifies stream +void SWR_API SwrSetSoFunc( + HANDLE hContext, + PFN_SO_FUNC pfnSoFunc, + uint32_t streamIndex); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set streamout state +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pSoState - Pointer to streamout state. +void SWR_API SwrSetSoState( + HANDLE hContext, + SWR_STREAMOUT_STATE* pSoState); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set streamout buffer state +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pSoBuffer - Pointer to streamout buffer. +/// @param slot - Slot to bind SO buffer to. +void SWR_API SwrSetSoBuffers( + HANDLE hContext, + SWR_STREAMOUT_BUFFER* pSoBuffer, + uint32_t slot); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set vertex shader pointer. +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pfnVertexFunc - Pointer to shader. +void SWR_API SwrSetVertexFunc( + HANDLE hContext, + PFN_VERTEX_FUNC pfnVertexFunc); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set frontend state. +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pState - Pointer to state +void SWR_API SwrSetFrontendState( + HANDLE hContext, + SWR_FRONTEND_STATE *pState); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set geometry shader state. +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pState - Pointer to state +void SWR_API SwrSetGsState( + HANDLE hContext, + SWR_GS_STATE *pState); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set geometry shader +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pState - Pointer to geometry shader function +void SWR_API SwrSetGsFunc( + HANDLE hContext, + PFN_GS_FUNC pfnGsFunc); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set compute shader +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pState - Pointer to compute shader function +/// @param totalThreadsInGroup - product of thread group dimensions. +void SWR_API SwrSetCsFunc( + HANDLE hContext, + PFN_CS_FUNC pfnCsFunc, + uint32_t totalThreadsInGroup); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set tessellation state. +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pState - Pointer to state +void SWR_API SwrSetTsState( + HANDLE hContext, + SWR_TS_STATE *pState); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set hull shader +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pfnFunc - Pointer to shader function +void SWR_API SwrSetHsFunc( + HANDLE hContext, + PFN_HS_FUNC pfnFunc); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set domain shader +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pfnFunc - Pointer to shader function +void SWR_API SwrSetDsFunc( + HANDLE hContext, + PFN_DS_FUNC pfnFunc); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set depth stencil state +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pState - Pointer to state. +void SWR_API SwrSetDepthStencilState( + HANDLE hContext, + SWR_DEPTH_STENCIL_STATE *pState); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set backend state +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pState - Pointer to state. +void SWR_API SwrSetBackendState( + HANDLE hContext, + SWR_BACKEND_STATE *pState); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set pixel shader state +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pState - Pointer to state. +void SWR_API SwrSetPixelShaderState( + HANDLE hContext, + SWR_PS_STATE *pState); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set blend state +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pState - Pointer to state. +void SWR_API SwrSetBlendState( + HANDLE hContext, + SWR_BLEND_STATE *pState); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set blend function +/// @param hContext - Handle passed back from SwrCreateContext +/// @param renderTarget - render target index +/// @param pfnBlendFunc - function pointer +void SWR_API SwrSetBlendFunc( + HANDLE hContext, + uint32_t renderTarget, + PFN_BLEND_JIT_FUNC pfnBlendFunc); + +////////////////////////////////////////////////////////////////////////// +/// @brief Set linkage mask +/// @param hContext - Handle passed back from SwrCreateContext +/// @param mask - Specifies which vertex outputs are are needed by PS. +/// @param pMap - (Optional)Linkage map to specify where FE attributes are +/// gathered from to supply PS attribute values. The length +/// of the map buffer needs to match the number of set bits +/// in "mask". +void SWR_API SwrSetLinkage( + HANDLE hContext, + uint32_t mask, + const uint8_t* pMap); + +////////////////////////////////////////////////////////////////////////// +/// @brief SwrDraw +/// @param hContext - Handle passed back from SwrCreateContext +/// @param topology - Specifies topology for draw. +/// @param startVertex - Specifies start vertex in vertex buffer for draw. +/// @param primCount - Number of vertices. +void SWR_API SwrDraw( + HANDLE hContext, + PRIMITIVE_TOPOLOGY topology, + uint32_t startVertex, + uint32_t primCount); + +////////////////////////////////////////////////////////////////////////// +/// @brief SwrDrawInstanced +/// @param hContext - Handle passed back from SwrCreateContext +/// @param topology - Specifies topology for draw. +/// @param numVertsPerInstance - How many vertices to read sequentially from vertex data. +/// @param numInstances - How many instances to render. +/// @param startVertex - Specifies start vertex for draw. (vertex data) +/// @param startInstance - Which instance to start sequentially fetching from in each buffer (instanced data) +void SWR_API SwrDrawInstanced( + HANDLE hContext, + PRIMITIVE_TOPOLOGY topology, + uint32_t numVertsPerInstance, + uint32_t numInstances, + uint32_t startVertex, + uint32_t startInstance); + +////////////////////////////////////////////////////////////////////////// +/// @brief DrawIndexed +/// @param hContext - Handle passed back from SwrCreateContext +/// @param topology - Specifies topology for draw. +/// @param numIndices - Number of indices to read sequentially from index buffer. +/// @param indexOffset - Starting index into index buffer. +/// @param baseVertex - Vertex in vertex buffer to consider as index "0". Note value is signed. +void SWR_API SwrDrawIndexed( + HANDLE hContext, + PRIMITIVE_TOPOLOGY topology, + uint32_t numIndices, + uint32_t indexOffset, + int32_t baseVertex); + +////////////////////////////////////////////////////////////////////////// +/// @brief SwrDrawIndexedInstanced +/// @param hContext - Handle passed back from SwrCreateContext +/// @param topology - Specifies topology for draw. +/// @param numIndices - Number of indices to read sequentially from index buffer. +/// @param numInstances - Number of instances to render. +/// @param indexOffset - Starting index into index buffer. +/// @param baseVertex - Vertex in vertex buffer to consider as index "0". Note value is signed. +/// @param startInstance - Which instance to start sequentially fetching from in each buffer (instanced data) +void SWR_API SwrDrawIndexedInstanced( + HANDLE hContext, + PRIMITIVE_TOPOLOGY topology, + uint32_t numIndices, + uint32_t numInstances, + uint32_t indexOffset, + int32_t baseVertex, + uint32_t startInstance); + +////////////////////////////////////////////////////////////////////////// +/// @brief SwrInvalidateTiles +/// @param hContext - Handle passed back from SwrCreateContext +/// @param attachmentMask - The mask specifies which surfaces attached to the hottiles to invalidate. +void SWR_API SwrInvalidateTiles( + HANDLE hContext, + uint32_t attachmentMask); + +////////////////////////////////////////////////////////////////////////// +/// @brief SwrDispatch +/// @param hContext - Handle passed back from SwrCreateContext +/// @param threadGroupCountX - Number of thread groups dispatched in X direction +/// @param threadGroupCountY - Number of thread groups dispatched in Y direction +/// @param threadGroupCountZ - Number of thread groups dispatched in Z direction +void SWR_API SwrDispatch( + HANDLE hContext, + uint32_t threadGroupCountX, + uint32_t threadGroupCountY, + uint32_t threadGroupCountZ); + + +enum SWR_TILE_STATE +{ + SWR_TILE_INVALID = 0, // tile is in unitialized state and should be loaded with surface contents before rendering + SWR_TILE_DIRTY = 2, // tile contains newer data than surface it represents + SWR_TILE_RESOLVED = 3, // is in sync with surface it represents +}; + +/// @todo Add a good description for what attachments are and when and why you would use the different SWR_TILE_STATEs. +void SWR_API SwrStoreTiles( + HANDLE hContext, + SWR_RENDERTARGET_ATTACHMENT attachment, + SWR_TILE_STATE postStoreTileState); + +void SWR_API SwrClearRenderTarget( + HANDLE hContext, + uint32_t clearMask, + const FLOAT clearColor[4], + float z, + BYTE stencil); + +void SWR_API SwrSetRastState( + HANDLE hContext, + const SWR_RASTSTATE *pRastState); + +////////////////////////////////////////////////////////////////////////// +/// @brief SwrSetViewports +/// @param hContext - Handle passed back from SwrCreateContext +/// @param numViewports - number of viewports passed in +/// @param pViewports - Specifies extents of viewport. +/// @param pMatrices - If not specified then SWR computes a default one. +void SWR_API SwrSetViewports( + HANDLE hContext, + uint32_t numViewports, + const SWR_VIEWPORT* pViewports, + const SWR_VIEWPORT_MATRIX* pMatrices); + +////////////////////////////////////////////////////////////////////////// +/// @brief SwrSetScissorRects +/// @param hContext - Handle passed back from SwrCreateContext +/// @param numScissors - number of scissors passed in +/// @param pScissors - array of scissors +void SWR_API SwrSetScissorRects( + HANDLE hContext, + uint32_t numScissors, + const BBOX* pScissors); + +////////////////////////////////////////////////////////////////////////// +/// @brief Returns a pointer to the private context state for the current +/// draw operation. This is used for external componets such as the +/// sampler. +/// +/// @note Client needs to resend private state prior to each draw call. +/// Also, SWR is responsible for the private state memory. +/// @param hContext - Handle passed back from SwrCreateContext +VOID* SWR_API SwrGetPrivateContextState( + HANDLE hContext); + +////////////////////////////////////////////////////////////////////////// +/// @brief Clients can use this to allocate memory for draw/dispatch +/// operations. The memory will automatically be freed once operation +/// has completed. Client can use this to allocate binding tables, +/// etc. needed for shader execution. +/// @param hContext - Handle passed back from SwrCreateContext +/// @param size - Size of allocation +/// @param align - Alignment needed for allocation. +VOID* SWR_API SwrAllocDrawContextMemory( + HANDLE hContext, + uint32_t size, + uint32_t align); + +////////////////////////////////////////////////////////////////////////// +/// @brief Returns pointer to SWR stats. +/// @note The counters are incremented by multiple threads. +/// When calling this, you need to ensure all previous operations +/// have completed. +/// @param hContext - Handle passed back from SwrCreateContext +/// @param pStats - SWR will fill this out for caller. +void SWR_API SwrGetStats( + HANDLE hContext, + SWR_STATS* pStats); + +////////////////////////////////////////////////////////////////////////// +/// @brief Enables stats counting +/// @param hContext - Handle passed back from SwrCreateContext +/// @param enable - If true then counts are incremented. +void SWR_API SwrEnableStats( + HANDLE hContext, + bool enable); + +////////////////////////////////////////////////////////////////////////// +/// @brief Mark end of frame - used for performance profiling +/// @param hContext - Handle passed back from SwrCreateContext +void SWR_API SwrEndFrame( + HANDLE hContext); +#endif//__SWR_API_H__ diff --git a/src/gallium/drivers/swr/rasterizer/core/arena.cpp b/src/gallium/drivers/swr/rasterizer/core/arena.cpp new file mode 100644 index 00000000000..8184c8d3f4c --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/arena.cpp @@ -0,0 +1,166 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file arena.cpp +* +* @brief Arena memory manager +* The arena is convenient and fast for managing allocations for any of +* our allocations that are associated with operations and can all be freed +* once when their operation has completed. Allocations are cheap since +* most of the time its simply an increment of an offset. Also, no need to +* free individual allocations. All of the arena memory can be freed at once. +* +******************************************************************************/ + +#include "context.h" +#include "arena.h" + +#include + +Arena::Arena() + : m_pCurBlock(nullptr), m_size(0) +{ + m_pMutex = new std::mutex(); +} + +Arena::~Arena() +{ + Reset(); // Reset just in case to avoid leaking memory. + + if (m_pCurBlock) + { + _aligned_free(m_pCurBlock->pMem); + delete m_pCurBlock; + } + + delete m_pMutex; +} + +///@todo Remove this when all users have stopped using this. +void Arena::Init() +{ + m_size = 0; + m_pCurBlock = nullptr; + + m_pMutex = new std::mutex(); +} + +void* Arena::AllocAligned(size_t size, size_t align) +{ + if (m_pCurBlock) + { + ArenaBlock* pCurBlock = m_pCurBlock; + pCurBlock->offset = AlignUp(pCurBlock->offset, align); + + if ((pCurBlock->offset + size) <= pCurBlock->blockSize) + { + void* pMem = PtrAdd(pCurBlock->pMem, pCurBlock->offset); + pCurBlock->offset += size; + m_size += size; + return pMem; + } + + // Not enough memory in this block, fall through to allocate + // a new block + } + + static const size_t ArenaBlockSize = 1024*1024; + size_t blockSize = std::max(m_size + ArenaBlockSize, std::max(size, ArenaBlockSize)); + blockSize = AlignUp(blockSize, KNOB_SIMD_WIDTH*4); + + void *pMem = _aligned_malloc(blockSize, KNOB_SIMD_WIDTH*4); // Arena blocks are always simd byte aligned. + SWR_ASSERT(pMem != nullptr); + + ArenaBlock* pNewBlock = new (std::nothrow) ArenaBlock(); + SWR_ASSERT(pNewBlock != nullptr); + + if (pNewBlock != nullptr) + { + pNewBlock->pNext = m_pCurBlock; + + m_pCurBlock = pNewBlock; + m_pCurBlock->pMem = pMem; + m_pCurBlock->blockSize = blockSize; + + } + + return AllocAligned(size, align); +} + +void* Arena::Alloc(size_t size) +{ + return AllocAligned(size, 1); +} + +void* Arena::AllocAlignedSync(size_t size, size_t align) +{ + void* pAlloc = nullptr; + + SWR_ASSERT(m_pMutex != nullptr); + + m_pMutex->lock(); + pAlloc = AllocAligned(size, align); + m_pMutex->unlock(); + + return pAlloc; +} + +void* Arena::AllocSync(size_t size) +{ + void* pAlloc = nullptr; + + SWR_ASSERT(m_pMutex != nullptr); + + m_pMutex->lock(); + pAlloc = Alloc(size); + m_pMutex->unlock(); + + return pAlloc; +} + +void Arena::Reset(bool removeAll) +{ + if (m_pCurBlock) + { + m_pCurBlock->offset = 0; + + ArenaBlock *pUsedBlocks = m_pCurBlock->pNext; + m_pCurBlock->pNext = nullptr; + while(pUsedBlocks) + { + ArenaBlock* pBlock = pUsedBlocks; + pUsedBlocks = pBlock->pNext; + + _aligned_free(pBlock->pMem); + delete pBlock; + } + + if (removeAll) + { + _aligned_free(m_pCurBlock->pMem); + delete m_pCurBlock; + m_pCurBlock = nullptr; + } + } + + m_size = 0; +} diff --git a/src/gallium/drivers/swr/rasterizer/core/arena.h b/src/gallium/drivers/swr/rasterizer/core/arena.h new file mode 100644 index 00000000000..76eee11fb08 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/arena.h @@ -0,0 +1,69 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file arena.h +* +* @brief Arena memory manager +* The arena is convenient and fast for managing allocations for any of +* our allocations that are associated with operations and can all be freed +* once when their operation has completed. Allocations are cheap since +* most of the time its simply an increment of an offset. Also, no need to +* free individual allocations. All of the arena memory can be freed at once. +* +******************************************************************************/ +#pragma once + +#include + +class Arena +{ +public: + Arena(); + ~Arena(); + + void Init(); + + void* AllocAligned(size_t size, size_t align); + void* Alloc(size_t size); + + void* AllocAlignedSync(size_t size, size_t align); + void* AllocSync(size_t size); + + void Reset(bool removeAll = false); + size_t Size() { return m_size; } + +private: + + struct ArenaBlock + { + void* pMem = nullptr; + size_t blockSize = 0; + size_t offset = 0; + ArenaBlock* pNext = nullptr; + }; + + ArenaBlock* m_pCurBlock = nullptr; + size_t m_size = 0; + + /// @note Mutex is only used by sync allocation functions. + std::mutex* m_pMutex; +}; diff --git a/src/gallium/drivers/swr/rasterizer/core/backend.cpp b/src/gallium/drivers/swr/rasterizer/core/backend.cpp new file mode 100644 index 00000000000..4a472bc9e5c --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/backend.cpp @@ -0,0 +1,1899 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file backend.cpp +* +* @brief Backend handles rasterization, pixel shading and output merger +* operations. +* +******************************************************************************/ + +#include + +#include "rdtsc_core.h" +#include "backend.h" +#include "depthstencil.h" +#include "tilemgr.h" +#include "memory/tilingtraits.h" +#include "core/multisample.h" + +#include + +const __m128 vTileOffsetsX = {0.5, KNOB_TILE_X_DIM - 0.5, 0.5, KNOB_TILE_X_DIM - 0.5}; +const __m128 vTileOffsetsY = {0.5, 0.5, KNOB_TILE_Y_DIM - 0.5, KNOB_TILE_Y_DIM - 0.5}; + +/// @todo move to common lib +#define MASKTOVEC(i3,i2,i1,i0) {-i0,-i1,-i2,-i3} +static const __m128 gMaskToVec[] = { + MASKTOVEC(0,0,0,0), + MASKTOVEC(0,0,0,1), + MASKTOVEC(0,0,1,0), + MASKTOVEC(0,0,1,1), + MASKTOVEC(0,1,0,0), + MASKTOVEC(0,1,0,1), + MASKTOVEC(0,1,1,0), + MASKTOVEC(0,1,1,1), + MASKTOVEC(1,0,0,0), + MASKTOVEC(1,0,0,1), + MASKTOVEC(1,0,1,0), + MASKTOVEC(1,0,1,1), + MASKTOVEC(1,1,0,0), + MASKTOVEC(1,1,0,1), + MASKTOVEC(1,1,1,0), + MASKTOVEC(1,1,1,1), +}; + +typedef void(*PFN_CLEAR_TILES)(DRAW_CONTEXT*, SWR_RENDERTARGET_ATTACHMENT rt, uint32_t, DWORD[4]); +static PFN_CLEAR_TILES sClearTilesTable[NUM_SWR_FORMATS]; + +////////////////////////////////////////////////////////////////////////// +/// @brief Process compute work. +/// @param pDC - pointer to draw context (dispatch). +/// @param workerId - The unique worker ID that is assigned to this thread. +/// @param threadGroupId - the linear index for the thread group within the dispatch. +void ProcessComputeBE(DRAW_CONTEXT* pDC, uint32_t workerId, uint32_t threadGroupId) +{ + RDTSC_START(BEDispatch); + + SWR_CONTEXT *pContext = pDC->pContext; + + const COMPUTE_DESC* pTaskData = (COMPUTE_DESC*)pDC->pDispatch->GetTasksData(); + SWR_ASSERT(pTaskData != nullptr); + + // Ensure spill fill memory has been allocated. + if (pDC->pSpillFill[workerId] == nullptr) + { + ///@todo Add state which indicates the spill fill size. + pDC->pSpillFill[workerId] = (uint8_t*)pDC->pArena->AllocAlignedSync(4096 * 1024, sizeof(float) * 8); + } + + const API_STATE& state = GetApiState(pDC); + + SWR_CS_CONTEXT csContext{ 0 }; + csContext.tileCounter = threadGroupId; + csContext.dispatchDims[0] = pTaskData->threadGroupCountX; + csContext.dispatchDims[1] = pTaskData->threadGroupCountY; + csContext.dispatchDims[2] = pTaskData->threadGroupCountZ; + csContext.pTGSM = pContext->pScratch[workerId]; + csContext.pSpillFillBuffer = pDC->pSpillFill[workerId]; + + state.pfnCsFunc(GetPrivateState(pDC), &csContext); + + UPDATE_STAT(CsInvocations, state.totalThreadsInGroup); + + RDTSC_STOP(BEDispatch, 1, 0); +} + +void ProcessSyncBE(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pUserData) +{ + SYNC_DESC *pSync = (SYNC_DESC*)pUserData; + + uint32_t x, y; + MacroTileMgr::getTileIndices(macroTile, x, y); + SWR_ASSERT(x == 0 && y == 0); + + if (pSync->pfnCallbackFunc != nullptr) + { + pSync->pfnCallbackFunc(pSync->userData, pSync->userData2, pSync->userData3); + } +} + +void ProcessQueryStatsBE(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pUserData) +{ + QUERY_DESC* pQueryDesc = (QUERY_DESC*)pUserData; + SWR_STATS* pStats = pQueryDesc->pStats; + SWR_CONTEXT *pContext = pDC->pContext; + + SWR_ASSERT(pStats != nullptr); + + for (uint32_t i = 0; i < pContext->NumWorkerThreads; ++i) + { + pStats->DepthPassCount += pContext->stats[i].DepthPassCount; + + pStats->IaVertices += pContext->stats[i].IaVertices; + pStats->IaPrimitives += pContext->stats[i].IaPrimitives; + pStats->VsInvocations += pContext->stats[i].VsInvocations; + pStats->HsInvocations += pContext->stats[i].HsInvocations; + pStats->DsInvocations += pContext->stats[i].DsInvocations; + pStats->GsInvocations += pContext->stats[i].GsInvocations; + pStats->PsInvocations += pContext->stats[i].PsInvocations; + pStats->CInvocations += pContext->stats[i].CInvocations; + pStats->CsInvocations += pContext->stats[i].CsInvocations; + pStats->CPrimitives += pContext->stats[i].CPrimitives; + pStats->GsPrimitives += pContext->stats[i].GsPrimitives; + + for (uint32_t stream = 0; stream < MAX_SO_STREAMS; ++stream) + { + pStats->SoWriteOffset[stream] += pContext->stats[i].SoWriteOffset[stream]; + + /// @note client is required to provide valid write offset before every draw, so we clear + /// out the contents of the write offset when storing stats + pContext->stats[i].SoWriteOffset[stream] = 0; + + pStats->SoPrimStorageNeeded[stream] += pContext->stats[i].SoPrimStorageNeeded[stream]; + pStats->SoNumPrimsWritten[stream] += pContext->stats[i].SoNumPrimsWritten[stream]; + } + } +} + +template +void ClearRasterTile(BYTE *pTileBuffer, simdvector &value) +{ + auto lambda = [&](int comp) + { + FormatTraits::storeSOA(comp, pTileBuffer, value.v[comp]); + pTileBuffer += (KNOB_SIMD_WIDTH * FormatTraits::GetBPC(comp) / 8); + }; + + const uint32_t numIter = (KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM) * (KNOB_TILE_X_DIM / SIMD_TILE_X_DIM); + for (uint32_t i = 0; i < numIter; ++i) + { + UnrollerL<0, FormatTraits::numComps, 1>::step(lambda); + } +} + +template +INLINE void ClearMacroTile(DRAW_CONTEXT *pDC, SWR_RENDERTARGET_ATTACHMENT rt, uint32_t macroTile, DWORD clear[4]) +{ + // convert clear color to hottile format + // clear color is in RGBA float/uint32 + simdvector vClear; + for (uint32_t comp = 0; comp < FormatTraits::numComps; ++comp) + { + simdscalar vComp; + vComp = _simd_load1_ps((const float*)&clear[comp]); + if (FormatTraits::isNormalized(comp)) + { + vComp = _simd_mul_ps(vComp, _simd_set1_ps(FormatTraits::fromFloat(comp))); + vComp = _simd_castsi_ps(_simd_cvtps_epi32(vComp)); + } + vComp = FormatTraits::pack(comp, vComp); + vClear.v[FormatTraits::swizzle(comp)] = vComp; + } + + uint32_t tileX, tileY; + MacroTileMgr::getTileIndices(macroTile, tileX, tileY); + const API_STATE& state = GetApiState(pDC); + + int top = KNOB_MACROTILE_Y_DIM_FIXED * tileY; + int bottom = top + KNOB_MACROTILE_Y_DIM_FIXED - 1; + int left = KNOB_MACROTILE_X_DIM_FIXED * tileX; + int right = left + KNOB_MACROTILE_X_DIM_FIXED - 1; + + // intersect with scissor + top = std::max(top, state.scissorInFixedPoint.top); + left = std::max(left, state.scissorInFixedPoint.left); + bottom = std::min(bottom, state.scissorInFixedPoint.bottom); + right = std::min(right, state.scissorInFixedPoint.right); + + // translate to local hottile origin + top -= KNOB_MACROTILE_Y_DIM_FIXED * tileY; + bottom -= KNOB_MACROTILE_Y_DIM_FIXED * tileY; + left -= KNOB_MACROTILE_X_DIM_FIXED * tileX; + right -= KNOB_MACROTILE_X_DIM_FIXED * tileX; + + // convert to raster tiles + top >>= (KNOB_TILE_Y_DIM_SHIFT + FIXED_POINT_SHIFT); + bottom >>= (KNOB_TILE_Y_DIM_SHIFT + FIXED_POINT_SHIFT); + left >>= (KNOB_TILE_X_DIM_SHIFT + FIXED_POINT_SHIFT); + right >>= (KNOB_TILE_X_DIM_SHIFT + FIXED_POINT_SHIFT); + + const int numSamples = GetNumSamples(pDC->pState->state.rastState.sampleCount); + // compute steps between raster tile samples / raster tiles / macro tile rows + const uint32_t rasterTileSampleStep = KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8; + const uint32_t rasterTileStep = (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * (FormatTraits::bpp / 8)) * numSamples; + const uint32_t macroTileRowStep = (KNOB_MACROTILE_X_DIM / KNOB_TILE_X_DIM) * rasterTileStep; + const uint32_t pitch = (FormatTraits::bpp * KNOB_MACROTILE_X_DIM / 8); + + HOTTILE *pHotTile = pDC->pContext->pHotTileMgr->GetHotTile(pDC->pContext, pDC, macroTile, rt, true, numSamples); + uint32_t rasterTileStartOffset = (ComputeTileOffset2D< TilingTraits::bpp > >(pitch, left, top)) * numSamples; + uint8_t* pRasterTileRow = pHotTile->pBuffer + rasterTileStartOffset; //(ComputeTileOffset2D< TilingTraits::bpp > >(pitch, x, y)) * numSamples; + + // loop over all raster tiles in the current hot tile + for (int y = top; y <= bottom; ++y) + { + uint8_t* pRasterTile = pRasterTileRow; + for (int x = left; x <= right; ++x) + { + for( int sampleNum = 0; sampleNum < numSamples; sampleNum++) + { + ClearRasterTile(pRasterTile, vClear); + pRasterTile += rasterTileSampleStep; + } + } + pRasterTileRow += macroTileRowStep; + } + + pHotTile->state = HOTTILE_DIRTY; +} + + +void ProcessClearBE(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pUserData) +{ + if (KNOB_FAST_CLEAR) + { + CLEAR_DESC *pClear = (CLEAR_DESC*)pUserData; + SWR_CONTEXT *pContext = pDC->pContext; + SWR_MULTISAMPLE_COUNT sampleCount = pDC->pState->state.rastState.sampleCount; + uint32_t numSamples = GetNumSamples(sampleCount); + + SWR_ASSERT(pClear->flags.bits != 0); // shouldn't be here without a reason. + + RDTSC_START(BEClear); + + if (pClear->flags.mask & SWR_CLEAR_COLOR) + { + HOTTILE *pHotTile = pContext->pHotTileMgr->GetHotTile(pContext, pDC, macroTile, SWR_ATTACHMENT_COLOR0, true, numSamples); + // All we want to do here is to mark the hot tile as being in a "needs clear" state. + pHotTile->clearData[0] = *(DWORD*)&(pClear->clearRTColor[0]); + pHotTile->clearData[1] = *(DWORD*)&(pClear->clearRTColor[1]); + pHotTile->clearData[2] = *(DWORD*)&(pClear->clearRTColor[2]); + pHotTile->clearData[3] = *(DWORD*)&(pClear->clearRTColor[3]); + pHotTile->state = HOTTILE_CLEAR; + } + + if (pClear->flags.mask & SWR_CLEAR_DEPTH) + { + HOTTILE *pHotTile = pContext->pHotTileMgr->GetHotTile(pContext, pDC, macroTile, SWR_ATTACHMENT_DEPTH, true, numSamples); + pHotTile->clearData[0] = *(DWORD*)&pClear->clearDepth; + pHotTile->state = HOTTILE_CLEAR; + } + + if (pClear->flags.mask & SWR_CLEAR_STENCIL) + { + HOTTILE *pHotTile = pContext->pHotTileMgr->GetHotTile(pContext, pDC, macroTile, SWR_ATTACHMENT_STENCIL, true, numSamples); + + pHotTile->clearData[0] = *(DWORD*)&pClear->clearStencil; + pHotTile->state = HOTTILE_CLEAR; + } + + RDTSC_STOP(BEClear, 0, 0); + } + else + { + // Legacy clear + CLEAR_DESC *pClear = (CLEAR_DESC*)pUserData; + RDTSC_START(BEClear); + + if (pClear->flags.mask & SWR_CLEAR_COLOR) + { + /// @todo clear data should come in as RGBA32_FLOAT + DWORD clearData[4]; + float clearFloat[4]; + clearFloat[0] = ((BYTE*)(&pClear->clearRTColor))[0] / 255.0f; + clearFloat[1] = ((BYTE*)(&pClear->clearRTColor))[1] / 255.0f; + clearFloat[2] = ((BYTE*)(&pClear->clearRTColor))[2] / 255.0f; + clearFloat[3] = ((BYTE*)(&pClear->clearRTColor))[3] / 255.0f; + clearData[0] = *(DWORD*)&clearFloat[0]; + clearData[1] = *(DWORD*)&clearFloat[1]; + clearData[2] = *(DWORD*)&clearFloat[2]; + clearData[3] = *(DWORD*)&clearFloat[3]; + + PFN_CLEAR_TILES pfnClearTiles = sClearTilesTable[KNOB_COLOR_HOT_TILE_FORMAT]; + SWR_ASSERT(pfnClearTiles != nullptr); + + pfnClearTiles(pDC, SWR_ATTACHMENT_COLOR0, macroTile, clearData); + } + + if (pClear->flags.mask & SWR_CLEAR_DEPTH) + { + DWORD clearData[4]; + clearData[0] = *(DWORD*)&pClear->clearDepth; + PFN_CLEAR_TILES pfnClearTiles = sClearTilesTable[KNOB_DEPTH_HOT_TILE_FORMAT]; + SWR_ASSERT(pfnClearTiles != nullptr); + + pfnClearTiles(pDC, SWR_ATTACHMENT_DEPTH, macroTile, clearData); + } + + if (pClear->flags.mask & SWR_CLEAR_STENCIL) + { + uint32_t value = pClear->clearStencil; + DWORD clearData[4]; + clearData[0] = *(DWORD*)&value; + PFN_CLEAR_TILES pfnClearTiles = sClearTilesTable[KNOB_STENCIL_HOT_TILE_FORMAT]; + + pfnClearTiles(pDC, SWR_ATTACHMENT_STENCIL, macroTile, clearData); + } + + RDTSC_STOP(BEClear, 0, 0); + } +} + + +void ProcessStoreTileBE(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pData) +{ + RDTSC_START(BEStoreTiles); + STORE_TILES_DESC *pDesc = (STORE_TILES_DESC*)pData; + SWR_CONTEXT *pContext = pDC->pContext; + +#ifdef KNOB_ENABLE_RDTSC + uint32_t numTiles = 0; +#endif + SWR_FORMAT srcFormat; + switch (pDesc->attachment) + { + case SWR_ATTACHMENT_COLOR0: + case SWR_ATTACHMENT_COLOR1: + case SWR_ATTACHMENT_COLOR2: + case SWR_ATTACHMENT_COLOR3: + case SWR_ATTACHMENT_COLOR4: + case SWR_ATTACHMENT_COLOR5: + case SWR_ATTACHMENT_COLOR6: + case SWR_ATTACHMENT_COLOR7: srcFormat = KNOB_COLOR_HOT_TILE_FORMAT; break; + case SWR_ATTACHMENT_DEPTH: srcFormat = KNOB_DEPTH_HOT_TILE_FORMAT; break; + case SWR_ATTACHMENT_STENCIL: srcFormat = KNOB_STENCIL_HOT_TILE_FORMAT; break; + default: SWR_ASSERT(false, "Unknown attachment: %d", pDesc->attachment); srcFormat = KNOB_COLOR_HOT_TILE_FORMAT; break; + } + + uint32_t x, y; + MacroTileMgr::getTileIndices(macroTile, x, y); + + // Only need to store the hottile if it's been rendered to... + HOTTILE *pHotTile = pContext->pHotTileMgr->GetHotTile(pContext, pDC, macroTile, pDesc->attachment, false); + if (pHotTile) + { + // clear if clear is pending (i.e., not rendered to), then mark as dirty for store. + if (pHotTile->state == HOTTILE_CLEAR) + { + PFN_CLEAR_TILES pfnClearTiles = sClearTilesTable[srcFormat]; + SWR_ASSERT(pfnClearTiles != nullptr); + + pfnClearTiles(pDC, pDesc->attachment, macroTile, pHotTile->clearData); + } + + if (pHotTile->state == HOTTILE_DIRTY || pDesc->postStoreTileState == (SWR_TILE_STATE)HOTTILE_DIRTY) + { + int destX = KNOB_MACROTILE_X_DIM * x; + int destY = KNOB_MACROTILE_Y_DIM * y; + + pContext->pfnStoreTile(GetPrivateState(pDC), srcFormat, + pDesc->attachment, destX, destY, pHotTile->renderTargetArrayIndex, pHotTile->pBuffer); + } + + + if (pHotTile->state == HOTTILE_DIRTY || pHotTile->state == HOTTILE_RESOLVED) + { + pHotTile->state = (HOTTILE_STATE)pDesc->postStoreTileState; + } + } + RDTSC_STOP(BEStoreTiles, numTiles, pDC->drawId); +} + + +void ProcessInvalidateTilesBE(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pData) +{ + INVALIDATE_TILES_DESC *pDesc = (INVALIDATE_TILES_DESC*)pData; + SWR_CONTEXT *pContext = pDC->pContext; + + for (uint32_t i = 0; i < SWR_NUM_ATTACHMENTS; ++i) + { + if (pDesc->attachmentMask & (1 << i)) + { + HOTTILE *pHotTile = pContext->pHotTileMgr->GetHotTile(pContext, pDC, macroTile, (SWR_RENDERTARGET_ATTACHMENT)i, false); + if (pHotTile) + { + pHotTile->state = HOTTILE_INVALID; + } + } + } +} + +#if KNOB_SIMD_WIDTH == 8 +const __m256 vQuadCenterOffsetsX = { 0.5, 1.5, 0.5, 1.5, 2.5, 3.5, 2.5, 3.5 }; +const __m256 vQuadCenterOffsetsY = { 0.5, 0.5, 1.5, 1.5, 0.5, 0.5, 1.5, 1.5 }; +const __m256 vQuadULOffsetsX ={0.0, 1.0, 0.0, 1.0, 2.0, 3.0, 2.0, 3.0}; +const __m256 vQuadULOffsetsY ={0.0, 0.0, 1.0, 1.0, 0.0, 0.0, 1.0, 1.0}; +#define MASK 0xff +#else +#error Unsupported vector width +#endif + +INLINE +bool CanEarlyZ(const SWR_PS_STATE *pPSState) +{ + return (pPSState->forceEarlyZ || (!pPSState->writesODepth && !pPSState->usesSourceDepth && !pPSState->usesUAV)); +} + +simdmask ComputeUserClipMask(uint8_t clipMask, float* pUserClipBuffer, simdscalar vI, simdscalar vJ) +{ + simdscalar vClipMask = _simd_setzero_ps(); + uint32_t numClipDistance = _mm_popcnt_u32(clipMask); + + for (uint32_t i = 0; i < numClipDistance; ++i) + { + // pull triangle clip distance values from clip buffer + simdscalar vA = _simd_broadcast_ss(pUserClipBuffer++); + simdscalar vB = _simd_broadcast_ss(pUserClipBuffer++); + simdscalar vC = _simd_broadcast_ss(pUserClipBuffer++); + + // interpolate + simdscalar vInterp = vplaneps(vA, vB, vC, vI, vJ); + + // clip if interpolated clip distance is < 0 || NAN + simdscalar vCull = _simd_cmp_ps(_simd_setzero_ps(), vInterp, _CMP_NLE_UQ); + + vClipMask = _simd_or_ps(vClipMask, vCull); + } + + return _simd_movemask_ps(vClipMask); +} + +template +INLINE void generateInputCoverage(const uint64_t *const coverageMask, uint32_t (&inputMask)[KNOB_SIMD_WIDTH], const uint32_t sampleMask) +{ + + // will need to update for avx512 + assert(KNOB_SIMD_WIDTH == 8); + + __m256i mask[2]; + __m256i sampleCoverage[2]; + if(bIsStandardPattern) + { + __m256i src = _mm256_set1_epi32(0); + __m256i index0 = _mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0), index1; + + if(MultisampleTraits::numSamples == 1) + { + mask[0] = _mm256_set_epi32(0, 0, 0, 0, 0, 0, 0, -1); + } + else if(MultisampleTraits::numSamples == 2) + { + mask[0] = _mm256_set_epi32(0, 0, 0, 0, 0, 0, -1, -1); + } + else if(MultisampleTraits::numSamples == 4) + { + mask[0] = _mm256_set_epi32(0, 0, 0, 0, -1, -1, -1, -1); + } + else if(MultisampleTraits::numSamples == 8) + { + mask[0] = _mm256_set1_epi32(-1); + } + else if(MultisampleTraits::numSamples == 16) + { + mask[0] = _mm256_set1_epi32(-1); + mask[1] = _mm256_set1_epi32(-1); + index1 = _mm256_set_epi32(15, 14, 13, 12, 11, 10, 9, 8); + } + + // gather coverage for samples 0-7 + sampleCoverage[0] = _mm256_castps_si256(_simd_mask_i32gather_ps(_mm256_castsi256_ps(src), (const float*)coverageMask, index0, _mm256_castsi256_ps(mask[0]), 8)); + if(MultisampleTraits::numSamples > 8) + { + // gather coverage for samples 8-15 + sampleCoverage[1] = _mm256_castps_si256(_simd_mask_i32gather_ps(_mm256_castsi256_ps(src), (const float*)coverageMask, index1, _mm256_castsi256_ps(mask[1]), 8)); + } + } + else + { + // center coverage is the same for all samples; just broadcast to the sample slots + uint32_t centerCoverage = ((uint32_t)(*coverageMask) & MASK); + if(MultisampleTraits::numSamples == 1) + { + sampleCoverage[0] = _mm256_set_epi32(0, 0, 0, 0, 0, 0, 0, centerCoverage); + } + else if(MultisampleTraits::numSamples == 2) + { + sampleCoverage[0] = _mm256_set_epi32(0, 0, 0, 0, 0, 0, centerCoverage, centerCoverage); + } + else if(MultisampleTraits::numSamples == 4) + { + sampleCoverage[0] = _mm256_set_epi32(0, 0, 0, 0, centerCoverage, centerCoverage, centerCoverage, centerCoverage); + } + else if(MultisampleTraits::numSamples == 8) + { + sampleCoverage[0] = _mm256_set1_epi32(centerCoverage); + } + else if(MultisampleTraits::numSamples == 16) + { + sampleCoverage[0] = _mm256_set1_epi32(centerCoverage); + sampleCoverage[1] = _mm256_set1_epi32(centerCoverage); + } + } + + mask[0] = _mm256_set_epi8(-1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 0xC, 0x8, 0x4, 0x0, + -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, 0xC, 0x8, 0x4, 0x0); + // pull out the the 8bit 4x2 coverage for samples 0-7 into the lower 32 bits of each 128bit lane + __m256i packedCoverage0 = _simd_shuffle_epi8(sampleCoverage[0], mask[0]); + + __m256i packedCoverage1; + if(MultisampleTraits::numSamples > 8) + { + // pull out the the 8bit 4x2 coverage for samples 8-15 into the lower 32 bits of each 128bit lane + packedCoverage1 = _simd_shuffle_epi8(sampleCoverage[1], mask[0]); + } + +#if (KNOB_ARCH == KNOB_ARCH_AVX) + // pack lower 32 bits of each 128 bit lane into lower 64 bits of single 128 bit lane + __m256i hiToLow = _mm256_permute2f128_si256(packedCoverage0, packedCoverage0, 0x83); + __m256 shufRes = _mm256_shuffle_ps(_mm256_castsi256_ps(hiToLow), _mm256_castsi256_ps(hiToLow), _MM_SHUFFLE(1, 1, 0, 1)); + packedCoverage0 = _mm256_castps_si256(_mm256_blend_ps(_mm256_castsi256_ps(packedCoverage0), shufRes, 0xFE)); + + __m256i packedSampleCoverage; + if(MultisampleTraits::numSamples > 8) + { + // pack lower 32 bits of each 128 bit lane into upper 64 bits of single 128 bit lane + hiToLow = _mm256_permute2f128_si256(packedCoverage1, packedCoverage1, 0x83); + shufRes = _mm256_shuffle_ps(_mm256_castsi256_ps(hiToLow), _mm256_castsi256_ps(hiToLow), _MM_SHUFFLE(1, 1, 0, 1)); + shufRes = _mm256_blend_ps(_mm256_castsi256_ps(packedCoverage1), shufRes, 0xFE); + packedCoverage1 = _mm256_castps_si256(_mm256_castpd_ps(_mm256_shuffle_pd(_mm256_castps_pd(shufRes), _mm256_castps_pd(shufRes), 0x01))); + packedSampleCoverage = _mm256_castps_si256(_mm256_blend_ps(_mm256_castsi256_ps(packedCoverage0), _mm256_castsi256_ps(packedCoverage1), 0xFC)); + } + else + { + packedSampleCoverage = packedCoverage0; + } +#else + __m256i permMask = _mm256_set_epi32(0x7, 0x7, 0x7, 0x7, 0x7, 0x7, 0x4, 0x0); + // pack lower 32 bits of each 128 bit lane into lower 64 bits of single 128 bit lane + packedCoverage0 = _mm256_permutevar8x32_epi32(packedCoverage0, permMask); + + __m256i packedSampleCoverage; + if(MultisampleTraits::numSamples > 8) + { + permMask = _mm256_set_epi32(0x7, 0x7, 0x7, 0x7, 0x4, 0x0, 0x7, 0x7); + // pack lower 32 bits of each 128 bit lane into upper 64 bits of single 128 bit lane + packedCoverage1 = _mm256_permutevar8x32_epi32(packedCoverage1, permMask); + + // blend coverage masks for samples 0-7 and samples 8-15 into single 128 bit lane + packedSampleCoverage = _mm256_blend_epi32(packedCoverage0, packedCoverage1, 0x0C); + } + else + { + packedSampleCoverage = packedCoverage0; + } +#endif + + for(int32_t i = KNOB_SIMD_WIDTH - 1; i >= 0; i--) + { + // convert packed sample coverage masks into single coverage masks for all samples for each pixel in the 4x2 + inputMask[i] = _simd_movemask_epi8(packedSampleCoverage); + + if(!bForcedSampleCount) + { + // input coverage has to be anded with sample mask if MSAA isn't forced on + inputMask[i] &= sampleMask; + } + + // shift to the next pixel in the 4x2 + packedSampleCoverage = _simd_slli_epi32(packedSampleCoverage, 1); + } +} + +template +INLINE void generateInputCoverage(const uint64_t *const coverageMask, __m256 &inputCoverage, const uint32_t sampleMask) +{ + uint32_t inputMask[KNOB_SIMD_WIDTH]; + generateInputCoverage(coverageMask, inputMask, sampleMask); + inputCoverage = _simd_castsi_ps(_mm256_set_epi32(inputMask[7], inputMask[6], inputMask[5], inputMask[4], inputMask[3], inputMask[2], inputMask[1], inputMask[0])); +} + +template +INLINE void CalcPixelBarycentrics(const BarycentricCoeffs& coeffs, SWR_PS_CONTEXT &psContext) +{ + if(perspMask) + { + // evaluate I,J + psContext.vI.center = vplaneps(coeffs.vIa, coeffs.vIb, coeffs.vIc, psContext.vX.center, psContext.vY.center); + psContext.vJ.center = vplaneps(coeffs.vJa, coeffs.vJb, coeffs.vJc, psContext.vX.center, psContext.vY.center); + psContext.vI.center = _simd_mul_ps(psContext.vI.center, coeffs.vRecipDet); + psContext.vJ.center = _simd_mul_ps(psContext.vJ.center, coeffs.vRecipDet); + + // interpolate 1/w + psContext.vOneOverW.center = vplaneps(coeffs.vAOneOverW, coeffs.vBOneOverW, coeffs.vCOneOverW, psContext.vI.center, psContext.vJ.center); + } +} + +template +INLINE void CalcSampleBarycentrics(const BarycentricCoeffs& coeffs, SWR_PS_CONTEXT &psContext) +{ + if(perspMask) + { + // evaluate I,J + psContext.vI.sample = vplaneps(coeffs.vIa, coeffs.vIb, coeffs.vIc, psContext.vX.sample, psContext.vY.sample); + psContext.vJ.sample = vplaneps(coeffs.vJa, coeffs.vJb, coeffs.vJc, psContext.vX.sample, psContext.vY.sample); + psContext.vI.sample = _simd_mul_ps(psContext.vI.sample, coeffs.vRecipDet); + psContext.vJ.sample = _simd_mul_ps(psContext.vJ.sample, coeffs.vRecipDet); + + // interpolate 1/w + psContext.vOneOverW.sample = vplaneps(coeffs.vAOneOverW, coeffs.vBOneOverW, coeffs.vCOneOverW, psContext.vI.sample, psContext.vJ.sample); + } +} + + +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +// Centroid behaves exactly as follows : +// (1) If all samples in the primitive are covered, the attribute is evaluated at the pixel center (even if the sample pattern does not happen to +// have a sample location there). +// (2) Else the attribute is evaluated at the first covered sample, in increasing order of sample index, where sample coverage is after ANDing the +// coverage with the SampleMask Rasterizer State. +// (3) If no samples are covered, such as on helper pixels executed off the bounds of a primitive to fill out 2x2 pixel stamps, the attribute is +// evaluated as follows : If the SampleMask Rasterizer state is a subset of the samples in the pixel, then the first sample covered by the +// SampleMask Rasterizer State is the evaluation point.Otherwise (full SampleMask), the pixel center is the evaluation point. +//////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////////// +template +INLINE void CalcCentroidPos(SWR_PS_CONTEXT &psContext, const uint64_t *const coverageMask, const uint32_t sampleMask, + const simdscalar vXSamplePosUL, const simdscalar vYSamplePosUL) +{ + uint32_t inputMask[KNOB_SIMD_WIDTH]; + + generateInputCoverage(coverageMask, inputMask, sampleMask); + + // Case (2) - partially covered pixel + + // scan for first covered sample per pixel in the 4x2 span + unsigned long sampleNum[KNOB_SIMD_WIDTH]; + (inputMask[0] > 0) ? (_BitScanForward(&sampleNum[0], inputMask[0])) : (sampleNum[0] = 0); + (inputMask[1] > 0) ? (_BitScanForward(&sampleNum[1], inputMask[1])) : (sampleNum[1] = 0); + (inputMask[2] > 0) ? (_BitScanForward(&sampleNum[2], inputMask[2])) : (sampleNum[2] = 0); + (inputMask[3] > 0) ? (_BitScanForward(&sampleNum[3], inputMask[3])) : (sampleNum[3] = 0); + (inputMask[4] > 0) ? (_BitScanForward(&sampleNum[4], inputMask[4])) : (sampleNum[4] = 0); + (inputMask[5] > 0) ? (_BitScanForward(&sampleNum[5], inputMask[5])) : (sampleNum[5] = 0); + (inputMask[6] > 0) ? (_BitScanForward(&sampleNum[6], inputMask[6])) : (sampleNum[6] = 0); + (inputMask[7] > 0) ? (_BitScanForward(&sampleNum[7], inputMask[7])) : (sampleNum[7] = 0); + + // look up and set the sample offsets from UL pixel corner for first covered sample + __m256 vXSample = _mm256_set_ps(MultisampleTraits::X(sampleNum[7]), + MultisampleTraits::X(sampleNum[6]), + MultisampleTraits::X(sampleNum[5]), + MultisampleTraits::X(sampleNum[4]), + MultisampleTraits::X(sampleNum[3]), + MultisampleTraits::X(sampleNum[2]), + MultisampleTraits::X(sampleNum[1]), + MultisampleTraits::X(sampleNum[0])); + + __m256 vYSample = _mm256_set_ps(MultisampleTraits::Y(sampleNum[7]), + MultisampleTraits::Y(sampleNum[6]), + MultisampleTraits::Y(sampleNum[5]), + MultisampleTraits::Y(sampleNum[4]), + MultisampleTraits::Y(sampleNum[3]), + MultisampleTraits::Y(sampleNum[2]), + MultisampleTraits::Y(sampleNum[1]), + MultisampleTraits::Y(sampleNum[0])); + // add sample offset to UL pixel corner + vXSample = _simd_add_ps(vXSamplePosUL, vXSample); + vYSample = _simd_add_ps(vYSamplePosUL, vYSample); + + // Case (1) and case (3b) - All samples covered or not covered with full SampleMask + static const __m256i vFullyCoveredMask = MultisampleTraits::FullSampleMask(); + __m256i vInputCoveragei = _mm256_set_epi32(inputMask[7], inputMask[6], inputMask[5], inputMask[4], inputMask[3], inputMask[2], inputMask[1], inputMask[0]); + __m256i vAllSamplesCovered = _simd_cmpeq_epi32(vInputCoveragei, vFullyCoveredMask); + + static const __m256i vZero = _simd_setzero_si(); + const __m256i vSampleMask = _simd_and_si(_simd_set1_epi32(sampleMask), vFullyCoveredMask); + __m256i vNoSamplesCovered = _simd_cmpeq_epi32(vInputCoveragei, vZero); + __m256i vIsFullSampleMask = _simd_cmpeq_epi32(vSampleMask, vFullyCoveredMask); + __m256i vCase3b = _simd_and_si(vNoSamplesCovered, vIsFullSampleMask); + + __m256i vEvalAtCenter = _simd_or_si(vAllSamplesCovered, vCase3b); + + // set the centroid position based on results from above + psContext.vX.centroid = _simd_blendv_ps(vXSample, psContext.vX.center, _simd_castsi_ps(vEvalAtCenter)); + psContext.vY.centroid = _simd_blendv_ps(vYSample, psContext.vY.center, _simd_castsi_ps(vEvalAtCenter)); + + // Case (3a) No samples covered and partial sample mask + __m256i vSomeSampleMaskSamples = _simd_cmplt_epi32(vSampleMask, vFullyCoveredMask); + // sample mask should never be all 0's for this case, but handle it anyways + unsigned long firstCoveredSampleMaskSample = 0; + (sampleMask > 0) ? (_BitScanForward(&firstCoveredSampleMaskSample, sampleMask)) : (firstCoveredSampleMaskSample = 0); + + __m256i vCase3a = _simd_and_si(vNoSamplesCovered, vSomeSampleMaskSamples); + + vXSample = _simd_set1_ps(MultisampleTraits::X(firstCoveredSampleMaskSample)); + vYSample = _simd_set1_ps(MultisampleTraits::Y(firstCoveredSampleMaskSample)); + + // blend in case 3a pixel locations + psContext.vX.centroid = _simd_blendv_ps(psContext.vX.centroid, vXSample, _simd_castsi_ps(vCase3a)); + psContext.vY.centroid = _simd_blendv_ps(psContext.vY.centroid, vYSample, _simd_castsi_ps(vCase3a)); +} + +template +INLINE void CalcCentroidBarycentrics(const BarycentricCoeffs& coeffs, SWR_PS_CONTEXT &psContext, + const uint64_t *const coverageMask, const uint32_t sampleMask, + const simdscalar vXSamplePosUL, const simdscalar vYSamplePosUL) +{ + static const bool bPersp = (bool)persp; + static const bool bIsStandardPattern = (bool)standardPattern; + static const bool bForcedMultisampleCount = (bool)forcedMultisampleCount; + + // calculate centroid positions + if(bPersp) + { + if(bIsStandardPattern) + { + ///@ todo: don't need to generate input coverage 2x if input coverage and centroid + CalcCentroidPos<(SWR_MULTISAMPLE_COUNT)sampleCount, bForcedMultisampleCount>(psContext, coverageMask, sampleMask, vXSamplePosUL, vYSamplePosUL); + } + else + { + static const __m256 pixelCenter = _simd_set1_ps(0.5f); + psContext.vX.centroid = _simd_add_ps(vXSamplePosUL, pixelCenter); + psContext.vY.centroid = _simd_add_ps(vYSamplePosUL, pixelCenter); + } + // evaluate I,J + psContext.vI.centroid = vplaneps(coeffs.vIa, coeffs.vIb, coeffs.vIc, psContext.vX.centroid, psContext.vY.centroid); + psContext.vJ.centroid = vplaneps(coeffs.vJa, coeffs.vJb, coeffs.vJc, psContext.vX.centroid, psContext.vY.centroid); + psContext.vI.centroid = _simd_mul_ps(psContext.vI.centroid, coeffs.vRecipDet); + psContext.vJ.centroid = _simd_mul_ps(psContext.vJ.centroid, coeffs.vRecipDet); + + // interpolate 1/w + psContext.vOneOverW.centroid = vplaneps(coeffs.vAOneOverW, coeffs.vBOneOverW, coeffs.vCOneOverW, psContext.vI.centroid, psContext.vJ.centroid); + } +} + +template +void OutputMerger(SWR_PS_CONTEXT &psContext, uint8_t* (&pColorBase)[SWR_NUM_RENDERTARGETS], uint32_t sample, const SWR_BLEND_STATE *pBlendState, + const PFN_BLEND_JIT_FUNC (&pfnBlendFunc)[SWR_NUM_RENDERTARGETS], simdscalar &coverageMask, simdscalar depthPassMask) +{ + // type safety guaranteed from template instantiation in BEChooser<>::GetFunc + static const SWR_MULTISAMPLE_COUNT sampleCount = (SWR_MULTISAMPLE_COUNT)sampleCountT; + uint32_t rasterTileColorOffset = MultisampleTraits::RasterTileColorOffset(sample); + for(uint32_t rt = 0; rt < NumRT; ++rt) + { + uint8_t *pColorSample; + if(sampleCount == SWR_MULTISAMPLE_1X) + { + pColorSample = pColorBase[rt]; + } + else + { + pColorSample = pColorBase[rt] + rasterTileColorOffset; + } + + const SWR_RENDER_TARGET_BLEND_STATE *pRTBlend = &pBlendState->renderTarget[rt]; + + // Blend outputs and update coverage mask for alpha test + if(pfnBlendFunc[rt] != nullptr) + { + pfnBlendFunc[rt]( + pBlendState, + psContext.shaded[rt], + psContext.shaded[1], + sample, + pColorSample, + psContext.shaded[rt], + &psContext.oMask, + (simdscalari*)&coverageMask); + } + + // final write mask + simdscalari outputMask = _simd_castps_si(_simd_and_ps(coverageMask, depthPassMask)); + + ///@todo can only use maskstore fast path if bpc is 32. Assuming hot tile is RGBA32_FLOAT. + static_assert(KNOB_COLOR_HOT_TILE_FORMAT == R32G32B32A32_FLOAT, "Unsupported hot tile format"); + + const uint32_t simd = KNOB_SIMD_WIDTH * sizeof(float); + + // store with color mask + if(!pRTBlend->writeDisableRed) + { + _simd_maskstore_ps((float*)pColorSample, outputMask, psContext.shaded[rt].x); + } + if(!pRTBlend->writeDisableGreen) + { + _simd_maskstore_ps((float*)(pColorSample + simd), outputMask, psContext.shaded[rt].y); + } + if(!pRTBlend->writeDisableBlue) + { + _simd_maskstore_ps((float*)(pColorSample + simd * 2), outputMask, psContext.shaded[rt].z); + } + if(!pRTBlend->writeDisableAlpha) + { + _simd_maskstore_ps((float*)(pColorSample + simd * 3), outputMask, psContext.shaded[rt].w); + } + } +} + +template +void BackendSingleSample(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t x, uint32_t y, SWR_TRIANGLE_DESC &work, RenderOutputBuffers &renderBuffers) +{ + RDTSC_START(BESetup); + // type safety guaranteed from template instantiation in BEChooser<>::GetFunc + static const bool bInputCoverage = (bool)inputCoverage; + static const bool bCentroidPos = (bool)centroidPos; + + SWR_CONTEXT *pContext = pDC->pContext; + const API_STATE& state = GetApiState(pDC); + const SWR_RASTSTATE& rastState = state.rastState; + const SWR_PS_STATE *pPSState = &state.psState; + const SWR_BLEND_STATE *pBlendState = &state.blendState; + const BACKEND_FUNCS& backendFuncs = pDC->pState->backendFuncs; + uint64_t coverageMask = work.coverageMask[0]; + + // broadcast scalars + BarycentricCoeffs coeffs; + coeffs.vIa = _simd_broadcast_ss(&work.I[0]); + coeffs.vIb = _simd_broadcast_ss(&work.I[1]); + coeffs.vIc = _simd_broadcast_ss(&work.I[2]); + + coeffs.vJa = _simd_broadcast_ss(&work.J[0]); + coeffs.vJb = _simd_broadcast_ss(&work.J[1]); + coeffs.vJc = _simd_broadcast_ss(&work.J[2]); + + coeffs.vZa = _simd_broadcast_ss(&work.Z[0]); + coeffs.vZb = _simd_broadcast_ss(&work.Z[1]); + coeffs.vZc = _simd_broadcast_ss(&work.Z[2]); + + coeffs.vRecipDet = _simd_broadcast_ss(&work.recipDet); + + coeffs.vAOneOverW = _simd_broadcast_ss(&work.OneOverW[0]); + coeffs.vBOneOverW = _simd_broadcast_ss(&work.OneOverW[1]); + coeffs.vCOneOverW = _simd_broadcast_ss(&work.OneOverW[2]); + + uint8_t *pColorBase[SWR_NUM_RENDERTARGETS]; + uint32_t NumRT = state.psState.numRenderTargets; + for(uint32_t rt = 0; rt < NumRT; ++rt) + { + pColorBase[rt] = renderBuffers.pColor[rt]; + } + uint8_t *pDepthBase = renderBuffers.pDepth, *pStencilBase = renderBuffers.pStencil; + RDTSC_STOP(BESetup, 0, 0); + + SWR_PS_CONTEXT psContext; + psContext.pAttribs = work.pAttribs; + psContext.pPerspAttribs = work.pPerspAttribs; + psContext.frontFace = work.triFlags.frontFacing; + psContext.primID = work.triFlags.primID; + + // save Ia/Ib/Ic and Ja/Jb/Jc if we need to reevaluate i/j/k in the shader because of pull attribs + psContext.I = work.I; + psContext.J = work.J; + psContext.recipDet = work.recipDet; + psContext.pRecipW = work.pRecipW; + psContext.pSamplePosX = (const float*)&MultisampleTraits::samplePosX; + psContext.pSamplePosY = (const float*)&MultisampleTraits::samplePosY; + + for(uint32_t yy = y; yy < y + KNOB_TILE_Y_DIM; yy += SIMD_TILE_Y_DIM) + { + // UL pixel corner + psContext.vY.UL = _simd_add_ps(vQuadULOffsetsY, _simd_set1_ps((float)yy)); + // pixel center + psContext.vY.center = _simd_add_ps(vQuadCenterOffsetsY, _simd_set1_ps((float)yy)); + + for(uint32_t xx = x; xx < x + KNOB_TILE_X_DIM; xx += SIMD_TILE_X_DIM) + { + if(bInputCoverage) + { + generateInputCoverage(&work.coverageMask[0], psContext.inputMask, pBlendState->sampleMask); + } + + if(coverageMask & MASK) + { + RDTSC_START(BEBarycentric); + psContext.vX.UL = _simd_add_ps(vQuadULOffsetsX, _simd_set1_ps((float)xx)); + // pixel center + psContext.vX.center = _simd_add_ps(vQuadCenterOffsetsX, _simd_set1_ps((float)xx)); + + backendFuncs.pfnCalcPixelBarycentrics(coeffs, psContext); + + if(bCentroidPos) + { + // for 1x case, centroid is pixel center + psContext.vX.centroid = psContext.vX.center; + psContext.vY.centroid = psContext.vY.center; + psContext.vI.centroid = psContext.vI.center; + psContext.vJ.centroid = psContext.vJ.center; + psContext.vOneOverW.centroid = psContext.vOneOverW.center; + } + + // interpolate z + psContext.vZ = vplaneps(coeffs.vZa, coeffs.vZb, coeffs.vZc, psContext.vI.center, psContext.vJ.center); + RDTSC_STOP(BEBarycentric, 0, 0); + + simdmask clipCoverageMask = coverageMask & MASK; + + // interpolate user clip distance if available + if(rastState.clipDistanceMask) + { + clipCoverageMask &= ~ComputeUserClipMask(rastState.clipDistanceMask, work.pUserClipBuffer, + psContext.vI.center, psContext.vJ.center); + } + + simdscalar vCoverageMask = vMask(clipCoverageMask); + simdscalar depthPassMask = vCoverageMask; + simdscalar stencilPassMask = vCoverageMask; + + // Early-Z? + if(CanEarlyZ(pPSState)) + { + RDTSC_START(BEEarlyDepthTest); + depthPassMask = DepthStencilTest(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, + psContext.vZ, pDepthBase, vCoverageMask, pStencilBase, &stencilPassMask); + RDTSC_STOP(BEEarlyDepthTest, 0, 0); + + // early-exit if no pixels passed depth or earlyZ is forced on + if(pPSState->forceEarlyZ || !_simd_movemask_ps(depthPassMask)) + { + DepthStencilWrite(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, psContext.vZ, + pDepthBase, depthPassMask, vCoverageMask, pStencilBase, stencilPassMask); + + if (!_simd_movemask_ps(depthPassMask)) + { + goto Endtile; + } + } + } + + psContext.sampleIndex = 0; + psContext.activeMask = _simd_castps_si(vCoverageMask); + + // execute pixel shader + RDTSC_START(BEPixelShader); + UPDATE_STAT(PsInvocations, _mm_popcnt_u32(_simd_movemask_ps(vCoverageMask))); + state.psState.pfnPixelShader(GetPrivateState(pDC), &psContext); + RDTSC_STOP(BEPixelShader, 0, 0); + + vCoverageMask = _simd_castsi_ps(psContext.activeMask); + + // late-Z + if(!CanEarlyZ(pPSState)) + { + RDTSC_START(BELateDepthTest); + depthPassMask = DepthStencilTest(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, + psContext.vZ, pDepthBase, vCoverageMask, pStencilBase, &stencilPassMask); + RDTSC_STOP(BELateDepthTest, 0, 0); + + if(!_simd_movemask_ps(depthPassMask)) + { + // need to call depth/stencil write for stencil write + DepthStencilWrite(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, psContext.vZ, + pDepthBase, depthPassMask, vCoverageMask, pStencilBase, stencilPassMask); + goto Endtile; + } + } + + uint32_t statMask = _simd_movemask_ps(depthPassMask); + uint32_t statCount = _mm_popcnt_u32(statMask); + UPDATE_STAT(DepthPassCount, statCount); + + // output merger + RDTSC_START(BEOutputMerger); + backendFuncs.pfnOutputMerger(psContext, pColorBase, 0, pBlendState, state.pfnBlendFunc, + vCoverageMask, depthPassMask); + + // do final depth write after all pixel kills + if (!pPSState->forceEarlyZ) + { + DepthStencilWrite(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, psContext.vZ, + pDepthBase, depthPassMask, vCoverageMask, pStencilBase, stencilPassMask); + } + RDTSC_STOP(BEOutputMerger, 0, 0); + } + +Endtile: + RDTSC_START(BEEndTile); + coverageMask >>= (SIMD_TILE_Y_DIM * SIMD_TILE_X_DIM); + pDepthBase += (KNOB_SIMD_WIDTH * FormatTraits::bpp) / 8; + pStencilBase += (KNOB_SIMD_WIDTH * FormatTraits::bpp) / 8; + + for(uint32_t rt = 0; rt < NumRT; ++rt) + { + pColorBase[rt] += (KNOB_SIMD_WIDTH * FormatTraits::bpp) / 8; + } + RDTSC_STOP(BEEndTile, 0, 0); + } + } +} + +template +void BackendSampleRate(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t x, uint32_t y, SWR_TRIANGLE_DESC &work, RenderOutputBuffers &renderBuffers) +{ + // type safety guaranteed from template instantiation in BEChooser<>::GetFunc + static const SWR_MULTISAMPLE_COUNT sampleCount = (SWR_MULTISAMPLE_COUNT)sampleCountT; + static const bool bInputCoverage = (bool)inputCoverage; + static const bool bCentroidPos = (bool)centroidPos; + + RDTSC_START(BESetup); + + SWR_CONTEXT *pContext = pDC->pContext; + const API_STATE& state = GetApiState(pDC); + const SWR_RASTSTATE& rastState = state.rastState; + const SWR_PS_STATE *pPSState = &state.psState; + const SWR_BLEND_STATE *pBlendState = &state.blendState; + const BACKEND_FUNCS& backendFuncs = pDC->pState->backendFuncs; + + // broadcast scalars + BarycentricCoeffs coeffs; + coeffs.vIa = _simd_broadcast_ss(&work.I[0]); + coeffs.vIb = _simd_broadcast_ss(&work.I[1]); + coeffs.vIc = _simd_broadcast_ss(&work.I[2]); + + coeffs.vJa = _simd_broadcast_ss(&work.J[0]); + coeffs.vJb = _simd_broadcast_ss(&work.J[1]); + coeffs.vJc = _simd_broadcast_ss(&work.J[2]); + + coeffs.vZa = _simd_broadcast_ss(&work.Z[0]); + coeffs.vZb = _simd_broadcast_ss(&work.Z[1]); + coeffs.vZc = _simd_broadcast_ss(&work.Z[2]); + + coeffs.vRecipDet = _simd_broadcast_ss(&work.recipDet); + + coeffs.vAOneOverW = _simd_broadcast_ss(&work.OneOverW[0]); + coeffs.vBOneOverW = _simd_broadcast_ss(&work.OneOverW[1]); + coeffs.vCOneOverW = _simd_broadcast_ss(&work.OneOverW[2]); + + uint8_t *pColorBase[SWR_NUM_RENDERTARGETS]; + uint32_t NumRT = state.psState.numRenderTargets; + for(uint32_t rt = 0; rt < NumRT; ++rt) + { + pColorBase[rt] = renderBuffers.pColor[rt]; + } + uint8_t *pDepthBase = renderBuffers.pDepth, *pStencilBase = renderBuffers.pStencil; + RDTSC_STOP(BESetup, 0, 0); + + SWR_PS_CONTEXT psContext; + psContext.pAttribs = work.pAttribs; + psContext.pPerspAttribs = work.pPerspAttribs; + psContext.pRecipW = work.pRecipW; + psContext.frontFace = work.triFlags.frontFacing; + psContext.primID = work.triFlags.primID; + + // save Ia/Ib/Ic and Ja/Jb/Jc if we need to reevaluate i/j/k in the shader because of pull attribs + psContext.I = work.I; + psContext.J = work.J; + psContext.recipDet = work.recipDet; + psContext.pSamplePosX = (const float*)&MultisampleTraits::samplePosX; + psContext.pSamplePosY = (const float*)&MultisampleTraits::samplePosY; + const uint32_t numSamples = MultisampleTraits::numSamples; + + for (uint32_t yy = y; yy < y + KNOB_TILE_Y_DIM; yy += SIMD_TILE_Y_DIM) + { + // UL pixel corner + psContext.vY.UL = _simd_add_ps(vQuadULOffsetsY, _simd_set1_ps((float)yy)); + // pixel center + psContext.vY.center = _simd_add_ps(vQuadCenterOffsetsY, _simd_set1_ps((float)yy)); + + for (uint32_t xx = x; xx < x + KNOB_TILE_X_DIM; xx += SIMD_TILE_X_DIM) + { + psContext.vX.UL = _simd_add_ps(vQuadULOffsetsX, _simd_set1_ps((float)xx)); + // pixel center + psContext.vX.center = _simd_add_ps(vQuadCenterOffsetsX, _simd_set1_ps((float)xx)); + + RDTSC_START(BEBarycentric); + backendFuncs.pfnCalcPixelBarycentrics(coeffs, psContext); + RDTSC_STOP(BEBarycentric, 0, 0); + + if(bInputCoverage) + { + generateInputCoverage(&work.coverageMask[0], psContext.inputMask, pBlendState->sampleMask); + } + + if(bCentroidPos) + { + ///@ todo: don't need to genererate input coverage 2x if input coverage and centroid + RDTSC_START(BEBarycentric); + backendFuncs.pfnCalcCentroidBarycentrics(coeffs, psContext, &work.coverageMask[0], pBlendState->sampleMask, psContext.vX.UL, psContext.vY.UL); + RDTSC_STOP(BEBarycentric, 0, 0); + } + + for(uint32_t sample = 0; sample < numSamples; sample++) + { + if (work.coverageMask[sample] & MASK) + { + RDTSC_START(BEBarycentric); + + // calculate per sample positions + psContext.vX.sample = _simd_add_ps(psContext.vX.UL, MultisampleTraits::vX(sample)); + psContext.vY.sample = _simd_add_ps(psContext.vY.UL, MultisampleTraits::vY(sample)); + + simdmask coverageMask = work.coverageMask[sample] & MASK; + simdscalar vCoverageMask = vMask(coverageMask); + + backendFuncs.pfnCalcSampleBarycentrics(coeffs, psContext); + + // interpolate z + psContext.vZ = vplaneps(coeffs.vZa, coeffs.vZb, coeffs.vZc, psContext.vI.sample, psContext.vJ.sample); + + RDTSC_STOP(BEBarycentric, 0, 0); + + // interpolate user clip distance if available + if (rastState.clipDistanceMask) + { + coverageMask &= ~ComputeUserClipMask(rastState.clipDistanceMask, work.pUserClipBuffer, + psContext.vI.sample, psContext.vJ.sample); + } + + simdscalar depthPassMask = vCoverageMask; + simdscalar stencilPassMask = vCoverageMask; + + // offset depth/stencil buffers current sample + uint8_t *pDepthSample = pDepthBase + MultisampleTraits::RasterTileDepthOffset(sample); + uint8_t *pStencilSample = pStencilBase + MultisampleTraits::RasterTileStencilOffset(sample); + + // Early-Z? + if (CanEarlyZ(pPSState)) + { + RDTSC_START(BEEarlyDepthTest); + depthPassMask = DepthStencilTest(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, + psContext.vZ, pDepthSample, vCoverageMask, pStencilSample, &stencilPassMask); + RDTSC_STOP(BEEarlyDepthTest, 0, 0); + + // early-exit if no samples passed depth or earlyZ is forced on. + if (pPSState->forceEarlyZ || !_simd_movemask_ps(depthPassMask)) + { + DepthStencilWrite(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, psContext.vZ, + pDepthSample, depthPassMask, vCoverageMask, pStencilSample, stencilPassMask); + + if (!_simd_movemask_ps(depthPassMask)) + { + work.coverageMask[sample] >>= (SIMD_TILE_Y_DIM * SIMD_TILE_X_DIM); + continue; + } + } + } + + psContext.sampleIndex = sample; + psContext.activeMask = _simd_castps_si(vCoverageMask); + + // execute pixel shader + RDTSC_START(BEPixelShader); + UPDATE_STAT(PsInvocations, _mm_popcnt_u32(_simd_movemask_ps(vCoverageMask))); + state.psState.pfnPixelShader(GetPrivateState(pDC), &psContext); + RDTSC_STOP(BEPixelShader, 0, 0); + + vCoverageMask = _simd_castsi_ps(psContext.activeMask); + + //// late-Z + if (!CanEarlyZ(pPSState)) + { + RDTSC_START(BELateDepthTest); + depthPassMask = DepthStencilTest(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, + psContext.vZ, pDepthSample, vCoverageMask, pStencilSample, &stencilPassMask); + RDTSC_STOP(BELateDepthTest, 0, 0); + + if (!_simd_movemask_ps(depthPassMask)) + { + // need to call depth/stencil write for stencil write + DepthStencilWrite(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, psContext.vZ, + pDepthSample, depthPassMask, vCoverageMask, pStencilSample, stencilPassMask); + + work.coverageMask[sample] >>= (SIMD_TILE_Y_DIM * SIMD_TILE_X_DIM); + continue; + } + } + + uint32_t statMask = _simd_movemask_ps(depthPassMask); + uint32_t statCount = _mm_popcnt_u32(statMask); + UPDATE_STAT(DepthPassCount, statCount); + + // output merger + RDTSC_START(BEOutputMerger); + backendFuncs.pfnOutputMerger(psContext, pColorBase, sample, pBlendState, state.pfnBlendFunc, + vCoverageMask, depthPassMask); + + // do final depth write after all pixel kills + if (!pPSState->forceEarlyZ) + { + DepthStencilWrite(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, psContext.vZ, + pDepthSample, depthPassMask, vCoverageMask, pStencilSample, stencilPassMask); + } + RDTSC_STOP(BEOutputMerger, 0, 0); + } + work.coverageMask[sample] >>= (SIMD_TILE_Y_DIM * SIMD_TILE_X_DIM); + } + RDTSC_START(BEEndTile); + pDepthBase += (KNOB_SIMD_WIDTH * FormatTraits::bpp) / 8; + pStencilBase += (KNOB_SIMD_WIDTH * FormatTraits::bpp) / 8; + + for (uint32_t rt = 0; rt < NumRT; ++rt) + { + pColorBase[rt] += (KNOB_SIMD_WIDTH * FormatTraits::bpp) / 8; + } + RDTSC_STOP(BEEndTile, 0, 0); + } + } +} + +template +void BackendPixelRate(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t x, uint32_t y, SWR_TRIANGLE_DESC &work, RenderOutputBuffers &renderBuffers) +{ + // type safety guaranteed from template instantiation in BEChooser<>::GetFunc + static const SWR_MULTISAMPLE_COUNT sampleCount = (SWR_MULTISAMPLE_COUNT)sampleCountT; + static const bool bIsStandardPattern = (bool)samplePattern; + static const bool bInputCoverage = (bool)inputCoverage; + static const bool bCentroidPos = (bool)centroidPos; + static const bool bForcedSampleCount = (bool)forcedSampleCount; + + RDTSC_START(BESetup); + + SWR_CONTEXT *pContext = pDC->pContext; + const API_STATE& state = GetApiState(pDC); + const SWR_RASTSTATE& rastState = state.rastState; + const SWR_PS_STATE *pPSState = &state.psState; + const SWR_BLEND_STATE *pBlendState = &state.blendState; + const BACKEND_FUNCS& backendFuncs = pDC->pState->backendFuncs; + + // broadcast scalars + BarycentricCoeffs coeffs; + coeffs.vIa = _simd_broadcast_ss(&work.I[0]); + coeffs.vIb = _simd_broadcast_ss(&work.I[1]); + coeffs.vIc = _simd_broadcast_ss(&work.I[2]); + + coeffs.vJa = _simd_broadcast_ss(&work.J[0]); + coeffs.vJb = _simd_broadcast_ss(&work.J[1]); + coeffs.vJc = _simd_broadcast_ss(&work.J[2]); + + coeffs.vZa = _simd_broadcast_ss(&work.Z[0]); + coeffs.vZb = _simd_broadcast_ss(&work.Z[1]); + coeffs.vZc = _simd_broadcast_ss(&work.Z[2]); + + coeffs.vRecipDet = _simd_broadcast_ss(&work.recipDet); + + coeffs.vAOneOverW = _simd_broadcast_ss(&work.OneOverW[0]); + coeffs.vBOneOverW = _simd_broadcast_ss(&work.OneOverW[1]); + coeffs.vCOneOverW = _simd_broadcast_ss(&work.OneOverW[2]); + + uint8_t *pColorBase[SWR_NUM_RENDERTARGETS]; + uint32_t NumRT = state.psState.numRenderTargets; + for(uint32_t rt = 0; rt < NumRT; ++rt) + { + pColorBase[rt] = renderBuffers.pColor[rt]; + } + uint8_t *pDepthBase = renderBuffers.pDepth, *pStencilBase = renderBuffers.pStencil; + RDTSC_STOP(BESetup, 0, 0); + + SWR_PS_CONTEXT psContext; + psContext.pAttribs = work.pAttribs; + psContext.pPerspAttribs = work.pPerspAttribs; + psContext.frontFace = work.triFlags.frontFacing; + psContext.primID = work.triFlags.primID; + psContext.pRecipW = work.pRecipW; + // save Ia/Ib/Ic and Ja/Jb/Jc if we need to reevaluate i/j/k in the shader because of pull attribs + psContext.I = work.I; + psContext.J = work.J; + psContext.recipDet = work.recipDet; + psContext.pSamplePosX = (const float*)&MultisampleTraits::samplePosX; + psContext.pSamplePosY = (const float*)&MultisampleTraits::samplePosY; + psContext.sampleIndex = 0; + + uint32_t numCoverageSamples; + if(bIsStandardPattern) + { + numCoverageSamples = MultisampleTraits::numSamples; + } + else + { + numCoverageSamples = 1; + } + + uint32_t numOMSamples; + // RT has to be single sample if we're in forcedMSAA mode + if(bForcedSampleCount && (sampleCount > SWR_MULTISAMPLE_1X)) + { + numOMSamples = 1; + } + // unless we're forced to single sample, in which case we run the OM at the sample count of the RT + else if(bForcedSampleCount && (sampleCount == SWR_MULTISAMPLE_1X)) + { + numOMSamples = GetNumSamples(pBlendState->sampleCount); + } + // else we're in normal MSAA mode and rasterizer and OM are running at the same sample count + else + { + numOMSamples = MultisampleTraits::numSamples; + } + + for(uint32_t yy = y; yy < y + KNOB_TILE_Y_DIM; yy += SIMD_TILE_Y_DIM) + { + psContext.vY.UL = _simd_add_ps(vQuadULOffsetsY, _simd_set1_ps((float)yy)); + psContext.vY.center = _simd_add_ps(vQuadCenterOffsetsY, _simd_set1_ps((float)yy)); + for(uint32_t xx = x; xx < x + KNOB_TILE_X_DIM; xx += SIMD_TILE_X_DIM) + { + simdscalar vZ[MultisampleTraits::numSamples]; + psContext.vX.UL = _simd_add_ps(vQuadULOffsetsX, _simd_set1_ps((float)xx)); + // set pixel center positions + psContext.vX.center = _simd_add_ps(vQuadCenterOffsetsX, _simd_set1_ps((float)xx)); + + if (bInputCoverage) + { + generateInputCoverage(&work.coverageMask[0], psContext.inputMask, pBlendState->sampleMask); + } + + if(bCentroidPos) + { + ///@ todo: don't need to genererate input coverage 2x if input coverage and centroid + RDTSC_START(BEBarycentric); + backendFuncs.pfnCalcCentroidBarycentrics(coeffs, psContext, &work.coverageMask[0], pBlendState->sampleMask, psContext.vX.UL, psContext.vY.UL); + RDTSC_STOP(BEBarycentric, 0, 0); + } + + // if oDepth written to, or there is a potential to discard any samples, we need to + // run the PS early, then interp or broadcast Z and test + if(pPSState->writesODepth || pPSState->killsPixel) + { + RDTSC_START(BEBarycentric); + backendFuncs.pfnCalcPixelBarycentrics(coeffs, psContext); + + // interpolate z + psContext.vZ = vplaneps(coeffs.vZa, coeffs.vZb, coeffs.vZc, psContext.vI.center, psContext.vJ.center); + RDTSC_STOP(BEBarycentric, 0, 0); + + // execute pixel shader + RDTSC_START(BEPixelShader); + state.psState.pfnPixelShader(GetPrivateState(pDC), &psContext); + RDTSC_STOP(BEPixelShader, 0, 0); + } + else + { + psContext.activeMask = _simd_set1_epi32(-1); + } + + // need to declare enough space for all samples + simdscalar vCoverageMask[MultisampleTraits::numSamples]; + simdscalar depthPassMask[MultisampleTraits::numSamples]; + simdscalar stencilPassMask[MultisampleTraits::numSamples]; + simdscalar anyDepthSamplePassed = _simd_setzero_ps(); + simdscalar anyStencilSamplePassed = _simd_setzero_ps(); + for(uint32_t sample = 0; sample < numCoverageSamples; sample++) + { + vCoverageMask[sample] = vMask(work.coverageMask[sample] & MASK); + + // pull mask back out for any discards and and with coverage + vCoverageMask[sample] = _simd_and_ps(vCoverageMask[sample], _simd_castsi_ps(psContext.activeMask)); + + if (!_simd_movemask_ps(vCoverageMask[sample])) + { + vCoverageMask[sample] = depthPassMask[sample] = stencilPassMask[sample] = _simd_setzero_ps(); + continue; + } + + if(bForcedSampleCount) + { + // candidate pixels (that passed coverage) will cause shader invocation if any bits in the samplemask are set + const simdscalar vSampleMask = _simd_castsi_ps(_simd_cmpgt_epi32(_simd_set1_epi32(pBlendState->sampleMask), _simd_setzero_si())); + anyDepthSamplePassed = _simd_or_ps(anyDepthSamplePassed, _simd_and_ps(vCoverageMask[sample], vSampleMask)); + continue; + } + + depthPassMask[sample] = vCoverageMask[sample]; + + // if oDepth isn't written to, we need to interpolate Z for each sample + // if clip distances are enabled, we need to interpolate for each sample + if(!pPSState->writesODepth || rastState.clipDistanceMask) + { + RDTSC_START(BEBarycentric); + if(bIsStandardPattern) + { + // calculate per sample positions + psContext.vX.sample = _simd_add_ps(psContext.vX.UL, MultisampleTraits::vX(sample)); + psContext.vY.sample = _simd_add_ps(psContext.vY.UL, MultisampleTraits::vY(sample)); + } + else + { + psContext.vX.sample = psContext.vX.center; + psContext.vY.sample = psContext.vY.center; + } + + // calc I & J per sample + backendFuncs.pfnCalcSampleBarycentrics(coeffs, psContext); + + // interpolate z + if (!pPSState->writesODepth) + { + vZ[sample] = vplaneps(coeffs.vZa, coeffs.vZb, coeffs.vZc, psContext.vI.sample, psContext.vJ.sample); + } + + ///@todo: perspective correct vs non-perspective correct clipping? + // interpolate clip distances + if (rastState.clipDistanceMask) + { + uint8_t clipMask = ComputeUserClipMask(rastState.clipDistanceMask, work.pUserClipBuffer, + psContext.vI.sample, psContext.vJ.sample); + vCoverageMask[sample] = _simd_and_ps(vCoverageMask[sample], vMask(~clipMask)); + } + RDTSC_STOP(BEBarycentric, 0, 0); + } + // else 'broadcast' and test psContext.vZ written from the PS each sample + else + { + vZ[sample] = psContext.vZ; + } + + // offset depth/stencil buffers current sample + uint8_t *pDepthSample = pDepthBase + MultisampleTraits::RasterTileDepthOffset(sample); + uint8_t * pStencilSample = pStencilBase + MultisampleTraits::RasterTileStencilOffset(sample); + + // ZTest for this sample + RDTSC_START(BEEarlyDepthTest); + stencilPassMask[sample] = vCoverageMask[sample]; + depthPassMask[sample] = DepthStencilTest(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, + vZ[sample], pDepthSample, vCoverageMask[sample], pStencilSample, &stencilPassMask[sample]); + RDTSC_STOP(BEEarlyDepthTest, 0, 0); + + anyDepthSamplePassed = _simd_or_ps(anyDepthSamplePassed, depthPassMask[sample]); + anyStencilSamplePassed = _simd_or_ps(anyStencilSamplePassed, stencilPassMask[sample]); + uint32_t statMask = _simd_movemask_ps(depthPassMask[sample]); + uint32_t statCount = _mm_popcnt_u32(statMask); + UPDATE_STAT(DepthPassCount, statCount); + } + + // if we didn't have to execute the PS early, and at least 1 sample passed the depth test, run the PS + if(!pPSState->writesODepth && !pPSState->killsPixel && _simd_movemask_ps(anyDepthSamplePassed)) + { + RDTSC_START(BEBarycentric); + backendFuncs.pfnCalcPixelBarycentrics(coeffs, psContext); + // interpolate z + psContext.vZ = vplaneps(coeffs.vZa, coeffs.vZb, coeffs.vZc, psContext.vI.center, psContext.vJ.center); + RDTSC_STOP(BEBarycentric, 0, 0); + + // execute pixel shader + RDTSC_START(BEPixelShader); + state.psState.pfnPixelShader(GetPrivateState(pDC), &psContext); + RDTSC_STOP(BEPixelShader, 0, 0); + } + ///@todo: make sure this works for kill pixel + else if(!_simd_movemask_ps(anyStencilSamplePassed)) + { + goto Endtile; + } + + // loop over all samples, broadcasting the results of the PS to all passing pixels + for(uint32_t sample = 0; sample < numOMSamples; sample++) + { + uint8_t *pDepthSample = pDepthBase + MultisampleTraits::RasterTileDepthOffset(sample); + uint8_t * pStencilSample = pStencilBase + MultisampleTraits::RasterTileStencilOffset(sample); + + // output merger + RDTSC_START(BEOutputMerger); + + // skip if none of the pixels for this sample passed + simdscalar coverageMaskSample; + simdscalar depthMaskSample; + simdscalar stencilMaskSample; + simdscalar vInterpolatedZ; + + // forcedSampleCount outputs to any pixels with covered samples not masked off by SampleMask + // depth test is disabled, so just set the z val to 0. + if(bForcedSampleCount) + { + coverageMaskSample = depthMaskSample = anyDepthSamplePassed; + vInterpolatedZ = _simd_setzero_ps(); + } + else if(bIsStandardPattern) + { + if(!_simd_movemask_ps(depthPassMask[sample])) + { + depthPassMask[sample] = _simd_setzero_ps(); + DepthStencilWrite(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, vZ[sample], pDepthSample, depthPassMask[sample], + vCoverageMask[sample], pStencilSample, stencilPassMask[sample]); + continue; + } + coverageMaskSample = vCoverageMask[sample]; + depthMaskSample = depthPassMask[sample]; + stencilMaskSample = stencilPassMask[sample]; + vInterpolatedZ = vZ[sample]; + } + else + { + // center pattern only needs to use a single depth test as all samples are at the same position + if(!_simd_movemask_ps(depthPassMask[0])) + { + depthPassMask[0] = _simd_setzero_ps(); + DepthStencilWrite(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, vZ[0], pDepthSample, depthPassMask[0], + vCoverageMask[0], pStencilSample, stencilPassMask[0]); + continue; + } + coverageMaskSample = (vCoverageMask[0]); + depthMaskSample = depthPassMask[0]; + stencilMaskSample = stencilPassMask[0]; + vInterpolatedZ = vZ[0]; + } + + // output merger + RDTSC_START(BEOutputMerger); + backendFuncs.pfnOutputMerger(psContext, pColorBase, sample, pBlendState, state.pfnBlendFunc, + coverageMaskSample, depthMaskSample); + + DepthStencilWrite(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, vInterpolatedZ, pDepthSample, depthMaskSample, + coverageMaskSample, pStencilSample, stencilMaskSample); + RDTSC_STOP(BEOutputMerger, 0, 0); + } + +Endtile: + RDTSC_START(BEEndTile); + for(uint32_t sample = 0; sample < numCoverageSamples; sample++) + { + work.coverageMask[sample] >>= (SIMD_TILE_Y_DIM * SIMD_TILE_X_DIM); + } + + pDepthBase += (KNOB_SIMD_WIDTH * FormatTraits::bpp) / 8; + pStencilBase += (KNOB_SIMD_WIDTH * FormatTraits::bpp) / 8; + + for(uint32_t rt = 0; rt < NumRT; ++rt) + { + pColorBase[rt] += (KNOB_SIMD_WIDTH * FormatTraits::bpp) / 8; + } + RDTSC_STOP(BEEndTile, 0, 0); + } + } +} +// optimized backend flow with NULL PS +template +void BackendNullPS(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t x, uint32_t y, SWR_TRIANGLE_DESC &work, RenderOutputBuffers &renderBuffers) +{ + RDTSC_START(BESetup); + + static const SWR_MULTISAMPLE_COUNT sampleCount = (SWR_MULTISAMPLE_COUNT)sampleCountT; + SWR_CONTEXT *pContext = pDC->pContext; + const API_STATE& state = GetApiState(pDC); + const BACKEND_FUNCS& backendFuncs = pDC->pState->backendFuncs; + + // broadcast scalars + BarycentricCoeffs coeffs; + coeffs.vIa = _simd_broadcast_ss(&work.I[0]); + coeffs.vIb = _simd_broadcast_ss(&work.I[1]); + coeffs.vIc = _simd_broadcast_ss(&work.I[2]); + + coeffs.vJa = _simd_broadcast_ss(&work.J[0]); + coeffs.vJb = _simd_broadcast_ss(&work.J[1]); + coeffs.vJc = _simd_broadcast_ss(&work.J[2]); + + coeffs.vZa = _simd_broadcast_ss(&work.Z[0]); + coeffs.vZb = _simd_broadcast_ss(&work.Z[1]); + coeffs.vZc = _simd_broadcast_ss(&work.Z[2]); + + coeffs.vRecipDet = _simd_broadcast_ss(&work.recipDet); + + BYTE *pDepthBase = renderBuffers.pDepth, *pStencilBase = renderBuffers.pStencil; + + RDTSC_STOP(BESetup, 0, 0); + + SWR_PS_CONTEXT psContext; + for (uint32_t yy = y; yy < y + KNOB_TILE_Y_DIM; yy += SIMD_TILE_Y_DIM) + { + // UL pixel corner + simdscalar vYSamplePosUL = _simd_add_ps(vQuadULOffsetsY, _simd_set1_ps((float)yy)); + + for (uint32_t xx = x; xx < x + KNOB_TILE_X_DIM; xx += SIMD_TILE_X_DIM) + { + // UL pixel corners + simdscalar vXSamplePosUL = _simd_add_ps(vQuadULOffsetsX, _simd_set1_ps((float)xx)); + + // iterate over active samples + unsigned long sample = 0; + uint32_t sampleMask = state.blendState.sampleMask; + while (_BitScanForward(&sample, sampleMask)) + { + sampleMask &= ~(1 << sample); + if (work.coverageMask[sample] & MASK) + { + RDTSC_START(BEBarycentric); + // calculate per sample positions + psContext.vX.sample = _simd_add_ps(vXSamplePosUL, MultisampleTraits::vX(sample)); + psContext.vY.sample = _simd_add_ps(vYSamplePosUL, MultisampleTraits::vY(sample)); + + backendFuncs.pfnCalcSampleBarycentrics(coeffs, psContext); + + // interpolate z + psContext.vZ = vplaneps(coeffs.vZa, coeffs.vZb, coeffs.vZc, psContext.vI.sample, psContext.vJ.sample); + + RDTSC_STOP(BEBarycentric, 0, 0); + + simdscalar vCoverageMask = vMask(work.coverageMask[sample] & MASK); + simdscalar stencilPassMask = vCoverageMask; + + // offset depth/stencil buffers current sample + uint8_t *pDepthSample = pDepthBase + MultisampleTraits::RasterTileDepthOffset(sample); + uint8_t *pStencilSample = pStencilBase + MultisampleTraits::RasterTileStencilOffset(sample); + + RDTSC_START(BEEarlyDepthTest); + simdscalar depthPassMask = DepthStencilTest(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, + psContext.vZ, pDepthSample, vCoverageMask, pStencilSample, &stencilPassMask); + DepthStencilWrite(&state.vp[0], &state.depthStencilState, work.triFlags.frontFacing, psContext.vZ, + pDepthSample, depthPassMask, vCoverageMask, pStencilSample, stencilPassMask); + RDTSC_STOP(BEEarlyDepthTest, 0, 0); + + uint32_t statMask = _simd_movemask_ps(depthPassMask); + uint32_t statCount = _mm_popcnt_u32(statMask); + UPDATE_STAT(DepthPassCount, statCount); + } + work.coverageMask[sample] >>= (SIMD_TILE_Y_DIM * SIMD_TILE_X_DIM); + } + pDepthBase += (KNOB_SIMD_WIDTH * FormatTraits::bpp) / 8; + pStencilBase += (KNOB_SIMD_WIDTH * FormatTraits::bpp) / 8; + } + } +} + +void InitClearTilesTable() +{ + memset(sClearTilesTable, 0, sizeof(sClearTilesTable)); + + sClearTilesTable[R8G8B8A8_UNORM] = ClearMacroTile; + sClearTilesTable[B8G8R8A8_UNORM] = ClearMacroTile; + sClearTilesTable[R32_FLOAT] = ClearMacroTile; + sClearTilesTable[R32G32B32A32_FLOAT] = ClearMacroTile; + sClearTilesTable[R8_UINT] = ClearMacroTile; +} + +PFN_BACKEND_FUNC gBackendNullPs[SWR_MULTISAMPLE_TYPE_MAX]; +PFN_BACKEND_FUNC gBackendSingleSample[2][2] = {}; +PFN_BACKEND_FUNC gBackendPixelRateTable[SWR_MULTISAMPLE_TYPE_MAX][SWR_MSAA_SAMPLE_PATTERN_MAX][SWR_INPUT_COVERAGE_MAX][2][2] = {}; +PFN_BACKEND_FUNC gBackendSampleRateTable[SWR_MULTISAMPLE_TYPE_MAX][SWR_INPUT_COVERAGE_MAX][2] = {}; +PFN_OUTPUT_MERGER gBackendOutputMergerTable[SWR_NUM_RENDERTARGETS+1][SWR_MULTISAMPLE_TYPE_MAX] = {}; +PFN_CALC_PIXEL_BARYCENTRICS gPixelBarycentricTable[2] = {}; +PFN_CALC_SAMPLE_BARYCENTRICS gSampleBarycentricTable[2] = {}; +PFN_CALC_CENTROID_BARYCENTRICS gCentroidBarycentricTable[SWR_MULTISAMPLE_TYPE_MAX][2][2][2] = {}; + +// Recursive template used to auto-nest conditionals. Converts dynamic enum function +// arguments to static template arguments. +template +struct OMChooser +{ + // Last Arg Terminator + static PFN_OUTPUT_MERGER GetFunc(SWR_MULTISAMPLE_COUNT tArg) + { + switch(tArg) + { + case SWR_MULTISAMPLE_1X: return OutputMerger; break; + case SWR_MULTISAMPLE_2X: return OutputMerger; break; + case SWR_MULTISAMPLE_4X: return OutputMerger; break; + case SWR_MULTISAMPLE_8X: return OutputMerger; break; + case SWR_MULTISAMPLE_16X: return OutputMerger; break; + default: + SWR_ASSERT(0 && "Invalid sample count\n"); + return nullptr; + break; + } + } + + // Recursively parse args + template + static PFN_OUTPUT_MERGER GetFunc(uint32_t tArg, TArgsT... remainingArgs) + { + switch(tArg) + { + case 0: return OMChooser::GetFunc(remainingArgs...); break; + case 1: return OMChooser::GetFunc(remainingArgs...); break; + case 2: return OMChooser::GetFunc(remainingArgs...); break; + case 3: return OMChooser::GetFunc(remainingArgs...); break; + case 4: return OMChooser::GetFunc(remainingArgs...); break; + case 5: return OMChooser::GetFunc(remainingArgs...); break; + case 6: return OMChooser::GetFunc(remainingArgs...); break; + case 7: return OMChooser::GetFunc(remainingArgs...); break; + case 8: return OMChooser::GetFunc(remainingArgs...); break; + default: + SWR_ASSERT(0 && "Invalid RT index\n"); + return nullptr; + break; + } + } +}; + +// Recursive template used to auto-nest conditionals. Converts dynamic enum function +// arguments to static template arguments. +template +struct BECentroidBarycentricChooser +{ + + // Last Arg Terminator + template + static PFN_CALC_CENTROID_BARYCENTRICS GetFunc(uint32_t tArg) + { + if(tArg > 0) + { + return CalcCentroidBarycentrics; + } + + return CalcCentroidBarycentrics; + } + + // Recursively parse args + template + static PFN_CALC_CENTROID_BARYCENTRICS GetFunc(SWR_MULTISAMPLE_COUNT tArg, TArgsT... remainingArgs) + { + switch(tArg) + { + case SWR_MULTISAMPLE_1X: return BECentroidBarycentricChooser::GetFunc(remainingArgs...); break; + case SWR_MULTISAMPLE_2X: return BECentroidBarycentricChooser::GetFunc(remainingArgs...); break; + case SWR_MULTISAMPLE_4X: return BECentroidBarycentricChooser::GetFunc(remainingArgs...); break; + case SWR_MULTISAMPLE_8X: return BECentroidBarycentricChooser::GetFunc(remainingArgs...); break; + case SWR_MULTISAMPLE_16X: return BECentroidBarycentricChooser::GetFunc(remainingArgs...); break; + default: + SWR_ASSERT(0 && "Invalid sample count\n"); + return nullptr; + break; + } + } + + // Recursively parse args + template + static PFN_CALC_CENTROID_BARYCENTRICS GetFunc(uint32_t tArg, TArgsT... remainingArgs) + { + if(tArg > 0) + { + return BECentroidBarycentricChooser::GetFunc(remainingArgs...); + } + + return BECentroidBarycentricChooser::GetFunc(remainingArgs...); + } +}; + +// Recursive template used to auto-nest conditionals. Converts dynamic enum function +// arguments to static template arguments. +template +struct BEChooser +{ + // Last Arg Terminator + static PFN_BACKEND_FUNC GetFunc(SWR_BACKEND_FUNCS tArg) + { + switch(tArg) + { + case SWR_BACKEND_SINGLE_SAMPLE: return BackendSingleSample; break; + case SWR_BACKEND_MSAA_PIXEL_RATE: return BackendPixelRate; break; + case SWR_BACKEND_MSAA_SAMPLE_RATE: return BackendSampleRate; break; + default: + SWR_ASSERT(0 && "Invalid backend func\n"); + return nullptr; + break; + } + } + + + // Recursively parse args + template + static PFN_BACKEND_FUNC GetFunc(SWR_MULTISAMPLE_COUNT tArg, TArgsT... remainingArgs) + { + switch(tArg) + { + case SWR_MULTISAMPLE_1X: return BEChooser::GetFunc(remainingArgs...); break; + case SWR_MULTISAMPLE_2X: return BEChooser::GetFunc(remainingArgs...); break; + case SWR_MULTISAMPLE_4X: return BEChooser::GetFunc(remainingArgs...); break; + case SWR_MULTISAMPLE_8X: return BEChooser::GetFunc(remainingArgs...); break; + case SWR_MULTISAMPLE_16X: return BEChooser::GetFunc(remainingArgs...); break; + default: + SWR_ASSERT(0 && "Invalid sample count\n"); + return nullptr; + break; + } + } + + // Recursively parse args + template + static PFN_BACKEND_FUNC GetFunc(uint32_t tArg, TArgsT... remainingArgs) + { + if(tArg > 0) + { + return BEChooser::GetFunc(remainingArgs...); + } + + return BEChooser::GetFunc(remainingArgs...); + } +}; + +template +void InitBackendOMFuncTable(PFN_OUTPUT_MERGER (&table)[numRenderTargets][numSampleRates]) +{ + for(uint32_t rtNum = SWR_ATTACHMENT_COLOR0; rtNum < numRenderTargets; rtNum++) + { + for(uint32_t sampleCount = SWR_MULTISAMPLE_1X; sampleCount < numSampleRates; sampleCount++) + { + table[rtNum][sampleCount] = + OMChooser<>::GetFunc((SWR_RENDERTARGET_ATTACHMENT)rtNum, (SWR_MULTISAMPLE_COUNT)sampleCount); + } + } +} + +template +void InitBackendBarycentricsTables(PFN_CALC_PIXEL_BARYCENTRICS (&pixelTable)[2], + PFN_CALC_SAMPLE_BARYCENTRICS (&sampleTable)[2], + PFN_CALC_CENTROID_BARYCENTRICS (¢roidTable)[numSampleRates][2][2][2]) +{ + pixelTable[0] = CalcPixelBarycentrics<0>; + pixelTable[1] = CalcPixelBarycentrics<1>; + + sampleTable[0] = CalcSampleBarycentrics<0>; + sampleTable[1] = CalcSampleBarycentrics<1>; + + for(uint32_t sampleCount = SWR_MULTISAMPLE_1X; sampleCount < numSampleRates; sampleCount++) + { + for(uint32_t baryMask = 0; baryMask < 2; baryMask++) + { + for(uint32_t patternNum = 0; patternNum < 2; patternNum++) + { + for(uint32_t forcedSampleEnable = 0; forcedSampleEnable < 2; forcedSampleEnable++) + { + centroidTable[sampleCount][baryMask][patternNum][forcedSampleEnable]= + BECentroidBarycentricChooser<>::GetFunc((SWR_MULTISAMPLE_COUNT)sampleCount, baryMask, patternNum, forcedSampleEnable); + } + } + } + } +} + +void InitBackendSampleFuncTable(PFN_BACKEND_FUNC (&table)[2][2]) +{ + gBackendSingleSample[0][0] = BEChooser<>::GetFunc(SWR_MULTISAMPLE_1X, SWR_MSAA_STANDARD_PATTERN, SWR_INPUT_COVERAGE_NONE, 0, 0, (SWR_BACKEND_FUNCS)SWR_BACKEND_SINGLE_SAMPLE); + gBackendSingleSample[0][1] = BEChooser<>::GetFunc(SWR_MULTISAMPLE_1X, SWR_MSAA_STANDARD_PATTERN, SWR_INPUT_COVERAGE_NONE, 1, 0, (SWR_BACKEND_FUNCS)SWR_BACKEND_SINGLE_SAMPLE); + gBackendSingleSample[1][0] = BEChooser<>::GetFunc(SWR_MULTISAMPLE_1X, SWR_MSAA_STANDARD_PATTERN, SWR_INPUT_COVERAGE_NORMAL, 0, 0, (SWR_BACKEND_FUNCS)SWR_BACKEND_SINGLE_SAMPLE); + gBackendSingleSample[1][1] = BEChooser<>::GetFunc(SWR_MULTISAMPLE_1X, SWR_MSAA_STANDARD_PATTERN, SWR_INPUT_COVERAGE_NORMAL, 1, 0, (SWR_BACKEND_FUNCS)SWR_BACKEND_SINGLE_SAMPLE); +} + +template +void InitBackendPixelFuncTable(PFN_BACKEND_FUNC (&table)[numSampleRates][numSamplePatterns][numCoverageModes][2][2]) +{ + for(uint32_t sampleCount = SWR_MULTISAMPLE_1X; sampleCount < numSampleRates; sampleCount++) + { + for(uint32_t samplePattern = SWR_MSAA_CENTER_PATTERN; samplePattern < numSamplePatterns; samplePattern++) + { + for(uint32_t inputCoverage = SWR_INPUT_COVERAGE_NONE; inputCoverage < numCoverageModes; inputCoverage++) + { + for(uint32_t isCentroid = 0; isCentroid < 2; isCentroid++) + { + table[sampleCount][samplePattern][inputCoverage][isCentroid][0] = + BEChooser<>::GetFunc((SWR_MULTISAMPLE_COUNT)sampleCount, samplePattern, inputCoverage, isCentroid, 0, (SWR_BACKEND_FUNCS)SWR_BACKEND_MSAA_PIXEL_RATE); + table[sampleCount][samplePattern][inputCoverage][isCentroid][1] = + BEChooser<>::GetFunc((SWR_MULTISAMPLE_COUNT)sampleCount, samplePattern, inputCoverage, isCentroid, 1, (SWR_BACKEND_FUNCS)SWR_BACKEND_MSAA_PIXEL_RATE); + } + } + } + } +} + +template +void InitBackendSampleFuncTable(PFN_BACKEND_FUNC (&table)[numSampleRates][numCoverageModes][2]) +{ + for(uint32_t sampleCount = SWR_MULTISAMPLE_1X; sampleCount < numSampleRates; sampleCount++) + { + for(uint32_t inputCoverage = SWR_INPUT_COVERAGE_NONE; inputCoverage < numCoverageModes; inputCoverage++) + { + table[sampleCount][inputCoverage][0] = + BEChooser<>::GetFunc((SWR_MULTISAMPLE_COUNT)sampleCount, SWR_MSAA_STANDARD_PATTERN, inputCoverage, 0, 0, (SWR_BACKEND_FUNCS)SWR_BACKEND_MSAA_SAMPLE_RATE); + table[sampleCount][inputCoverage][1] = + BEChooser<>::GetFunc((SWR_MULTISAMPLE_COUNT)sampleCount, SWR_MSAA_STANDARD_PATTERN, inputCoverage, 1, 0, (SWR_BACKEND_FUNCS)SWR_BACKEND_MSAA_SAMPLE_RATE); + } + } +} + +void InitBackendFuncTables() +{ + InitBackendSampleFuncTable(gBackendSingleSample); + InitBackendPixelFuncTable<(SWR_MULTISAMPLE_COUNT)SWR_MULTISAMPLE_TYPE_MAX, SWR_MSAA_SAMPLE_PATTERN_MAX, SWR_INPUT_COVERAGE_MAX>(gBackendPixelRateTable); + InitBackendSampleFuncTable(gBackendSampleRateTable); + InitBackendOMFuncTable(gBackendOutputMergerTable); + InitBackendBarycentricsTables<(SWR_MULTISAMPLE_COUNT)(SWR_MULTISAMPLE_TYPE_MAX)>(gPixelBarycentricTable, gSampleBarycentricTable, gCentroidBarycentricTable); + + gBackendNullPs[SWR_MULTISAMPLE_1X] = &BackendNullPS < SWR_MULTISAMPLE_1X > ; + gBackendNullPs[SWR_MULTISAMPLE_2X] = &BackendNullPS < SWR_MULTISAMPLE_2X > ; + gBackendNullPs[SWR_MULTISAMPLE_4X] = &BackendNullPS < SWR_MULTISAMPLE_4X > ; + gBackendNullPs[SWR_MULTISAMPLE_8X] = &BackendNullPS < SWR_MULTISAMPLE_8X > ; + gBackendNullPs[SWR_MULTISAMPLE_16X] = &BackendNullPS < SWR_MULTISAMPLE_16X > ; +} diff --git a/src/gallium/drivers/swr/rasterizer/core/backend.h b/src/gallium/drivers/swr/rasterizer/core/backend.h new file mode 100644 index 00000000000..53089e5047b --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/backend.h @@ -0,0 +1,59 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file backend.h +* +* @brief Backend handles rasterization, pixel shading and output merger +* operations. +* +******************************************************************************/ +#pragma once + +#include "common/os.h" +#include "core/context.h" + +void ProcessComputeBE(DRAW_CONTEXT* pDC, uint32_t workerId, uint32_t threadGroupId); +void ProcessSyncBE(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pUserData); +void ProcessQueryStatsBE(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pUserData); +void ProcessClearBE(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pUserData); +void ProcessStoreTileBE(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pData); +void ProcessInvalidateTilesBE(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pData); +void BackendNullPS(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t x, uint32_t y, SWR_TRIANGLE_DESC &work, RenderOutputBuffers &renderBuffers); +void InitClearTilesTable(); + +enum SWR_BACKEND_FUNCS +{ + SWR_BACKEND_SINGLE_SAMPLE, + SWR_BACKEND_MSAA_PIXEL_RATE, + SWR_BACKEND_MSAA_SAMPLE_RATE, + SWR_BACKEND_FUNCS_MAX, +}; +void InitBackendFuncTables(); + +extern PFN_BACKEND_FUNC gBackendNullPs[SWR_MULTISAMPLE_TYPE_MAX]; +extern PFN_BACKEND_FUNC gBackendSingleSample[2][2]; +extern PFN_BACKEND_FUNC gBackendPixelRateTable[SWR_MULTISAMPLE_TYPE_MAX][SWR_MSAA_SAMPLE_PATTERN_MAX][SWR_INPUT_COVERAGE_MAX][2][2]; +extern PFN_BACKEND_FUNC gBackendSampleRateTable[SWR_MULTISAMPLE_TYPE_MAX][SWR_INPUT_COVERAGE_MAX][2]; +extern PFN_OUTPUT_MERGER gBackendOutputMergerTable[SWR_NUM_RENDERTARGETS+1][SWR_MULTISAMPLE_TYPE_MAX]; +extern PFN_CALC_PIXEL_BARYCENTRICS gPixelBarycentricTable[2]; +extern PFN_CALC_SAMPLE_BARYCENTRICS gSampleBarycentricTable[2]; +extern PFN_CALC_CENTROID_BARYCENTRICS gCentroidBarycentricTable[SWR_MULTISAMPLE_TYPE_MAX][2][2][2]; diff --git a/src/gallium/drivers/swr/rasterizer/core/blend.h b/src/gallium/drivers/swr/rasterizer/core/blend.h new file mode 100644 index 00000000000..626c237d75b --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/blend.h @@ -0,0 +1,318 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file blend.cpp +* +* @brief Implementation for blending operations. +* +******************************************************************************/ +#include "state.h" + +template +INLINE +void GenerateBlendFactor(SWR_BLEND_FACTOR func, simdvector &constantColor, simdvector &src, simdvector &src1, simdvector &dst, simdvector &out) +{ + simdvector result; + + switch (func) + { + case BLENDFACTOR_ZERO: + result.x = _simd_setzero_ps(); + result.y = _simd_setzero_ps(); + result.z = _simd_setzero_ps(); + result.w = _simd_setzero_ps(); + break; + + case BLENDFACTOR_ONE: + result.x = _simd_set1_ps(1.0); + result.y = _simd_set1_ps(1.0); + result.z = _simd_set1_ps(1.0); + result.w = _simd_set1_ps(1.0); + break; + + case BLENDFACTOR_SRC_COLOR: + result = src; + break; + + case BLENDFACTOR_DST_COLOR: + result = dst; + break; + + case BLENDFACTOR_INV_SRC_COLOR: + result.x = _simd_sub_ps(_simd_set1_ps(1.0), src.x); + result.y = _simd_sub_ps(_simd_set1_ps(1.0), src.y); + result.z = _simd_sub_ps(_simd_set1_ps(1.0), src.z); + result.w = _simd_sub_ps(_simd_set1_ps(1.0), src.w); + break; + + case BLENDFACTOR_INV_DST_COLOR: + result.x = _simd_sub_ps(_simd_set1_ps(1.0), dst.x); + result.y = _simd_sub_ps(_simd_set1_ps(1.0), dst.y); + result.z = _simd_sub_ps(_simd_set1_ps(1.0), dst.z); + result.w = _simd_sub_ps(_simd_set1_ps(1.0), dst.w); + break; + + case BLENDFACTOR_SRC_ALPHA: result.x = src.w; + result.y = src.w; + result.z = src.w; + result.w = src.w; + break; + + case BLENDFACTOR_INV_SRC_ALPHA: + { + simdscalar oneMinusSrcA = _simd_sub_ps(_simd_set1_ps(1.0), src.w); + result.x = oneMinusSrcA; + result.y = oneMinusSrcA; + result.z = oneMinusSrcA; + result.w = oneMinusSrcA; + break; + } + + case BLENDFACTOR_DST_ALPHA: result.x = dst.w; + result.y = dst.w; + result.z = dst.w; + result.w = dst.w; + break; + + case BLENDFACTOR_INV_DST_ALPHA: + { + simdscalar oneMinusDstA = _simd_sub_ps(_simd_set1_ps(1.0), dst.w); + result.x = oneMinusDstA; + result.y = oneMinusDstA; + result.z = oneMinusDstA; + result.w = oneMinusDstA; + break; + } + + case BLENDFACTOR_SRC_ALPHA_SATURATE: + { + simdscalar sat = _simd_min_ps(src.w, _simd_sub_ps(_simd_set1_ps(1.0), dst.w)); + result.x = sat; + result.y = sat; + result.z = sat; + result.w = _simd_set1_ps(1.0); + break; + } + + case BLENDFACTOR_CONST_COLOR: + result.x = constantColor[0]; + result.y = constantColor[1]; + result.z = constantColor[2]; + result.w = constantColor[3]; + break; + + case BLENDFACTOR_CONST_ALPHA: + result.x = result.y = result.z = result.w = constantColor[3]; + break; + + case BLENDFACTOR_INV_CONST_COLOR: + { + result.x = _simd_sub_ps(_simd_set1_ps(1.0f), constantColor[0]); + result.y = _simd_sub_ps(_simd_set1_ps(1.0f), constantColor[1]); + result.z = _simd_sub_ps(_simd_set1_ps(1.0f), constantColor[2]); + result.w = _simd_sub_ps(_simd_set1_ps(1.0f), constantColor[3]); + break; + } + + case BLENDFACTOR_INV_CONST_ALPHA: + { + result.x = result.y = result.z = result.w = _simd_sub_ps(_simd_set1_ps(1.0f), constantColor[3]); + break; + } + + case BLENDFACTOR_SRC1_COLOR: + result.x = src1.x; + result.y = src1.y; + result.z = src1.z; + result.w = src1.w; + break; + + case BLENDFACTOR_SRC1_ALPHA: + result.x = result.y = result.z = result.w = src1.w; + break; + + case BLENDFACTOR_INV_SRC1_COLOR: + result.x = _simd_sub_ps(_simd_set1_ps(1.0f), src1.x); + result.y = _simd_sub_ps(_simd_set1_ps(1.0f), src1.y); + result.z = _simd_sub_ps(_simd_set1_ps(1.0f), src1.z); + result.w = _simd_sub_ps(_simd_set1_ps(1.0f), src1.w); + break; + + case BLENDFACTOR_INV_SRC1_ALPHA: + result.x = result.y = result.z = result.w = _simd_sub_ps(_simd_set1_ps(1.0f), src1.w); + break; + + default: SWR_ASSERT(false, "Unimplemented blend factor: %d", func); + } + + if (Color) + { + out.x = result.x; + out.y = result.y; + out.z = result.z; + } + if (Alpha) + { + out.w = result.w; + } + +} + +template +INLINE void BlendFunc(SWR_BLEND_OP blendOp, simdvector &src, simdvector &srcFactor, simdvector &dst, simdvector &dstFactor, simdvector &out) +{ + simdvector result; + + switch (blendOp) + { + case BLENDOP_ADD: + result.x = _simd_fmadd_ps(srcFactor.x, src.x, _simd_mul_ps(dstFactor.x, dst.x)); + result.y = _simd_fmadd_ps(srcFactor.y, src.y, _simd_mul_ps(dstFactor.y, dst.y)); + result.z = _simd_fmadd_ps(srcFactor.z, src.z, _simd_mul_ps(dstFactor.z, dst.z)); + result.w = _simd_fmadd_ps(srcFactor.w, src.w, _simd_mul_ps(dstFactor.w, dst.w)); + break; + + case BLENDOP_SUBTRACT: + result.x = _simd_fmsub_ps(srcFactor.x, src.x, _simd_mul_ps(dstFactor.x, dst.x)); + result.y = _simd_fmsub_ps(srcFactor.y, src.y, _simd_mul_ps(dstFactor.y, dst.y)); + result.z = _simd_fmsub_ps(srcFactor.z, src.z, _simd_mul_ps(dstFactor.z, dst.z)); + result.w = _simd_fmsub_ps(srcFactor.w, src.w, _simd_mul_ps(dstFactor.w, dst.w)); + break; + + case BLENDOP_REVSUBTRACT: + result.x = _simd_fmsub_ps(dstFactor.x, dst.x, _simd_mul_ps(srcFactor.x, src.x)); + result.y = _simd_fmsub_ps(dstFactor.y, dst.y, _simd_mul_ps(srcFactor.y, src.y)); + result.z = _simd_fmsub_ps(dstFactor.z, dst.z, _simd_mul_ps(srcFactor.z, src.z)); + result.w = _simd_fmsub_ps(dstFactor.w, dst.w, _simd_mul_ps(srcFactor.w, src.w)); + break; + + case BLENDOP_MIN: + result.x = _simd_min_ps(_simd_mul_ps(srcFactor.x, src.x), _simd_mul_ps(dstFactor.x, dst.x)); + result.y = _simd_min_ps(_simd_mul_ps(srcFactor.y, src.y), _simd_mul_ps(dstFactor.y, dst.y)); + result.z = _simd_min_ps(_simd_mul_ps(srcFactor.z, src.z), _simd_mul_ps(dstFactor.z, dst.z)); + result.w = _simd_min_ps(_simd_mul_ps(srcFactor.w, src.w), _simd_mul_ps(dstFactor.w, dst.w)); + break; + + case BLENDOP_MAX: + result.x = _simd_max_ps(_simd_mul_ps(srcFactor.x, src.x), _simd_mul_ps(dstFactor.x, dst.x)); + result.y = _simd_max_ps(_simd_mul_ps(srcFactor.y, src.y), _simd_mul_ps(dstFactor.y, dst.y)); + result.z = _simd_max_ps(_simd_mul_ps(srcFactor.z, src.z), _simd_mul_ps(dstFactor.z, dst.z)); + result.w = _simd_max_ps(_simd_mul_ps(srcFactor.w, src.w), _simd_mul_ps(dstFactor.w, dst.w)); + break; + + default: + SWR_ASSERT(false, "Unimplemented blend function: %d", blendOp); + } + + if (Color) + { + out.x = result.x; + out.y = result.y; + out.z = result.z; + } + if (Alpha) + { + out.w = result.w; + } +} + +template +INLINE void Clamp(simdvector &src) +{ + switch (type) + { + case SWR_TYPE_FLOAT: + break; + + case SWR_TYPE_UNORM: + src.x = _simd_max_ps(src.x, _simd_setzero_ps()); + src.x = _simd_min_ps(src.x, _simd_set1_ps(1.0f)); + + src.y = _simd_max_ps(src.y, _simd_setzero_ps()); + src.y = _simd_min_ps(src.y, _simd_set1_ps(1.0f)); + + src.z = _simd_max_ps(src.z, _simd_setzero_ps()); + src.z = _simd_min_ps(src.z, _simd_set1_ps(1.0f)); + + src.w = _simd_max_ps(src.w, _simd_setzero_ps()); + src.w = _simd_min_ps(src.w, _simd_set1_ps(1.0f)); + break; + + case SWR_TYPE_SNORM: + src.x = _simd_max_ps(src.x, _simd_set1_ps(-1.0f)); + src.x = _simd_min_ps(src.x, _simd_set1_ps(1.0f)); + + src.y = _simd_max_ps(src.y, _simd_set1_ps(-1.0f)); + src.y = _simd_min_ps(src.y, _simd_set1_ps(1.0f)); + + src.z = _simd_max_ps(src.z, _simd_set1_ps(-1.0f)); + src.z = _simd_min_ps(src.z, _simd_set1_ps(1.0f)); + + src.w = _simd_max_ps(src.w, _simd_set1_ps(-1.0f)); + src.w = _simd_min_ps(src.w, _simd_set1_ps(1.0f)); + break; + + default: + SWR_ASSERT(false, "Unimplemented clamp: %d", type); + break; + } +} + +template +void Blend(const SWR_BLEND_STATE *pBlendState, const SWR_RENDER_TARGET_BLEND_STATE *pState, simdvector &src, simdvector& src1, BYTE *pDst, simdvector &result) +{ + // load render target + simdvector dst; + LoadSOA(pDst, dst); + + simdvector constColor; + constColor.x = _simd_broadcast_ss(&pBlendState->constantColor[0]); + constColor.y = _simd_broadcast_ss(&pBlendState->constantColor[1]); + constColor.z = _simd_broadcast_ss(&pBlendState->constantColor[2]); + constColor.w = _simd_broadcast_ss(&pBlendState->constantColor[3]); + + // clamp src/dst/constant + Clamp(src); + Clamp(src1); + Clamp(dst); + Clamp(constColor); + + simdvector srcFactor, dstFactor; + if (pBlendState->independentAlphaBlendEnable) + { + GenerateBlendFactor((SWR_BLEND_FACTOR)pState->sourceBlendFactor, constColor, src, src1, dst, srcFactor); + GenerateBlendFactor((SWR_BLEND_FACTOR)pState->sourceAlphaBlendFactor, constColor, src, src1, dst, srcFactor); + + GenerateBlendFactor((SWR_BLEND_FACTOR)pState->destBlendFactor, constColor, src, src1, dst, dstFactor); + GenerateBlendFactor((SWR_BLEND_FACTOR)pState->destAlphaBlendFactor, constColor, src, src1, dst, dstFactor); + + BlendFunc((SWR_BLEND_OP)pState->colorBlendFunc, src, srcFactor, dst, dstFactor, result); + BlendFunc((SWR_BLEND_OP)pState->alphaBlendFunc, src, srcFactor, dst, dstFactor, result); + } + else + { + GenerateBlendFactor((SWR_BLEND_FACTOR)pState->sourceBlendFactor, constColor, src, src1, dst, srcFactor); + GenerateBlendFactor((SWR_BLEND_FACTOR)pState->destBlendFactor, constColor, src, src1, dst, dstFactor); + + BlendFunc((SWR_BLEND_OP)pState->colorBlendFunc, src, srcFactor, dst, dstFactor, result); + } +} diff --git a/src/gallium/drivers/swr/rasterizer/core/clip.cpp b/src/gallium/drivers/swr/rasterizer/core/clip.cpp new file mode 100644 index 00000000000..ce27bf71d3c --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/clip.cpp @@ -0,0 +1,201 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file clip.cpp +* +* @brief Implementation for clipping +* +******************************************************************************/ + +#include + +#include "common/os.h" +#include "core/clip.h" + +float ComputeInterpFactor(float boundaryCoord0, float boundaryCoord1) +{ + return (boundaryCoord0 / (boundaryCoord0 - boundaryCoord1)); +} + +template +inline void intersect( + int s, // index to first edge vertex v0 in pInPts. + int p, // index to second edge vertex v1 in pInPts. + const float *pInPts, // array of all the input positions. + const float *pInAttribs, // array of all attributes for all vertex. All the attributes for each vertex is contiguous. + int numInAttribs, // number of attributes per vertex. + int i, // output index. + float *pOutPts, // array of output positions. We'll write our new intersection point at i*4. + float *pOutAttribs) // array of output attributes. We'll write our new attributes at i*numInAttribs. +{ + float t; + + // Find the parameter of the intersection. + // t = (v1.w - v1.x) / ((v2.x - v1.x) - (v2.w - v1.w)) for x = w (RIGHT) plane, etc. + const float *v1 = &pInPts[s*4]; + const float *v2 = &pInPts[p*4]; + + switch (ClippingPlane) + { + case FRUSTUM_LEFT: t = ComputeInterpFactor(v1[3] + v1[0], v2[3] + v2[0]); break; + case FRUSTUM_RIGHT: t = ComputeInterpFactor(v1[3] - v1[0], v2[3] - v2[0]); break; + case FRUSTUM_TOP: t = ComputeInterpFactor(v1[3] + v1[1], v2[3] + v2[1]); break; + case FRUSTUM_BOTTOM: t = ComputeInterpFactor(v1[3] - v1[1], v2[3] - v2[1]); break; + case FRUSTUM_NEAR: t = ComputeInterpFactor(v1[2], v2[2]); break; + case FRUSTUM_FAR: t = ComputeInterpFactor(v1[3] - v1[2], v2[3] - v2[2]); break; + default: SWR_ASSERT(false, "invalid clipping plane: %d", ClippingPlane); + }; + + + const float *a1 = &pInAttribs[s*numInAttribs]; + const float *a2 = &pInAttribs[p*numInAttribs]; + + float *pOutP = &pOutPts[i*4]; + float *pOutA = &pOutAttribs[i*numInAttribs]; + + // Interpolate new position. + for(int j = 0; j < 4; ++j) + { + pOutP[j] = v1[j] + (v2[j]-v1[j])*t; + } + + // Interpolate Attributes + for(int attr = 0; attr < numInAttribs; ++attr) + { + pOutA[attr] = a1[attr] + (a2[attr]-a1[attr])*t; + } +} + + +// Checks whether vertex v lies inside clipping plane +// in homogenous coords check -w < {x,y,z} < w; +// +template +inline int inside(const float v[4]) +{ + switch (ClippingPlane) + { + case FRUSTUM_LEFT : return (v[0]>=-v[3]); + case FRUSTUM_RIGHT : return (v[0]<= v[3]); + case FRUSTUM_TOP : return (v[1]>=-v[3]); + case FRUSTUM_BOTTOM : return (v[1]<= v[3]); + case FRUSTUM_NEAR : return (v[2]>=0.0f); + case FRUSTUM_FAR : return (v[2]<= v[3]); + default: + SWR_ASSERT(false, "invalid clipping plane: %d", ClippingPlane); + return 0; + } +} + + +// Clips a polygon in homogenous coordinates to a particular clipping plane. +// Takes in vertices of the polygon (InPts) and the clipping plane +// Puts the vertices of the clipped polygon in OutPts +// Returns number of points in clipped polygon +// +template +int ClipTriToPlane( const float *pInPts, int numInPts, + const float *pInAttribs, int numInAttribs, + float *pOutPts, float *pOutAttribs) +{ + int i=0; // index number of OutPts, # of vertices in OutPts = i div 4; + + for (int j = 0; j < numInPts; ++j) + { + int s = j; + int p = (j + 1) % numInPts; + + int s_in = inside(&pInPts[s*4]); + int p_in = inside(&pInPts[p*4]); + + // test if vertex is to be added to output vertices + if (s_in != p_in) // edge crosses clipping plane + { + // find point of intersection + intersect(s, p, pInPts, pInAttribs, numInAttribs, i, pOutPts, pOutAttribs); + i++; + } + if (p_in) // 2nd vertex is inside clipping volume, add it to output + { + // Copy 2nd vertex position of edge over to output. + for(int k = 0; k < 4; ++k) + { + pOutPts[i*4 + k] = pInPts[p*4 + k]; + } + // Copy 2nd vertex attributes of edge over to output. + for(int attr = 0; attr < numInAttribs; ++attr) + { + pOutAttribs[i*numInAttribs+attr] = pInAttribs[p*numInAttribs+attr]; + } + i++; + } + // edge does not cross clipping plane and vertex outside clipping volume + // => do not add vertex + } + return i; +} + + + +void Clip(const float *pTriangle, const float *pAttribs, int numAttribs, float *pOutTriangles, int *numVerts, float *pOutAttribs) +{ + // temp storage to hold at least 6 sets of vertices, the max number that can be created during clipping + OSALIGN(float, 16) tempPts[6 * 4]; + OSALIGN(float, 16) tempAttribs[6 * KNOB_NUM_ATTRIBUTES * 4]; + + // we opt to clip to viewport frustum to produce smaller triangles for rasterization precision + int NumOutPts = ClipTriToPlane(pTriangle, 3, pAttribs, numAttribs, tempPts, tempAttribs); + NumOutPts = ClipTriToPlane(tempPts, NumOutPts, tempAttribs, numAttribs, pOutTriangles, pOutAttribs); + NumOutPts = ClipTriToPlane(pOutTriangles, NumOutPts, pOutAttribs, numAttribs, tempPts, tempAttribs); + NumOutPts = ClipTriToPlane(tempPts, NumOutPts, tempAttribs, numAttribs, pOutTriangles, pOutAttribs); + NumOutPts = ClipTriToPlane(pOutTriangles, NumOutPts, pOutAttribs, numAttribs, tempPts, tempAttribs); + NumOutPts = ClipTriToPlane(tempPts, NumOutPts, tempAttribs, numAttribs, pOutTriangles, pOutAttribs); + + SWR_ASSERT(NumOutPts <= 6); + + *numVerts = NumOutPts; + return; +} + +void ClipTriangles(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[], uint32_t primMask, simdscalari primId) +{ + RDTSC_START(FEClipTriangles); + Clipper<3> clipper(workerId, pDC); + clipper.ExecuteStage(pa, prims, primMask, primId); + RDTSC_STOP(FEClipTriangles, 1, 0); +} + +void ClipLines(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[], uint32_t primMask, simdscalari primId) +{ + RDTSC_START(FEClipLines); + Clipper<2> clipper(workerId, pDC); + clipper.ExecuteStage(pa, prims, primMask, primId); + RDTSC_STOP(FEClipLines, 1, 0); +} +void ClipPoints(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[], uint32_t primMask, simdscalari primId) +{ + RDTSC_START(FEClipPoints); + Clipper<1> clipper(workerId, pDC); + clipper.ExecuteStage(pa, prims, primMask, primId); + RDTSC_STOP(FEClipPoints, 1, 0); +} + diff --git a/src/gallium/drivers/swr/rasterizer/core/clip.h b/src/gallium/drivers/swr/rasterizer/core/clip.h new file mode 100644 index 00000000000..49494a4e374 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/clip.h @@ -0,0 +1,868 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file clip.h +* +* @brief Definitions for clipping +* +******************************************************************************/ +#pragma once + +#include "common/simdintrin.h" +#include "core/context.h" +#include "core/pa.h" +#include "rdtsc_core.h" + +enum SWR_CLIPCODES +{ + // Shift clip codes out of the mantissa to prevent denormalized values when used in float compare. + // Guardband is able to use a single high-bit with 4 separate LSBs, because it computes a union, rather than intersection, of clipcodes. +#define CLIPCODE_SHIFT 23 + FRUSTUM_LEFT = (0x01 << CLIPCODE_SHIFT), + FRUSTUM_TOP = (0x02 << CLIPCODE_SHIFT), + FRUSTUM_RIGHT = (0x04 << CLIPCODE_SHIFT), + FRUSTUM_BOTTOM = (0x08 << CLIPCODE_SHIFT), + + FRUSTUM_NEAR = (0x10 << CLIPCODE_SHIFT), + FRUSTUM_FAR = (0x20 << CLIPCODE_SHIFT), + + NEGW = (0x40 << CLIPCODE_SHIFT), + + GUARDBAND_LEFT = (0x80 << CLIPCODE_SHIFT | 0x1), + GUARDBAND_TOP = (0x80 << CLIPCODE_SHIFT | 0x2), + GUARDBAND_RIGHT = (0x80 << CLIPCODE_SHIFT | 0x4), + GUARDBAND_BOTTOM = (0x80 << CLIPCODE_SHIFT | 0x8) +}; + +#define FRUSTUM_CLIP_MASK (FRUSTUM_LEFT|FRUSTUM_TOP|FRUSTUM_RIGHT|FRUSTUM_BOTTOM|FRUSTUM_NEAR|FRUSTUM_FAR) +#define GUARDBAND_CLIP_MASK (FRUSTUM_NEAR|FRUSTUM_FAR|GUARDBAND_LEFT|GUARDBAND_TOP|GUARDBAND_RIGHT|GUARDBAND_BOTTOM|NEGW) + +void Clip(const float *pTriangle, const float *pAttribs, int numAttribs, float *pOutTriangles, + int *numVerts, float *pOutAttribs); + +INLINE +void ComputeClipCodes(DRIVER_TYPE type, const API_STATE& state, const simdvector& vertex, simdscalar& clipCodes) +{ + clipCodes = _simd_setzero_ps(); + + // -w + simdscalar vNegW = _simd_mul_ps(vertex.w, _simd_set1_ps(-1.0f)); + + // FRUSTUM_LEFT + simdscalar vRes = _simd_cmplt_ps(vertex.x, vNegW); + clipCodes = _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_LEFT))); + + // FRUSTUM_TOP + vRes = _simd_cmplt_ps(vertex.y, vNegW); + clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_TOP)))); + + // FRUSTUM_RIGHT + vRes = _simd_cmpgt_ps(vertex.x, vertex.w); + clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_RIGHT)))); + + // FRUSTUM_BOTTOM + vRes = _simd_cmpgt_ps(vertex.y, vertex.w); + clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_BOTTOM)))); + + if (state.rastState.depthClipEnable) + { + // FRUSTUM_NEAR + // DX clips depth [0..w], GL clips [-w..w] + if (type == DX) + { + vRes = _simd_cmplt_ps(vertex.z, _simd_setzero_ps()); + } + else + { + vRes = _simd_cmplt_ps(vertex.z, vNegW); + } + clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_NEAR)))); + + // FRUSTUM_FAR + vRes = _simd_cmpgt_ps(vertex.z, vertex.w); + clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(FRUSTUM_FAR)))); + } + + // NEGW + vRes = _simd_cmple_ps(vertex.w, _simd_setzero_ps()); + clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(NEGW)))); + + // GUARDBAND_LEFT + simdscalar gbMult = _simd_mul_ps(vNegW, _simd_set1_ps(state.gbState.left)); + vRes = _simd_cmplt_ps(vertex.x, gbMult); + clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(GUARDBAND_LEFT)))); + + // GUARDBAND_TOP + gbMult = _simd_mul_ps(vNegW, _simd_set1_ps(state.gbState.top)); + vRes = _simd_cmplt_ps(vertex.y, gbMult); + clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(GUARDBAND_TOP)))); + + // GUARDBAND_RIGHT + gbMult = _simd_mul_ps(vertex.w, _simd_set1_ps(state.gbState.right)); + vRes = _simd_cmpgt_ps(vertex.x, gbMult); + clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(GUARDBAND_RIGHT)))); + + // GUARDBAND_BOTTOM + gbMult = _simd_mul_ps(vertex.w, _simd_set1_ps(state.gbState.bottom)); + vRes = _simd_cmpgt_ps(vertex.y, gbMult); + clipCodes = _simd_or_ps(clipCodes, _simd_and_ps(vRes, _simd_castsi_ps(_simd_set1_epi32(GUARDBAND_BOTTOM)))); +} + +template +class Clipper +{ +public: + Clipper(uint32_t in_workerId, DRAW_CONTEXT* in_pDC) : + workerId(in_workerId), driverType(in_pDC->pContext->driverType), pDC(in_pDC), state(GetApiState(in_pDC)) + { + static_assert(NumVertsPerPrim >= 1 && NumVertsPerPrim <= 3, "Invalid NumVertsPerPrim"); + } + + void ComputeClipCodes(simdvector vertex[]) + { + for (uint32_t i = 0; i < NumVertsPerPrim; ++i) + { + ::ComputeClipCodes(this->driverType, this->state, vertex[i], this->clipCodes[i]); + } + } + + simdscalar ComputeClipCodeIntersection() + { + simdscalar result = this->clipCodes[0]; + for (uint32_t i = 1; i < NumVertsPerPrim; ++i) + { + result = _simd_and_ps(result, this->clipCodes[i]); + } + return result; + } + + simdscalar ComputeClipCodeUnion() + { + simdscalar result = this->clipCodes[0]; + for (uint32_t i = 1; i < NumVertsPerPrim; ++i) + { + result = _simd_or_ps(result, this->clipCodes[i]); + } + return result; + } + + int ComputeNegWMask() + { + simdscalar clipCodeUnion = ComputeClipCodeUnion(); + clipCodeUnion = _simd_and_ps(clipCodeUnion, _simd_castsi_ps(_simd_set1_epi32(NEGW))); + return _simd_movemask_ps(_simd_cmpneq_ps(clipCodeUnion, _simd_setzero_ps())); + } + + int ComputeClipMask() + { + simdscalar clipUnion = ComputeClipCodeUnion(); + clipUnion = _simd_and_ps(clipUnion, _simd_castsi_ps(_simd_set1_epi32(GUARDBAND_CLIP_MASK))); + return _simd_movemask_ps(_simd_cmpneq_ps(clipUnion, _simd_setzero_ps())); + } + + // clipper is responsible for culling any prims with NAN coordinates + int ComputeNaNMask(simdvector prim[]) + { + simdscalar vNanMask = _simd_setzero_ps(); + for (uint32_t e = 0; e < NumVertsPerPrim; ++e) + { + simdscalar vNan01 = _simd_cmp_ps(prim[e].v[0], prim[e].v[1], _CMP_UNORD_Q); + vNanMask = _simd_or_ps(vNanMask, vNan01); + simdscalar vNan23 = _simd_cmp_ps(prim[e].v[2], prim[e].v[3], _CMP_UNORD_Q); + vNanMask = _simd_or_ps(vNanMask, vNan23); + } + + return _simd_movemask_ps(vNanMask); + } + + int ComputeUserClipCullMask(PA_STATE& pa, simdvector prim[]) + { + uint8_t cullMask = this->state.rastState.cullDistanceMask; + simdscalar vClipCullMask = _simd_setzero_ps(); + DWORD index; + + simdvector vClipCullDistLo[3]; + simdvector vClipCullDistHi[3]; + + pa.Assemble(VERTEX_CLIPCULL_DIST_LO_SLOT, vClipCullDistLo); + pa.Assemble(VERTEX_CLIPCULL_DIST_HI_SLOT, vClipCullDistHi); + while (_BitScanForward(&index, cullMask)) + { + cullMask &= ~(1 << index); + uint32_t slot = index >> 2; + uint32_t component = index & 0x3; + + simdscalar vCullMaskElem = _simd_set1_ps(-1.0f); + for (uint32_t e = 0; e < NumVertsPerPrim; ++e) + { + simdscalar vCullComp; + if (slot == 0) + { + vCullComp = vClipCullDistLo[e][component]; + } + else + { + vCullComp = vClipCullDistHi[e][component]; + } + + // cull if cull distance < 0 || NAN + simdscalar vCull = _simd_cmp_ps(_mm256_setzero_ps(), vCullComp, _CMP_NLE_UQ); + vCullMaskElem = _simd_and_ps(vCullMaskElem, vCull); + } + vClipCullMask = _simd_or_ps(vClipCullMask, vCullMaskElem); + } + + // clipper should also discard any primitive with NAN clip distance + uint8_t clipMask = this->state.rastState.clipDistanceMask; + while (_BitScanForward(&index, clipMask)) + { + clipMask &= ~(1 << index); + uint32_t slot = index >> 2; + uint32_t component = index & 0x3; + + for (uint32_t e = 0; e < NumVertsPerPrim; ++e) + { + simdscalar vClipComp; + if (slot == 0) + { + vClipComp = vClipCullDistLo[e][component]; + } + else + { + vClipComp = vClipCullDistHi[e][component]; + } + + simdscalar vClip = _simd_cmp_ps(vClipComp, vClipComp, _CMP_UNORD_Q); + vClipCullMask = _simd_or_ps(vClipCullMask, vClip); + } + } + + return _simd_movemask_ps(vClipCullMask); + } + + // clip a single primitive + int ClipScalar(PA_STATE& pa, uint32_t primIndex, float* pOutPos, float* pOutAttribs) + { + OSALIGN(float, 16) inVerts[3 * 4]; + OSALIGN(float, 16) inAttribs[3 * KNOB_NUM_ATTRIBUTES * 4]; + + // transpose primitive position + __m128 verts[3]; + pa.AssembleSingle(VERTEX_POSITION_SLOT, primIndex, verts); + _mm_store_ps(&inVerts[0], verts[0]); + _mm_store_ps(&inVerts[4], verts[1]); + _mm_store_ps(&inVerts[8], verts[2]); + + // transpose attribs + uint32_t numScalarAttribs = this->state.linkageCount * 4; + + int idx = 0; + DWORD slot = 0; + uint32_t mapIdx = 0; + uint32_t tmpLinkage = uint32_t(this->state.linkageMask); + while (_BitScanForward(&slot, tmpLinkage)) + { + tmpLinkage &= ~(1 << slot); + // Compute absolute attrib slot in vertex array + uint32_t inputSlot = VERTEX_ATTRIB_START_SLOT + this->state.linkageMap[mapIdx++]; + __m128 attrib[3]; // triangle attribs (always 4 wide) + pa.AssembleSingle(inputSlot, primIndex, attrib); + _mm_store_ps(&inAttribs[idx], attrib[0]); + _mm_store_ps(&inAttribs[idx + numScalarAttribs], attrib[1]); + _mm_store_ps(&inAttribs[idx + numScalarAttribs * 2], attrib[2]); + idx += 4; + } + + int numVerts; + Clip(inVerts, inAttribs, numScalarAttribs, pOutPos, &numVerts, pOutAttribs); + + return numVerts; + } + + // clip SIMD primitives + void ClipSimd(const simdscalar& vPrimMask, const simdscalar& vClipMask, PA_STATE& pa, const simdscalari& vPrimId) + { + // input/output vertex store for clipper + simdvertex vertices[7]; // maximum 7 verts generated per triangle + + LONG constantInterpMask = this->state.backendState.constantInterpolationMask; + uint32_t provokingVertex = 0; + if(pa.binTopology == TOP_TRIANGLE_FAN) + { + provokingVertex = this->state.frontendState.provokingVertex.triFan; + } + ///@todo: line topology for wireframe? + + // assemble pos + simdvector tmpVector[NumVertsPerPrim]; + pa.Assemble(VERTEX_POSITION_SLOT, tmpVector); + for (uint32_t i = 0; i < NumVertsPerPrim; ++i) + { + vertices[i].attrib[VERTEX_POSITION_SLOT] = tmpVector[i]; + } + + // assemble attribs + DWORD slot = 0; + uint32_t mapIdx = 0; + uint32_t tmpLinkage = this->state.linkageMask; + + int32_t maxSlot = -1; + while (_BitScanForward(&slot, tmpLinkage)) + { + tmpLinkage &= ~(1 << slot); + // Compute absolute attrib slot in vertex array + uint32_t mapSlot = this->state.linkageMap[mapIdx++]; + maxSlot = std::max(maxSlot, mapSlot); + uint32_t inputSlot = VERTEX_ATTRIB_START_SLOT + mapSlot; + + pa.Assemble(inputSlot, tmpVector); + + // if constant interpolation enabled for this attribute, assign the provoking + // vertex values to all edges + if (_bittest(&constantInterpMask, slot)) + { + for (uint32_t i = 0; i < NumVertsPerPrim; ++i) + { + vertices[i].attrib[inputSlot] = tmpVector[provokingVertex]; + } + } + else + { + for (uint32_t i = 0; i < NumVertsPerPrim; ++i) + { + vertices[i].attrib[inputSlot] = tmpVector[i]; + } + } + } + + uint32_t numAttribs = maxSlot + 1; + + simdscalari vNumClippedVerts = ClipPrims((float*)&vertices[0], vPrimMask, vClipMask, numAttribs); + + // set up new PA for binning clipped primitives + PFN_PROCESS_PRIMS pfnBinFunc = nullptr; + PRIMITIVE_TOPOLOGY clipTopology = TOP_UNKNOWN; + if (NumVertsPerPrim == 3) + { + pfnBinFunc = BinTriangles; + clipTopology = TOP_TRIANGLE_FAN; + + // so that the binner knows to bloat wide points later + if (pa.binTopology == TOP_POINT_LIST) + clipTopology = TOP_POINT_LIST; + } + else if (NumVertsPerPrim == 2) + { + pfnBinFunc = BinLines; + clipTopology = TOP_LINE_LIST; + } + else + { + SWR_ASSERT(0 && "Unexpected points in clipper."); + } + + + uint32_t* pVertexCount = (uint32_t*)&vNumClippedVerts; + uint32_t* pPrimitiveId = (uint32_t*)&vPrimId; + + const simdscalari vOffsets = _mm256_set_epi32( + 0 * sizeof(simdvertex), // unused lane + 6 * sizeof(simdvertex), + 5 * sizeof(simdvertex), + 4 * sizeof(simdvertex), + 3 * sizeof(simdvertex), + 2 * sizeof(simdvertex), + 1 * sizeof(simdvertex), + 0 * sizeof(simdvertex)); + + // only need to gather 7 verts + // @todo dynamic mask based on actual # of verts generated per lane + const simdscalar vMask = _mm256_set_ps(0, -1, -1, -1, -1, -1, -1, -1); + + uint32_t numClippedPrims = 0; + for (uint32_t inputPrim = 0; inputPrim < pa.NumPrims(); ++inputPrim) + { + uint32_t numEmittedVerts = pVertexCount[inputPrim]; + if (numEmittedVerts < NumVertsPerPrim) + { + continue; + } + SWR_ASSERT(numEmittedVerts <= 7, "Unexpected vertex count from clipper."); + + uint32_t numEmittedPrims = GetNumPrims(clipTopology, numEmittedVerts); + numClippedPrims += numEmittedPrims; + + // tranpose clipper output so that each lane's vertices are in SIMD order + // set aside space for 2 vertices, as the PA will try to read up to 16 verts + // for triangle fan + simdvertex transposedPrims[2]; + + // transpose pos + uint8_t* pBase = (uint8_t*)(&vertices[0].attrib[VERTEX_POSITION_SLOT]) + sizeof(float) * inputPrim; + for (uint32_t c = 0; c < 4; ++c) + { + transposedPrims[0].attrib[VERTEX_POSITION_SLOT][c] = _simd_mask_i32gather_ps(_mm256_undefined_ps(), (const float*)pBase, vOffsets, vMask, 1); + pBase += sizeof(simdscalar); + } + + // transpose attribs + pBase = (uint8_t*)(&vertices[0].attrib[VERTEX_ATTRIB_START_SLOT]) + sizeof(float) * inputPrim; + for (uint32_t attrib = 0; attrib < numAttribs; ++attrib) + { + uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + attrib; + for (uint32_t c = 0; c < 4; ++c) + { + transposedPrims[0].attrib[attribSlot][c] = _simd_mask_i32gather_ps(_mm256_undefined_ps(), (const float*)pBase, vOffsets, vMask, 1); + pBase += sizeof(simdscalar); + } + } + + PA_STATE_OPT clipPa(this->pDC, numEmittedPrims, (uint8_t*)&transposedPrims[0], numEmittedVerts, true, clipTopology); + + while (clipPa.GetNextStreamOutput()) + { + do + { + simdvector attrib[NumVertsPerPrim]; + bool assemble = clipPa.Assemble(VERTEX_POSITION_SLOT, attrib); + if (assemble) + { + static const uint32_t primMaskMap[] = { 0x0, 0x1, 0x3, 0x7, 0xf, 0x1f, 0x3f, 0x7f, 0xff }; + pfnBinFunc(this->pDC, clipPa, this->workerId, attrib, primMaskMap[numEmittedPrims], _simd_set1_epi32(pPrimitiveId[inputPrim])); + } + } while (clipPa.NextPrim()); + } + } + + // update global pipeline stat + SWR_CONTEXT* pContext = this->pDC->pContext; + UPDATE_STAT(CPrimitives, numClippedPrims); + } + + // execute the clipper stage + void ExecuteStage(PA_STATE& pa, simdvector prim[], uint32_t primMask, simdscalari primId) + { + // set up binner based on PA state + PFN_PROCESS_PRIMS pfnBinner; + switch (pa.binTopology) + { + case TOP_POINT_LIST: + pfnBinner = BinPoints; + break; + case TOP_LINE_LIST: + case TOP_LINE_STRIP: + case TOP_LINE_LOOP: + case TOP_LINE_LIST_ADJ: + case TOP_LISTSTRIP_ADJ: + pfnBinner = BinLines; + break; + default: + pfnBinner = BinTriangles; + break; + }; + + // update clipper invocations pipeline stat + SWR_CONTEXT* pContext = this->pDC->pContext; + uint32_t numInvoc = _mm_popcnt_u32(primMask); + UPDATE_STAT(CInvocations, numInvoc); + + ComputeClipCodes(prim); + + // cull prims with NAN coords + primMask &= ~ComputeNaNMask(prim); + + // user cull distance cull + if (this->state.rastState.cullDistanceMask) + { + primMask &= ~ComputeUserClipCullMask(pa, prim); + } + + // cull prims outside view frustum + simdscalar clipIntersection = ComputeClipCodeIntersection(); + int validMask = primMask & _simd_movemask_ps(_simd_cmpeq_ps(clipIntersection, _simd_setzero_ps())); + + // skip clipping for points + uint32_t clipMask = 0; + if (NumVertsPerPrim != 1) + { + clipMask = primMask & ComputeClipMask(); + } + + if (clipMask) + { + RDTSC_START(FEGuardbandClip); + // we have to clip tris, execute the clipper, which will also + // call the binner + ClipSimd(vMask(primMask), vMask(clipMask), pa, primId); + RDTSC_STOP(FEGuardbandClip, 1, 0); + } + else if (validMask) + { + // update CPrimitives pipeline state + SWR_CONTEXT* pContext = this->pDC->pContext; + UPDATE_STAT(CPrimitives, _mm_popcnt_u32(validMask)); + + // forward valid prims directly to binner + pfnBinner(this->pDC, pa, this->workerId, prim, validMask, primId); + } + } + +private: + inline simdscalar ComputeInterpFactor(simdscalar boundaryCoord0, simdscalar boundaryCoord1) + { + return _simd_div_ps(boundaryCoord0, _simd_sub_ps(boundaryCoord0, boundaryCoord1)); + } + + inline simdscalari ComputeOffsets(uint32_t attrib, simdscalari vIndices, uint32_t component) + { + const uint32_t simdVertexStride = sizeof(simdvertex); + const uint32_t componentStride = sizeof(simdscalar); + const uint32_t attribStride = sizeof(simdvector); + const __m256i vElemOffset = _mm256_set_epi32(7 * sizeof(float), 6 * sizeof(float), 5 * sizeof(float), 4 * sizeof(float), + 3 * sizeof(float), 2 * sizeof(float), 1 * sizeof(float), 0 * sizeof(float)); + + // step to the simdvertex + simdscalari vOffsets = _simd_mullo_epi32(vIndices, _simd_set1_epi32(simdVertexStride)); + + // step to the attribute and component + vOffsets = _simd_add_epi32(vOffsets, _simd_set1_epi32(attribStride * attrib + componentStride * component)); + + // step to the lane + vOffsets = _simd_add_epi32(vOffsets, vElemOffset); + + return vOffsets; + } + + // gathers a single component for a given attribute for each SIMD lane + inline simdscalar GatherComponent(const float* pBuffer, uint32_t attrib, simdscalar vMask, simdscalari vIndices, uint32_t component) + { + simdscalari vOffsets = ComputeOffsets(attrib, vIndices, component); + simdscalar vSrc = _mm256_undefined_ps(); + return _simd_mask_i32gather_ps(vSrc, pBuffer, vOffsets, vMask, 1); + } + + inline void ScatterComponent(const float* pBuffer, uint32_t attrib, simdscalar vMask, simdscalari vIndices, uint32_t component, simdscalar vSrc) + { + simdscalari vOffsets = ComputeOffsets(attrib, vIndices, component); + + uint32_t* pOffsets = (uint32_t*)&vOffsets; + float* pSrc = (float*)&vSrc; + uint32_t mask = _simd_movemask_ps(vMask); + DWORD lane; + while (_BitScanForward(&lane, mask)) + { + mask &= ~(1 << lane); + uint8_t* pBuf = (uint8_t*)pBuffer + pOffsets[lane]; + *(float*)pBuf = pSrc[lane]; + } + } + + template + inline void intersect( + const simdscalar& vActiveMask, // active lanes to operate on + const simdscalari& s, // index to first edge vertex v0 in pInPts. + const simdscalari& p, // index to second edge vertex v1 in pInPts. + const simdvector& v1, // vertex 0 position + const simdvector& v2, // vertex 1 position + simdscalari& outIndex, // output index. + const float *pInVerts, // array of all the input positions. + uint32_t numInAttribs, // number of attributes per vertex. + float *pOutVerts) // array of output positions. We'll write our new intersection point at i*4. + { + // compute interpolation factor + simdscalar t; + switch (ClippingPlane) + { + case FRUSTUM_LEFT: t = ComputeInterpFactor(_simd_add_ps(v1[3], v1[0]), _simd_add_ps(v2[3], v2[0])); break; + case FRUSTUM_RIGHT: t = ComputeInterpFactor(_simd_sub_ps(v1[3], v1[0]), _simd_sub_ps(v2[3], v2[0])); break; + case FRUSTUM_TOP: t = ComputeInterpFactor(_simd_add_ps(v1[3], v1[1]), _simd_add_ps(v2[3], v2[1])); break; + case FRUSTUM_BOTTOM: t = ComputeInterpFactor(_simd_sub_ps(v1[3], v1[1]), _simd_sub_ps(v2[3], v2[1])); break; + case FRUSTUM_NEAR: + // DX Znear plane is 0, GL is -w + if (this->driverType == DX) + { + t = ComputeInterpFactor(v1[2], v2[2]); + } + else + { + t = ComputeInterpFactor(_simd_add_ps(v1[3], v1[2]), _simd_add_ps(v2[3], v2[2])); + } + break; + case FRUSTUM_FAR: t = ComputeInterpFactor(_simd_sub_ps(v1[3], v1[2]), _simd_sub_ps(v2[3], v2[2])); break; + default: SWR_ASSERT(false, "invalid clipping plane: %d", ClippingPlane); + }; + + // interpolate position and store + for (uint32_t c = 0; c < 4; ++c) + { + simdscalar vOutPos = _simd_fmadd_ps(_simd_sub_ps(v2[c], v1[c]), t, v1[c]); + ScatterComponent(pOutVerts, VERTEX_POSITION_SLOT, vActiveMask, outIndex, c, vOutPos); + } + + // interpolate attributes and store + for (uint32_t a = 0; a < numInAttribs; ++a) + { + uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a; + for (uint32_t c = 0; c < 4; ++c) + { + simdscalar vAttrib0 = GatherComponent(pInVerts, attribSlot, vActiveMask, s, c); + simdscalar vAttrib1 = GatherComponent(pInVerts, attribSlot, vActiveMask, p, c); + simdscalar vOutAttrib = _simd_fmadd_ps(_simd_sub_ps(vAttrib1, vAttrib0), t, vAttrib0); + ScatterComponent(pOutVerts, attribSlot, vActiveMask, outIndex, c, vOutAttrib); + } + } + } + + template + inline simdscalar inside(const simdvector& v) + { + switch (ClippingPlane) + { + case FRUSTUM_LEFT: return _simd_cmpge_ps(v[0], _simd_mul_ps(v[3], _simd_set1_ps(-1.0f))); + case FRUSTUM_RIGHT: return _simd_cmple_ps(v[0], v[3]); + case FRUSTUM_TOP: return _simd_cmpge_ps(v[1], _simd_mul_ps(v[3], _simd_set1_ps(-1.0f))); + case FRUSTUM_BOTTOM: return _simd_cmple_ps(v[1], v[3]); + case FRUSTUM_NEAR: return _simd_cmpge_ps(v[2], this->driverType == DX ? _simd_setzero_ps() : _simd_mul_ps(v[3], _simd_set1_ps(-1.0f))); + case FRUSTUM_FAR: return _simd_cmple_ps(v[2], v[3]); + default: + SWR_ASSERT(false, "invalid clipping plane: %d", ClippingPlane); + return _simd_setzero_ps(); + } + } + + template + simdscalari ClipTriToPlane(const float* pInVerts, const simdscalari& vNumInPts, uint32_t numInAttribs, float* pOutVerts) + { + simdscalari vCurIndex = _simd_setzero_si(); + simdscalari vOutIndex = _simd_setzero_si(); + simdscalar vActiveMask = _simd_castsi_ps(_simd_cmplt_epi32(vCurIndex, vNumInPts)); + + while (!_simd_testz_ps(vActiveMask, vActiveMask)) // loop until activeMask is empty + { + simdscalari s = vCurIndex; + simdscalari p = _simd_add_epi32(s, _simd_set1_epi32(1)); + simdscalari underFlowMask = _simd_cmpgt_epi32(vNumInPts, p); + p = _simd_castps_si(_simd_blendv_ps(_simd_setzero_ps(), _simd_castsi_ps(p), _simd_castsi_ps(underFlowMask))); + + // gather position + simdvector vInPos0, vInPos1; + for (uint32_t c = 0; c < 4; ++c) + { + vInPos0[c] = GatherComponent(pInVerts, VERTEX_POSITION_SLOT, vActiveMask, s, c); + vInPos1[c] = GatherComponent(pInVerts, VERTEX_POSITION_SLOT, vActiveMask, p, c); + } + + // compute inside mask + simdscalar s_in = inside(vInPos0); + simdscalar p_in = inside(vInPos1); + + // compute intersection mask (s_in != p_in) + simdscalar intersectMask = _simd_xor_ps(s_in, p_in); + intersectMask = _simd_and_ps(intersectMask, vActiveMask); + + // store s if inside + s_in = _simd_and_ps(s_in, vActiveMask); + if (!_simd_testz_ps(s_in, s_in)) + { + // store position + for (uint32_t c = 0; c < 4; ++c) + { + ScatterComponent(pOutVerts, VERTEX_POSITION_SLOT, s_in, vOutIndex, c, vInPos0[c]); + } + + // store attribs + for (uint32_t a = 0; a < numInAttribs; ++a) + { + uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a; + for (uint32_t c = 0; c < 4; ++c) + { + simdscalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c); + ScatterComponent(pOutVerts, attribSlot, s_in, vOutIndex, c, vAttrib); + } + } + + // increment outIndex + vOutIndex = _simd_blendv_epi32(vOutIndex, _simd_add_epi32(vOutIndex, _simd_set1_epi32(1)), s_in); + } + + // compute and store intersection + if (!_simd_testz_ps(intersectMask, intersectMask)) + { + intersect(intersectMask, s, p, vInPos0, vInPos1, vOutIndex, pInVerts, numInAttribs, pOutVerts); + + // increment outIndex for active lanes + vOutIndex = _simd_blendv_epi32(vOutIndex, _simd_add_epi32(vOutIndex, _simd_set1_epi32(1)), intersectMask); + } + + // increment loop index and update active mask + vCurIndex = _simd_add_epi32(vCurIndex, _simd_set1_epi32(1)); + vActiveMask = _simd_castsi_ps(_simd_cmplt_epi32(vCurIndex, vNumInPts)); + } + + return vOutIndex; + } + + template + simdscalari ClipLineToPlane(const float* pInVerts, const simdscalari& vNumInPts, uint32_t numInAttribs, float* pOutVerts) + { + simdscalari vCurIndex = _simd_setzero_si(); + simdscalari vOutIndex = _simd_setzero_si(); + simdscalar vActiveMask = _simd_castsi_ps(_simd_cmplt_epi32(vCurIndex, vNumInPts)); + + if (!_simd_testz_ps(vActiveMask, vActiveMask)) + { + simdscalari s = vCurIndex; + simdscalari p = _simd_add_epi32(s, _simd_set1_epi32(1)); + + // gather position + simdvector vInPos0, vInPos1; + for (uint32_t c = 0; c < 4; ++c) + { + vInPos0[c] = GatherComponent(pInVerts, VERTEX_POSITION_SLOT, vActiveMask, s, c); + vInPos1[c] = GatherComponent(pInVerts, VERTEX_POSITION_SLOT, vActiveMask, p, c); + } + + // compute inside mask + simdscalar s_in = inside(vInPos0); + simdscalar p_in = inside(vInPos1); + + // compute intersection mask (s_in != p_in) + simdscalar intersectMask = _simd_xor_ps(s_in, p_in); + intersectMask = _simd_and_ps(intersectMask, vActiveMask); + + // store s if inside + s_in = _simd_and_ps(s_in, vActiveMask); + if (!_simd_testz_ps(s_in, s_in)) + { + for (uint32_t c = 0; c < 4; ++c) + { + ScatterComponent(pOutVerts, VERTEX_POSITION_SLOT, s_in, vOutIndex, c, vInPos0[c]); + } + + // interpolate attributes and store + for (uint32_t a = 0; a < numInAttribs; ++a) + { + uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a; + for (uint32_t c = 0; c < 4; ++c) + { + simdscalar vAttrib = GatherComponent(pInVerts, attribSlot, s_in, s, c); + ScatterComponent(pOutVerts, attribSlot, s_in, vOutIndex, c, vAttrib); + } + } + + // increment outIndex + vOutIndex = _simd_blendv_epi32(vOutIndex, _simd_add_epi32(vOutIndex, _simd_set1_epi32(1)), s_in); + } + + // compute and store intersection + if (!_simd_testz_ps(intersectMask, intersectMask)) + { + intersect(intersectMask, s, p, vInPos0, vInPos1, vOutIndex, pInVerts, numInAttribs, pOutVerts); + + // increment outIndex for active lanes + vOutIndex = _simd_blendv_epi32(vOutIndex, _simd_add_epi32(vOutIndex, _simd_set1_epi32(1)), intersectMask); + } + + // store p if inside + p_in = _simd_and_ps(p_in, vActiveMask); + if (!_simd_testz_ps(p_in, p_in)) + { + for (uint32_t c = 0; c < 4; ++c) + { + ScatterComponent(pOutVerts, VERTEX_POSITION_SLOT, p_in, vOutIndex, c, vInPos1[c]); + } + + // interpolate attributes and store + for (uint32_t a = 0; a < numInAttribs; ++a) + { + uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + a; + for (uint32_t c = 0; c < 4; ++c) + { + simdscalar vAttrib = GatherComponent(pInVerts, attribSlot, p_in, p, c); + ScatterComponent(pOutVerts, attribSlot, p_in, vOutIndex, c, vAttrib); + } + } + + // increment outIndex + vOutIndex = _simd_blendv_epi32(vOutIndex, _simd_add_epi32(vOutIndex, _simd_set1_epi32(1)), p_in); + } + } + + return vOutIndex; + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Vertical clipper. Clips SIMD primitives at a time + /// @param pVertices - pointer to vertices in SOA form. Clipper will read input and write results to this buffer + /// @param vPrimMask - mask of valid input primitives, including non-clipped prims + /// @param numAttribs - number of valid input attribs, including position + simdscalari ClipPrims(float* pVertices, const simdscalar& vPrimMask, const simdscalar& vClipMask, int numAttribs) + { + // temp storage + simdvertex tempVertices[7]; + float* pTempVerts = (float*)&tempVertices[0]; + + // zero out num input verts for non-active lanes + simdscalari vNumInPts = _simd_set1_epi32(NumVertsPerPrim); + vNumInPts = _simd_blendv_epi32(_simd_setzero_si(), vNumInPts, vClipMask); + + // clip prims to frustum + simdscalari vNumOutPts; + if (NumVertsPerPrim == 3) + { + vNumOutPts = ClipTriToPlane(pVertices, vNumInPts, numAttribs, pTempVerts); + vNumOutPts = ClipTriToPlane(pTempVerts, vNumOutPts, numAttribs, pVertices); + vNumOutPts = ClipTriToPlane(pVertices, vNumOutPts, numAttribs, pTempVerts); + vNumOutPts = ClipTriToPlane(pTempVerts, vNumOutPts, numAttribs, pVertices); + vNumOutPts = ClipTriToPlane(pVertices, vNumOutPts, numAttribs, pTempVerts); + vNumOutPts = ClipTriToPlane(pTempVerts, vNumOutPts, numAttribs, pVertices); + } + else + { + SWR_ASSERT(NumVertsPerPrim == 2); + vNumOutPts = ClipLineToPlane(pVertices, vNumInPts, numAttribs, pTempVerts); + vNumOutPts = ClipLineToPlane(pTempVerts, vNumOutPts, numAttribs, pVertices); + vNumOutPts = ClipLineToPlane(pVertices, vNumOutPts, numAttribs, pTempVerts); + vNumOutPts = ClipLineToPlane(pTempVerts, vNumOutPts, numAttribs, pVertices); + vNumOutPts = ClipLineToPlane(pVertices, vNumOutPts, numAttribs, pTempVerts); + vNumOutPts = ClipLineToPlane(pTempVerts, vNumOutPts, numAttribs, pVertices); + } + + // restore num verts for non-clipped, active lanes + simdscalar vNonClippedMask = _simd_andnot_ps(vClipMask, vPrimMask); + vNumOutPts = _simd_blendv_epi32(vNumOutPts, _simd_set1_epi32(NumVertsPerPrim), vNonClippedMask); + + return vNumOutPts; + } + + const uint32_t workerId; + const DRIVER_TYPE driverType; + DRAW_CONTEXT* pDC; + const API_STATE& state; + simdscalar clipCodes[NumVertsPerPrim]; +}; + + +// pipeline stage functions +void ClipTriangles(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[], uint32_t primMask, simdscalari primId); +void ClipLines(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[], uint32_t primMask, simdscalari primId); +void ClipPoints(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[], uint32_t primMask, simdscalari primId); diff --git a/src/gallium/drivers/swr/rasterizer/core/context.h b/src/gallium/drivers/swr/rasterizer/core/context.h new file mode 100644 index 00000000000..4a214aff1c8 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/context.h @@ -0,0 +1,495 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file context.h +* +* @brief Definitions for SWR_CONTEXT and DRAW_CONTEXT +* The SWR_CONTEXT is our global context and contains the DC ring, +* thread state, etc. +* +* The DRAW_CONTEXT contains all state associated with a draw operation. +* +******************************************************************************/ +#pragma once + +#include +#include + +#include "core/api.h" +#include "core/utils.h" +#include "core/arena.h" +#include "core/fifo.hpp" +#include "core/knobs.h" +#include "common/simdintrin.h" +#include "core/threads.h" + +// x.8 fixed point precision values +#define FIXED_POINT_SHIFT 8 +#define FIXED_POINT_SCALE 256 + +// x.16 fixed point precision values +#define FIXED_POINT16_SHIFT 16 +#define FIXED_POINT16_SCALE 65536 + +struct SWR_CONTEXT; +struct DRAW_CONTEXT; + +struct TRI_FLAGS +{ + uint32_t frontFacing : 1; + uint32_t yMajor : 1; + uint32_t coverageMask : (SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM); + uint32_t reserved : 32 - 1 - 1 - (SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM); + float pointSize; + uint32_t primID; + uint32_t renderTargetArrayIndex; +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_TRIANGLE_DESC +///////////////////////////////////////////////////////////////////////// +struct SWR_TRIANGLE_DESC +{ + float I[3]; + float J[3]; + float Z[3]; + float OneOverW[3]; + float recipDet; + + float *pRecipW; + float *pAttribs; + float *pPerspAttribs; + float *pSamplePos; + float *pUserClipBuffer; + + uint64_t coverageMask[SWR_MAX_NUM_MULTISAMPLES]; + + TRI_FLAGS triFlags; +}; + +struct TRIANGLE_WORK_DESC +{ + float *pTriBuffer; + float *pAttribs; + float *pUserClipBuffer; + uint32_t numAttribs; + TRI_FLAGS triFlags; +}; + +union CLEAR_FLAGS +{ + struct + { + uint32_t mask : 3; + }; + uint32_t bits; +}; + +struct CLEAR_DESC +{ + CLEAR_FLAGS flags; + float clearRTColor[4]; // RGBA_32F + float clearDepth; // [0..1] + BYTE clearStencil; +}; + +struct INVALIDATE_TILES_DESC +{ + uint32_t attachmentMask; +}; + +struct SYNC_DESC +{ + PFN_CALLBACK_FUNC pfnCallbackFunc; + uint64_t userData; + uint64_t userData2; + uint64_t userData3; +}; + +struct QUERY_DESC +{ + SWR_STATS* pStats; +}; + +struct STORE_TILES_DESC +{ + SWR_RENDERTARGET_ATTACHMENT attachment; + SWR_TILE_STATE postStoreTileState; +}; + +struct COMPUTE_DESC +{ + uint32_t threadGroupCountX; + uint32_t threadGroupCountY; + uint32_t threadGroupCountZ; +}; + +typedef void(*PFN_WORK_FUNC)(DRAW_CONTEXT* pDC, uint32_t workerId, uint32_t macroTile, void* pDesc); + +enum WORK_TYPE +{ + SYNC, + DRAW, + CLEAR, + INVALIDATETILES, + STORETILES, + QUERYSTATS, +}; + +struct BE_WORK +{ + WORK_TYPE type; + PFN_WORK_FUNC pfnWork; + union + { + SYNC_DESC sync; + TRIANGLE_WORK_DESC tri; + CLEAR_DESC clear; + INVALIDATE_TILES_DESC invalidateTiles; + STORE_TILES_DESC storeTiles; + QUERY_DESC queryStats; + } desc; +}; + +struct DRAW_WORK +{ + DRAW_CONTEXT* pDC; + union + { + uint32_t numIndices; // DrawIndexed: Number of indices for draw. + uint32_t numVerts; // Draw: Number of verts (triangles, lines, etc) + }; + union + { + const int32_t* pIB; // DrawIndexed: App supplied indices + uint32_t startVertex; // Draw: Starting vertex in VB to render from. + }; + int32_t baseVertex; + uint32_t numInstances; // Number of instances + uint32_t startInstance; // Instance offset + uint32_t startPrimID; // starting primitiveID for this draw batch + uint32_t startVertexID; // starting VertexID for this draw batch (only needed for non-indexed draws) + SWR_FORMAT type; // index buffer type +}; + +typedef void(*PFN_FE_WORK_FUNC)(SWR_CONTEXT* pContext, DRAW_CONTEXT* pDC, uint32_t workerId, void* pDesc); +struct FE_WORK +{ + WORK_TYPE type; + PFN_FE_WORK_FUNC pfnWork; + union + { + SYNC_DESC sync; + DRAW_WORK draw; + CLEAR_DESC clear; + INVALIDATE_TILES_DESC invalidateTiles; + STORE_TILES_DESC storeTiles; + QUERY_DESC queryStats; + } desc; +}; + +struct GUARDBAND +{ + float left, right, top, bottom; +}; + +struct PA_STATE; + +// function signature for pipeline stages that execute after primitive assembly +typedef void(*PFN_PROCESS_PRIMS)(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[], + uint32_t primMask, simdscalari primID); + +OSALIGNLINE(struct) API_STATE +{ + // Vertex Buffers + SWR_VERTEX_BUFFER_STATE vertexBuffers[KNOB_NUM_STREAMS]; + + // Index Buffer + SWR_INDEX_BUFFER_STATE indexBuffer; + + // FS - Fetch Shader State + PFN_FETCH_FUNC pfnFetchFunc; + + // VS - Vertex Shader State + PFN_VERTEX_FUNC pfnVertexFunc; + + // GS - Geometry Shader State + PFN_GS_FUNC pfnGsFunc; + SWR_GS_STATE gsState; + + // CS - Compute Shader + PFN_CS_FUNC pfnCsFunc; + uint32_t totalThreadsInGroup; + + // FE - Frontend State + SWR_FRONTEND_STATE frontendState; + + // SOS - Streamout Shader State + PFN_SO_FUNC pfnSoFunc[MAX_SO_STREAMS]; + + // Streamout state + SWR_STREAMOUT_STATE soState; + mutable SWR_STREAMOUT_BUFFER soBuffer[MAX_SO_STREAMS]; + + // Tessellation State + PFN_HS_FUNC pfnHsFunc; + PFN_DS_FUNC pfnDsFunc; + SWR_TS_STATE tsState; + + // Specifies which VS outputs are sent to PS. + // Does not include position + uint32_t linkageMask; + uint32_t linkageCount; + uint8_t linkageMap[MAX_ATTRIBUTES]; + + // attrib mask, specifies the total set of attributes used + // by the frontend (vs, so, gs) + uint32_t feAttribMask; + + PRIMITIVE_TOPOLOGY topology; + bool forceFront; + + // RS - Rasterizer State + SWR_RASTSTATE rastState; + // floating point multisample offsets + float samplePos[SWR_MAX_NUM_MULTISAMPLES * 2]; + + GUARDBAND gbState; + + SWR_VIEWPORT vp[KNOB_NUM_VIEWPORTS_SCISSORS]; + SWR_VIEWPORT_MATRIX vpMatrix[KNOB_NUM_VIEWPORTS_SCISSORS]; + + BBOX scissorRects[KNOB_NUM_VIEWPORTS_SCISSORS]; + BBOX scissorInFixedPoint; + + // Backend state + SWR_BACKEND_STATE backendState; + + // PS - Pixel shader state + SWR_PS_STATE psState; + + SWR_DEPTH_STENCIL_STATE depthStencilState; + + // OM - Output Merger State + SWR_BLEND_STATE blendState; + PFN_BLEND_JIT_FUNC pfnBlendFunc[SWR_NUM_RENDERTARGETS]; + + // Stats are incremented when this is true. + bool enableStats; + + struct + { + uint32_t colorHottileEnable : 8; + uint32_t depthHottileEnable: 1; + uint32_t stencilHottileEnable : 1; + }; +}; + +class MacroTileMgr; +class DispatchQueue; + +struct RenderOutputBuffers +{ + uint8_t* pColor[SWR_NUM_RENDERTARGETS]; + uint8_t* pDepth; + uint8_t* pStencil; +}; + +// Plane equation A/B/C coeffs used to evaluate I/J barycentric coords +struct BarycentricCoeffs +{ + simdscalar vIa; + simdscalar vIb; + simdscalar vIc; + + simdscalar vJa; + simdscalar vJb; + simdscalar vJc; + + simdscalar vZa; + simdscalar vZb; + simdscalar vZc; + + simdscalar vRecipDet; + + simdscalar vAOneOverW; + simdscalar vBOneOverW; + simdscalar vCOneOverW; +}; + +// pipeline function pointer types +typedef void(*PFN_BACKEND_FUNC)(DRAW_CONTEXT*, uint32_t, uint32_t, uint32_t, SWR_TRIANGLE_DESC&, RenderOutputBuffers&); +typedef void(*PFN_OUTPUT_MERGER)(SWR_PS_CONTEXT &, uint8_t* (&)[SWR_NUM_RENDERTARGETS], uint32_t, const SWR_BLEND_STATE*, + const PFN_BLEND_JIT_FUNC (&)[SWR_NUM_RENDERTARGETS], simdscalar&, simdscalar); +typedef void(*PFN_CALC_PIXEL_BARYCENTRICS)(const BarycentricCoeffs&, SWR_PS_CONTEXT &); +typedef void(*PFN_CALC_SAMPLE_BARYCENTRICS)(const BarycentricCoeffs&, SWR_PS_CONTEXT&); +typedef void(*PFN_CALC_CENTROID_BARYCENTRICS)(const BarycentricCoeffs&, SWR_PS_CONTEXT &, const uint64_t *const, const uint32_t, + const simdscalar, const simdscalar); + +struct BACKEND_FUNCS +{ + PFN_BACKEND_FUNC pfnBackend; + PFN_CALC_PIXEL_BARYCENTRICS pfnCalcPixelBarycentrics; + PFN_CALC_SAMPLE_BARYCENTRICS pfnCalcSampleBarycentrics; + PFN_CALC_CENTROID_BARYCENTRICS pfnCalcCentroidBarycentrics; + PFN_OUTPUT_MERGER pfnOutputMerger; +}; + +// Draw State +struct DRAW_STATE +{ + API_STATE state; + + void* pPrivateState; // Its required the driver sets this up for each draw. + + // pipeline function pointers, filled in by API thread when setting up the draw + BACKEND_FUNCS backendFuncs; + PFN_PROCESS_PRIMS pfnProcessPrims; + + Arena* pArena; // This should only be used by API thread. +}; + +// Draw Context +// The api thread sets up a draw context that exists for the life of the draw. +// This draw context maintains all of the state needed for the draw operation. +struct DRAW_CONTEXT +{ + SWR_CONTEXT *pContext; + + uint64_t drawId; + + bool isCompute; // Is this DC a compute context? + + FE_WORK FeWork; + volatile OSALIGNLINE(uint32_t) FeLock; + volatile OSALIGNLINE(bool) inUse; + volatile OSALIGNLINE(bool) doneFE; // Is FE work done for this draw? + + // Have all worker threads moved past draw in DC ring? + volatile OSALIGNLINE(uint32_t) threadsDoneFE; + volatile OSALIGNLINE(uint32_t) threadsDoneBE; + + uint64_t dependency; + + MacroTileMgr* pTileMgr; + + // The following fields are valid if isCompute is true. + volatile OSALIGNLINE(bool) doneCompute; // Is this dispatch done? (isCompute) + DispatchQueue* pDispatch; // Queue for thread groups. (isCompute) + + DRAW_STATE* pState; + Arena* pArena; + + uint8_t* pSpillFill[KNOB_MAX_NUM_THREADS]; // Scratch space used for spill fills. +}; + +INLINE const API_STATE& GetApiState(const DRAW_CONTEXT* pDC) +{ + SWR_ASSERT(pDC != nullptr); + SWR_ASSERT(pDC->pState != nullptr); + + return pDC->pState->state; +} + +INLINE void* GetPrivateState(const DRAW_CONTEXT* pDC) +{ + SWR_ASSERT(pDC != nullptr); + SWR_ASSERT(pDC->pState != nullptr); + + return pDC->pState->pPrivateState; +} + +class HotTileMgr; + +struct SWR_CONTEXT +{ + // Draw Context Ring + // Each draw needs its own state in order to support mulitple draws in flight across multiple threads. + // We maintain N draw contexts configured as a ring. The size of the ring limits the maximum number + // of draws that can be in flight at any given time. + // + // Description: + // 1. State - When an application first sets state we'll request a new draw context to use. + // a. If there are no available draw contexts then we'll have to wait until one becomes free. + // b. If one is available then set pCurDrawContext to point to it and mark it in use. + // c. All state calls set state on pCurDrawContext. + // 2. Draw - Creates submits a work item that is associated with current draw context. + // a. Set pPrevDrawContext = pCurDrawContext + // b. Set pCurDrawContext to NULL. + // 3. State - When an applications sets state after draw + // a. Same as step 1. + // b. State is copied from prev draw context to current. + DRAW_CONTEXT* dcRing; + + DRAW_CONTEXT *pCurDrawContext; // This points to DC entry in ring for an unsubmitted draw. + DRAW_CONTEXT *pPrevDrawContext; // This points to DC entry for the previous context submitted that we can copy state from. + + // Draw State Ring + // When draw are very large (lots of primitives) then the API thread will break these up. + // These split draws all have identical state. So instead of storing the state directly + // in the Draw Context (DC) we instead store it in a Draw State (DS). This allows multiple DCs + // to reference a single entry in the DS ring. + DRAW_STATE* dsRing; + + uint32_t curStateId; // Current index to the next available entry in the DS ring. + + DRAW_STATE* subCtxSave; // Save area for inactive contexts. + uint32_t curSubCtxId; // Current index for active state subcontext. + uint32_t numSubContexts; // Number of available subcontexts + + uint32_t NumWorkerThreads; + + THREAD_POOL threadPool; // Thread pool associated with this context + + std::condition_variable FifosNotEmpty; + std::mutex WaitLock; + + // Draw Contexts will get a unique drawId generated from this + uint64_t nextDrawId; + + // most recent draw id enqueued by the API thread + // written by api thread, read by multiple workers + OSALIGNLINE(volatile uint64_t) DrawEnqueued; + + DRIVER_TYPE driverType; + + uint32_t privateStateSize; + + HotTileMgr *pHotTileMgr; + + // tile load/store functions, passed in at create context time + PFN_LOAD_TILE pfnLoadTile; + PFN_STORE_TILE pfnStoreTile; + PFN_CLEAR_TILE pfnClearTile; + + // Global Stats + SWR_STATS stats[KNOB_MAX_NUM_THREADS]; + + // Scratch space for workers. + uint8_t* pScratch[KNOB_MAX_NUM_THREADS]; +}; + +void WaitForDependencies(SWR_CONTEXT *pContext, uint64_t drawId); +void WakeAllThreads(SWR_CONTEXT *pContext); + +#define UPDATE_STAT(name, count) if (GetApiState(pDC).enableStats) { pContext->stats[workerId].name += count; } +#define SET_STAT(name, count) if (GetApiState(pDC).enableStats) { pContext->stats[workerId].name = count; } diff --git a/src/gallium/drivers/swr/rasterizer/core/depthstencil.h b/src/gallium/drivers/swr/rasterizer/core/depthstencil.h new file mode 100644 index 00000000000..4f245c8c53e --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/depthstencil.h @@ -0,0 +1,245 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file depthstencil.h +* +* @brief Implements depth/stencil functionality +* +******************************************************************************/ +#pragma once +#include "common/os.h" +#include "format_conversion.h" + +INLINE +void StencilOp(SWR_STENCILOP op, simdscalar mask, simdscalar stencilRefps, simdscalar &stencilps) +{ + simdscalari stencil = _simd_castps_si(stencilps); + + switch (op) + { + case STENCILOP_KEEP: + break; + case STENCILOP_ZERO: + stencilps = _simd_blendv_ps(stencilps, _simd_setzero_ps(), mask); + break; + case STENCILOP_REPLACE: + stencilps = _simd_blendv_ps(stencilps, stencilRefps, mask); + break; + case STENCILOP_INCRSAT: + { + simdscalari stencilincr = _simd_adds_epu8(stencil, _simd_set1_epi32(1)); + stencilps = _simd_blendv_ps(stencilps, _simd_castsi_ps(stencilincr), mask); + break; + } + case STENCILOP_DECRSAT: + { + simdscalari stencildecr = _simd_subs_epu8(stencil, _simd_set1_epi32(1)); + stencilps = _simd_blendv_ps(stencilps, _simd_castsi_ps(stencildecr), mask); + break; + } + case STENCILOP_INCR: + { + simdscalari stencilincr = _simd_add_epi8(stencil, _simd_set1_epi32(1)); + stencilps = _simd_blendv_ps(stencilps, _simd_castsi_ps(stencilincr), mask); + break; + } + case STENCILOP_DECR: + { + simdscalari stencildecr = _simd_add_epi8(stencil, _simd_set1_epi32((-1) & 0xff)); + stencilps = _simd_blendv_ps(stencilps, _simd_castsi_ps(stencildecr), mask); + break; + } + case STENCILOP_INVERT: + { + simdscalar stencilinvert = _simd_andnot_ps(stencilps, _simd_cmpeq_ps(_simd_setzero_ps(), _simd_setzero_ps())); + stencilps = _simd_blendv_ps(stencilps, stencilinvert, mask); + break; + } + default: + break; + } +} + + +INLINE +simdscalar DepthStencilTest(const SWR_VIEWPORT* pViewport, const SWR_DEPTH_STENCIL_STATE* pDSState, + bool frontFacing, simdscalar interpZ, BYTE* pDepthBase, simdscalar coverageMask, BYTE *pStencilBase, + simdscalar* pStencilMask) +{ + static_assert(KNOB_DEPTH_HOT_TILE_FORMAT == R32_FLOAT, "Unsupported depth hot tile format"); + static_assert(KNOB_STENCIL_HOT_TILE_FORMAT == R8_UINT, "Unsupported stencil hot tile format"); + + simdscalar depthResult = _simd_set1_ps(-1.0f); + simdscalar zbuf; + + // clamp Z to viewport [minZ..maxZ] + simdscalar vMinZ = _simd_broadcast_ss(&pViewport->minZ); + simdscalar vMaxZ = _simd_broadcast_ss(&pViewport->maxZ); + interpZ = _simd_min_ps(vMaxZ, _simd_max_ps(vMinZ, interpZ)); + + if (pDSState->depthTestEnable) + { + switch (pDSState->depthTestFunc) + { + case ZFUNC_NEVER: depthResult = _simd_setzero_ps(); break; + case ZFUNC_ALWAYS: break; + default: + zbuf = _simd_load_ps((const float*)pDepthBase); + } + + switch (pDSState->depthTestFunc) + { + case ZFUNC_LE: depthResult = _simd_cmple_ps(interpZ, zbuf); break; + case ZFUNC_LT: depthResult = _simd_cmplt_ps(interpZ, zbuf); break; + case ZFUNC_GT: depthResult = _simd_cmpgt_ps(interpZ, zbuf); break; + case ZFUNC_GE: depthResult = _simd_cmpge_ps(interpZ, zbuf); break; + case ZFUNC_EQ: depthResult = _simd_cmpeq_ps(interpZ, zbuf); break; + case ZFUNC_NE: depthResult = _simd_cmpneq_ps(interpZ, zbuf); break; + } + } + + simdscalar stencilMask = _simd_set1_ps(-1.0f); + + if (pDSState->stencilTestEnable) + { + uint8_t stencilRefValue; + uint32_t stencilTestFunc; + uint8_t stencilTestMask; + if (frontFacing || !pDSState->doubleSidedStencilTestEnable) + { + stencilRefValue = pDSState->stencilRefValue; + stencilTestFunc = pDSState->stencilTestFunc; + stencilTestMask = pDSState->stencilTestMask; + } + else + { + stencilRefValue = pDSState->backfaceStencilRefValue; + stencilTestFunc = pDSState->backfaceStencilTestFunc; + stencilTestMask = pDSState->backfaceStencilTestMask; + } + + simdvector sbuf; + simdscalar stencilWithMask; + simdscalar stencilRef; + switch(stencilTestFunc) + { + case ZFUNC_NEVER: stencilMask = _simd_setzero_ps(); break; + case ZFUNC_ALWAYS: break; + default: + LoadSOA(pStencilBase, sbuf); + + // apply stencil read mask + stencilWithMask = _simd_castsi_ps(_simd_and_si(_simd_castps_si(sbuf.v[0]), _simd_set1_epi32(stencilTestMask))); + + // do stencil compare in float to avoid simd integer emulation in AVX1 + stencilWithMask = _simd_cvtepi32_ps(_simd_castps_si(stencilWithMask)); + + stencilRef = _simd_set1_ps((float)(stencilRefValue & stencilTestMask)); + break; + } + + switch(stencilTestFunc) + { + case ZFUNC_LE: stencilMask = _simd_cmple_ps(stencilRef, stencilWithMask); break; + case ZFUNC_LT: stencilMask = _simd_cmplt_ps(stencilRef, stencilWithMask); break; + case ZFUNC_GT: stencilMask = _simd_cmpgt_ps(stencilRef, stencilWithMask); break; + case ZFUNC_GE: stencilMask = _simd_cmpge_ps(stencilRef, stencilWithMask); break; + case ZFUNC_EQ: stencilMask = _simd_cmpeq_ps(stencilRef, stencilWithMask); break; + case ZFUNC_NE: stencilMask = _simd_cmpneq_ps(stencilRef, stencilWithMask); break; + } + } + + simdscalar depthWriteMask = _simd_and_ps(depthResult, stencilMask); + depthWriteMask = _simd_and_ps(depthWriteMask, coverageMask); + + *pStencilMask = stencilMask; + return depthWriteMask; +} + +INLINE +void DepthStencilWrite(const SWR_VIEWPORT* pViewport, const SWR_DEPTH_STENCIL_STATE* pDSState, + bool frontFacing, simdscalar interpZ, BYTE* pDepthBase, const simdscalar& depthMask, const simdscalar& coverageMask, + BYTE *pStencilBase, const simdscalar& stencilMask) +{ + if (pDSState->depthWriteEnable) + { + // clamp Z to viewport [minZ..maxZ] + simdscalar vMinZ = _simd_broadcast_ss(&pViewport->minZ); + simdscalar vMaxZ = _simd_broadcast_ss(&pViewport->maxZ); + interpZ = _simd_min_ps(vMaxZ, _simd_max_ps(vMinZ, interpZ)); + + simdscalar vMask = _simd_and_ps(depthMask, coverageMask); + _simd_maskstore_ps((float*)pDepthBase, _simd_castps_si(vMask), interpZ); + } + + if (pDSState->stencilWriteEnable) + { + simdvector sbuf; + LoadSOA(pStencilBase, sbuf); + simdscalar stencilbuf = sbuf.v[0]; + + uint8_t stencilRefValue; + uint32_t stencilFailOp; + uint32_t stencilPassDepthPassOp; + uint32_t stencilPassDepthFailOp; + uint8_t stencilWriteMask; + if (frontFacing || !pDSState->doubleSidedStencilTestEnable) + { + stencilRefValue = pDSState->stencilRefValue; + stencilFailOp = pDSState->stencilFailOp; + stencilPassDepthPassOp = pDSState->stencilPassDepthPassOp; + stencilPassDepthFailOp = pDSState->stencilPassDepthFailOp; + stencilWriteMask = pDSState->stencilWriteMask; + } + else + { + stencilRefValue = pDSState->backfaceStencilRefValue; + stencilFailOp = pDSState->backfaceStencilFailOp; + stencilPassDepthPassOp = pDSState->backfaceStencilPassDepthPassOp; + stencilPassDepthFailOp = pDSState->backfaceStencilPassDepthFailOp; + stencilWriteMask = pDSState->backfaceStencilWriteMask; + } + + simdscalar stencilps = stencilbuf; + simdscalar stencilRefps = _simd_castsi_ps(_simd_set1_epi32(stencilRefValue)); + + simdscalar stencilFailMask = _simd_andnot_ps(stencilMask, coverageMask); + simdscalar stencilPassDepthPassMask = _simd_and_ps(stencilMask, depthMask); + simdscalar stencilPassDepthFailMask = _simd_and_ps(stencilMask, _simd_andnot_ps(depthMask, _simd_set1_ps(-1))); + + simdscalar origStencil = stencilps; + + StencilOp((SWR_STENCILOP)stencilFailOp, stencilFailMask, stencilRefps, stencilps); + StencilOp((SWR_STENCILOP)stencilPassDepthFailOp, stencilPassDepthFailMask, stencilRefps, stencilps); + StencilOp((SWR_STENCILOP)stencilPassDepthPassOp, stencilPassDepthPassMask, stencilRefps, stencilps); + + // apply stencil write mask + simdscalari vWriteMask = _simd_set1_epi32(stencilWriteMask); + stencilps = _simd_and_ps(stencilps, _simd_castsi_ps(vWriteMask)); + stencilps = _simd_or_ps(_simd_andnot_ps(_simd_castsi_ps(vWriteMask), origStencil), stencilps); + + simdvector stencilResult; + stencilResult.v[0] = _simd_blendv_ps(origStencil, stencilps, coverageMask); + StoreSOA(stencilResult, pStencilBase); + } + +} diff --git a/src/gallium/drivers/swr/rasterizer/core/fifo.hpp b/src/gallium/drivers/swr/rasterizer/core/fifo.hpp new file mode 100644 index 00000000000..7e556012e6b --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/fifo.hpp @@ -0,0 +1,136 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file fifo.hpp +* +* @brief Definitions for our fifos used for thread communication. +* +******************************************************************************/ +#pragma once + + +#include "common/os.h" +#include "arena.h" + +#include +#include + +template +struct QUEUE +{ + OSALIGNLINE(volatile uint32_t) mLock{ 0 }; + OSALIGNLINE(volatile uint32_t) mNumEntries{ 0 }; + std::vector mBlocks; + T* mCurBlock{ nullptr }; + uint32_t mHead{ 0 }; + uint32_t mTail{ 0 }; + uint32_t mCurBlockIdx{ 0 }; + + // power of 2 + static const uint32_t mBlockSizeShift = 6; + static const uint32_t mBlockSize = 1 << mBlockSizeShift; + + void clear(Arena& arena) + { + mHead = 0; + mTail = 0; + mBlocks.clear(); + T* pNewBlock = (T*)arena.Alloc(sizeof(T)*mBlockSize); + mBlocks.push_back(pNewBlock); + mCurBlock = pNewBlock; + mCurBlockIdx = 0; + + mNumEntries = 0; + _ReadWriteBarrier(); + mLock = 0; + } + + uint32_t getNumQueued() + { + return mNumEntries; + } + + bool tryLock() + { + if (mLock) + { + return false; + } + + // try to lock the FIFO + LONG initial = InterlockedCompareExchange(&mLock, 1, 0); + return (initial == 0); + } + + void unlock() + { + mLock = 0; + } + + T* peek() + { + if (mNumEntries == 0) + { + return nullptr; + } + uint32_t block = mHead >> mBlockSizeShift; + return &mBlocks[block][mHead & (mBlockSize-1)]; + } + + void dequeue_noinc() + { + mHead ++; + mNumEntries --; + } + + bool enqueue_try_nosync(Arena& arena, const T* entry) + { + memcpy(&mCurBlock[mTail], entry, sizeof(T)); + + mTail ++; + if (mTail == mBlockSize) + { + if (++mCurBlockIdx < mBlocks.size()) + { + mCurBlock = mBlocks[mCurBlockIdx]; + } + else + { + T* newBlock = (T*)arena.Alloc(sizeof(T)*mBlockSize); + SWR_ASSERT(newBlock); + + mBlocks.push_back(newBlock); + mCurBlock = newBlock; + } + + mTail = 0; + } + + mNumEntries ++; + return true; + } + + void destroy() + { + } + +}; diff --git a/src/gallium/drivers/swr/rasterizer/core/format_conversion.h b/src/gallium/drivers/swr/rasterizer/core/format_conversion.h new file mode 100644 index 00000000000..83d85fc86d8 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/format_conversion.h @@ -0,0 +1,196 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file format_conversion.h +* +* @brief API implementation +* +******************************************************************************/ +#include "format_types.h" +#include "format_traits.h" + +////////////////////////////////////////////////////////////////////////// +/// @brief Load SIMD packed pixels in SOA format and converts to +/// SOA RGBA32_FLOAT format. +/// @param pSrc - source data in SOA form +/// @param dst - output data in SOA form +template +INLINE void LoadSOA(const BYTE *pSrc, simdvector &dst) +{ + // fast path for float32 + if ((FormatTraits::GetType(0) == SWR_TYPE_FLOAT) && (FormatTraits::GetBPC(0) == 32)) + { + auto lambda = [&](int comp) + { + simdscalar vComp = _simd_load_ps((const float*)(pSrc + comp*sizeof(simdscalar))); + + dst.v[FormatTraits::swizzle(comp)] = vComp; + }; + + UnrollerL<0, FormatTraits::numComps, 1>::step(lambda); + return; + } + + auto lambda = [&](int comp) + { + // load SIMD components + simdscalar vComp = FormatTraits::loadSOA(comp, pSrc); + + // unpack + vComp = FormatTraits::unpack(comp, vComp); + + // convert + if (FormatTraits::isNormalized(comp)) + { + vComp = _simd_cvtepi32_ps(_simd_castps_si(vComp)); + vComp = _simd_mul_ps(vComp, _simd_set1_ps(FormatTraits::toFloat(comp))); + } + + dst.v[FormatTraits::swizzle(comp)] = vComp; + + pSrc += (FormatTraits::GetBPC(comp) * KNOB_SIMD_WIDTH) / 8; + }; + + UnrollerL<0, FormatTraits::numComps, 1>::step(lambda); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Clamps the given component based on the requirements on the +/// Format template arg +/// @param vComp - SIMD vector of floats +/// @param Component - component +template +INLINE simdscalar Clamp(simdscalar vComp, uint32_t Component) +{ + if (FormatTraits::isNormalized(Component)) + { + if (FormatTraits::GetType(Component) == SWR_TYPE_UNORM) + { + vComp = _simd_max_ps(vComp, _simd_setzero_ps()); + } + + if (FormatTraits::GetType(Component) == SWR_TYPE_SNORM) + { + vComp = _simd_max_ps(vComp, _simd_set1_ps(-1.0f)); + } + vComp = _simd_min_ps(vComp, _simd_set1_ps(1.0f)); + } + else if (FormatTraits::GetBPC(Component) < 32) + { + if (FormatTraits::GetType(Component) == SWR_TYPE_UINT) + { + int iMax = (1 << FormatTraits::GetBPC(Component)) - 1; + int iMin = 0; + simdscalari vCompi = _simd_castps_si(vComp); + vCompi = _simd_max_epu32(vCompi, _simd_set1_epi32(iMin)); + vCompi = _simd_min_epu32(vCompi, _simd_set1_epi32(iMax)); + vComp = _simd_castsi_ps(vCompi); + } + else if (FormatTraits::GetType(Component) == SWR_TYPE_SINT) + { + int iMax = (1 << (FormatTraits::GetBPC(Component) - 1)) - 1; + int iMin = -1 - iMax; + simdscalari vCompi = _simd_castps_si(vComp); + vCompi = _simd_max_epi32(vCompi, _simd_set1_epi32(iMin)); + vCompi = _simd_min_epi32(vCompi, _simd_set1_epi32(iMax)); + vComp = _simd_castsi_ps(vCompi); + } + } + + return vComp; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Normalize the given component based on the requirements on the +/// Format template arg +/// @param vComp - SIMD vector of floats +/// @param Component - component +template +INLINE simdscalar Normalize(simdscalar vComp, uint32_t Component) +{ + if (FormatTraits::isNormalized(Component)) + { + vComp = _simd_mul_ps(vComp, _simd_set1_ps(FormatTraits::fromFloat(Component))); + vComp = _simd_castsi_ps(_simd_cvtps_epi32(vComp)); + } + return vComp; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Convert and store simdvector of pixels in SOA +/// RGBA32_FLOAT to SOA format +/// @param src - source data in SOA form +/// @param dst - output data in SOA form +template +INLINE void StoreSOA(const simdvector &src, BYTE *pDst) +{ + // fast path for float32 + if ((FormatTraits::GetType(0) == SWR_TYPE_FLOAT) && (FormatTraits::GetBPC(0) == 32)) + { + for (uint32_t comp = 0; comp < FormatTraits::numComps; ++comp) + { + simdscalar vComp = src.v[FormatTraits::swizzle(comp)]; + + // Gamma-correct + if (FormatTraits::isSRGB) + { + if (comp < 3) // Input format is always RGBA32_FLOAT. + { + vComp = FormatTraits::convertSrgb(comp, vComp); + } + } + + _simd_store_ps((float*)(pDst + comp*sizeof(simdscalar)), vComp); + } + return; + } + + auto lambda = [&](int comp) + { + simdscalar vComp = src.v[FormatTraits::swizzle(comp)]; + + // Gamma-correct + if (FormatTraits::isSRGB) + { + if (comp < 3) // Input format is always RGBA32_FLOAT. + { + vComp = FormatTraits::convertSrgb(comp, vComp); + } + } + + // clamp + vComp = Clamp(vComp, comp); + + // normalize + vComp = Normalize(vComp, comp); + + // pack + vComp = FormatTraits::pack(comp, vComp); + + // store + FormatTraits::storeSOA(comp, pDst, vComp); + + pDst += (FormatTraits::GetBPC(comp) * KNOB_SIMD_WIDTH) / 8; + }; + + UnrollerL<0, FormatTraits::numComps, 1>::step(lambda); +} diff --git a/src/gallium/drivers/swr/rasterizer/core/format_traits.h b/src/gallium/drivers/swr/rasterizer/core/format_traits.h new file mode 100644 index 00000000000..52340f4987a --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/format_traits.h @@ -0,0 +1,3548 @@ + +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file format_traits.h +* +* @brief auto-generated file +* +* DO NOT EDIT +* +******************************************************************************/ + +#pragma once + +#include "format_types.h" +#include "utils.h" + + +////////////////////////////////////////////////////////////////////////// +/// FormatSwizzle - Component swizzle selects +////////////////////////////////////////////////////////////////////////// +template +struct FormatSwizzle +{ + // Return swizzle select for component. + INLINE static uint32_t swizzle(UINT c) + { + static const uint32_t s[4] = { comp0, comp1, comp2, comp3 }; + return s[c]; + } +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits +////////////////////////////////////////////////////////////////////////// +template +struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0> +{ + static const uint32_t bpp{ 0 }; + static const uint32_t numComps{ 0 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{1}; + static const uint32_t bcHeight{1}; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32B32A32_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32_32_32 TransposeT; + typedef Format4<32, 32, 32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32B32A32_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32_32_32 TransposeT; + typedef Format4<32, 32, 32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32B32A32_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32_32_32 TransposeT; + typedef Format4<32, 32, 32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32B32X32_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32_32_32 TransposeT; + typedef Format4<32, 32, 32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32B32A32_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32_32_32 TransposeT; + typedef Format4<32, 32, 32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32B32A32_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32_32_32 TransposeT; + typedef Format4<32, 32, 32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32B32_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 96 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32_32 TransposeT; + typedef Format3<32, 32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32B32_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 96 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32_32 TransposeT; + typedef Format3<32, 32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32B32_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 96 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32_32 TransposeT; + typedef Format3<32, 32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32B32_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 96 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32_32 TransposeT; + typedef Format3<32, 32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32B32_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 96 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32_32 TransposeT; + typedef Format3<32, 32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16A16_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16_16 TransposeT; + typedef Format4<16, 16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16A16_SNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16_16 TransposeT; + typedef Format4<16, 16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16A16_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16_16 TransposeT; + typedef Format4<16, 16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16A16_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16_16 TransposeT; + typedef Format4<16, 16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16A16_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16_16 TransposeT; + typedef Format4<16, 16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32 TransposeT; + typedef Format2<32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32 TransposeT; + typedef Format2<32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32 TransposeT; + typedef Format2<32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32_FLOAT_X8X24_TYPELESS +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32 TransposeT; + typedef Format2<32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for X32_TYPELESS_G8X24_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32 TransposeT; + typedef Format2<32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L32A32_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 1 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32 TransposeT; + typedef Format2<32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16X16_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16_16 TransposeT; + typedef Format4<16, 16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16X16_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16_16 TransposeT; + typedef Format4<16, 16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L32X32_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32 TransposeT; + typedef Format2<32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for I32X32_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32 TransposeT; + typedef Format2<32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16A16_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16_16 TransposeT; + typedef Format4<16, 16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16A16_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16_16 TransposeT; + typedef Format4<16, 16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32 TransposeT; + typedef Format2<32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32G32_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32 TransposeT; + typedef Format2<32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32_FLOAT_X8X24_TYPELESS_LD +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose32_32 TransposeT; + typedef Format2<32, 32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B8G8R8A8_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B8G8R8A8_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ true }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R10G10B10A2_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R10G10B10A2_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ true }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R10G10B10A2_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8A8_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8A8_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ true }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8A8_SNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8A8_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8A8_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16 TransposeT; + typedef Format2<16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16_SNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16 TransposeT; + typedef Format2<16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16 TransposeT; + typedef Format2<16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16 TransposeT; + typedef Format2<16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16 TransposeT; + typedef Format2<16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B10G10R10A2_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B10G10R10A2_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ true }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R11G11B10_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose11_11_10 TransposeT; + typedef Format3<11, 11, 10> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<32> TransposeT; + typedef Format1<32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<32> TransposeT; + typedef Format1<32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<32> TransposeT; + typedef Format1<32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R24_UNORM_X8_TYPELESS +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<32> TransposeT; + typedef Format1<24> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R24_UNORM_X8_TYPELESS_LD +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<32> TransposeT; + typedef Format1<24> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L16A16_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 1 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16 TransposeT; + typedef Format2<16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for I24X8_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose24_8 TransposeT; + typedef Format2<24, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L24X8_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose24_8 TransposeT; + typedef Format2<24, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for I32_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<32> TransposeT; + typedef Format1<32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L32_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<32> TransposeT; + typedef Format1<32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for A32_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<32> TransposeT; + typedef Format1<32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B8G8R8X8_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B8G8R8X8_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ true }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8X8_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8X8_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ true }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R9G9B9E5_SHAREDEXP +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose9_9_9_5 TransposeT; + typedef Format4<9, 9, 9, 5> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B10G10R10X2_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L16A16_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 1 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16 TransposeT; + typedef Format2<16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R10G10B10X2_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8A8_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8A8_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16 TransposeT; + typedef Format2<16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16 TransposeT; + typedef Format2<16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<32> TransposeT; + typedef Format1<32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R32_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<32> TransposeT; + typedef Format1<32> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B5G6R5_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose5_6_5 TransposeT; + typedef Format3<5, 6, 5> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B5G6R5_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ true }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose5_6_5 TransposeT; + typedef Format3<5, 6, 5> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B5G5R5A1_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose5_5_5_1 TransposeT; + typedef Format4<5, 5, 5, 1> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B5G5R5A1_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ true }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose5_5_5_1 TransposeT; + typedef Format4<5, 5, 5, 1> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B4G4R4A4_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose4_4_4_4 TransposeT; + typedef Format4<4, 4, 4, 4> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B4G4R4A4_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ true }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose4_4_4_4 TransposeT; + typedef Format4<4, 4, 4, 4> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8 TransposeT; + typedef Format2<8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8_SNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8 TransposeT; + typedef Format2<8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8 TransposeT; + typedef Format2<8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8 TransposeT; + typedef Format2<8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<16> TransposeT; + typedef Format1<16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16_SNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<16> TransposeT; + typedef Format1<16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<16> TransposeT; + typedef Format1<16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<16> TransposeT; + typedef Format1<16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<16> TransposeT; + typedef Format1<16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for I16_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<16> TransposeT; + typedef Format1<16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L16_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<16> TransposeT; + typedef Format1<16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for A16_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<16> TransposeT; + typedef Format1<16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L8A8_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 1 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8 TransposeT; + typedef Format2<8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for I16_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<16> TransposeT; + typedef Format1<16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L16_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<16> TransposeT; + typedef Format1<16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for A16_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<16> TransposeT; + typedef Format1<16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L8A8_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 1 }; + static const bool isSRGB{ true }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8 TransposeT; + typedef Format2<8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B5G5R5X1_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose5_5_5_1 TransposeT; + typedef Format4<5, 5, 5, 1> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B5G5R5X1_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ true }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose5_5_5_1 TransposeT; + typedef Format4<5, 5, 5, 1> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8 TransposeT; + typedef Format2<8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8 TransposeT; + typedef Format2<8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<16> TransposeT; + typedef Format1<16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<16> TransposeT; + typedef Format1<16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L8A8_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 1 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8 TransposeT; + typedef Format2<8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L8A8_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 16 }; + static const uint32_t numComps{ 2 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 1 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8 TransposeT; + typedef Format2<8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8_SNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for A8_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for I8_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L8_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L8_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ true }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L8_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for L8_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for I8_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for I8_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 8 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef TransposeSingleComponent<8> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for YCRCB_SWAPUVY +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ true }; + static const uint32_t bcWidth{ 2 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC1_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<64> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC2_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<128> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC3_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<128> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC4_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<64> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC5_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<128> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC1_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ true }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<64> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC2_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ true }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<128> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC3_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ true }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<128> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for YCRCB_SWAPUV +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ true }; + static const uint32_t bcWidth{ 2 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8_8 TransposeT; + typedef Format4<8, 8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 24 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8 TransposeT; + typedef Format3<8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8_SNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 24 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8 TransposeT; + typedef Format3<8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 24 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8 TransposeT; + typedef Format3<8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 24 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8 TransposeT; + typedef Format3<8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC4_SNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 64 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<64> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC5_SNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<128> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16_FLOAT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 48 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16 TransposeT; + typedef Format3<16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 48 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16 TransposeT; + typedef Format3<16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16_SNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 48 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16 TransposeT; + typedef Format3<16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 48 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16 TransposeT; + typedef Format3<16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 48 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16 TransposeT; + typedef Format3<16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC6H_SF16 +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<128> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC7_UNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<128> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC7_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ true }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<128> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for BC6H_UF16 +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 128 }; + static const uint32_t numComps{ 1 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ true }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 4 }; + static const uint32_t bcHeight{ 4 }; + + typedef TransposeSingleComponent<128> TransposeT; + typedef Format1<8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8_UNORM_SRGB +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 24 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ true }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8 TransposeT; + typedef Format3<8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 48 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16 TransposeT; + typedef Format3<16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R16G16B16_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 48 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose16_16_16 TransposeT; + typedef Format3<16, 16, 16> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R10G10B10A2_SNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R10G10B10A2_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R10G10B10A2_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R10G10B10A2_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B10G10R10A2_SNORM +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B10G10R10A2_USCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B10G10R10A2_SSCALED +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x3f800000> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B10G10R10A2_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for B10G10R10A2_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<2, 1, 0, 3>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 32 }; + static const uint32_t numComps{ 4 }; + static const bool hasAlpha{ true }; + static const uint32_t alphaComp{ 3 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose10_10_10_2 TransposeT; + typedef Format4<10, 10, 10, 2> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8_UINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 24 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8 TransposeT; + typedef Format3<8, 8, 8> FormatT; +}; + +////////////////////////////////////////////////////////////////////////// +/// FormatTraits - Format traits specialization for R8G8B8_SINT +////////////////////////////////////////////////////////////////////////// +template<> struct FormatTraits : + ComponentTraits, + FormatSwizzle<0, 1, 2>, + Defaults<0, 0, 0, 0x1> +{ + static const uint32_t bpp{ 24 }; + static const uint32_t numComps{ 3 }; + static const bool hasAlpha{ false }; + static const uint32_t alphaComp{ 0 }; + static const bool isSRGB{ false }; + static const bool isBC{ false }; + static const bool isSubsampled{ false }; + static const uint32_t bcWidth{ 1 }; + static const uint32_t bcHeight{ 1 }; + + typedef Transpose8_8_8 TransposeT; + typedef Format3<8, 8, 8> FormatT; +}; + diff --git a/src/gallium/drivers/swr/rasterizer/core/format_types.h b/src/gallium/drivers/swr/rasterizer/core/format_types.h new file mode 100644 index 00000000000..aa350259a15 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/format_types.h @@ -0,0 +1,1075 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file formats.h +* +* @brief Definitions for SWR_FORMAT functions. +* +******************************************************************************/ +#pragma once + +////////////////////////////////////////////////////////////////////////// +/// PackTraits - Helpers for packing / unpacking same pixel sizes +////////////////////////////////////////////////////////////////////////// +template +struct PackTraits +{ + static const uint32_t MyNumBits = NumBits; + static simdscalar loadSOA(const BYTE *pSrc) = delete; + static void storeSOA(BYTE *pDst, simdscalar src) = delete; + static simdscalar unpack(simdscalar &in) = delete; + static simdscalar pack(simdscalar &in) = delete; +}; + +////////////////////////////////////////////////////////////////////////// +/// PackTraits - Helpers for packing / unpacking unused channels +////////////////////////////////////////////////////////////////////////// +template <> +struct PackTraits<0, false> +{ + static const uint32_t MyNumBits = 0; + + static simdscalar loadSOA(const BYTE *pSrc) { return _simd_setzero_ps(); } + static void storeSOA(BYTE *pDst, simdscalar src) { return; } + static simdscalar unpack(simdscalar &in) { return _simd_setzero_ps(); } + static simdscalar pack(simdscalar &in) { return _simd_setzero_ps(); } +}; + + +////////////////////////////////////////////////////////////////////////// +/// PackTraits - Helpers for packing / unpacking 8 bit unsigned channels +////////////////////////////////////////////////////////////////////////// +template <> +struct PackTraits<8, false> +{ + static const uint32_t MyNumBits = 8; + + static simdscalar loadSOA(const BYTE *pSrc) + { +#if KNOB_SIMD_WIDTH == 8 + __m256 result = _mm256_setzero_ps(); + __m128 vLo = _mm_castpd_ps(_mm_load_sd((double*)pSrc)); + return _mm256_insertf128_ps(result, vLo, 0); +#else +#error Unsupported vector width +#endif + } + + static void storeSOA(BYTE *pDst, simdscalar src) + { + // store simd bytes +#if KNOB_SIMD_WIDTH == 8 + _mm_storel_pd((double*)pDst, _mm_castps_pd(_mm256_castps256_ps128(src))); +#else +#error Unsupported vector width +#endif + } + + static simdscalar unpack(simdscalar &in) + { +#if KNOB_SIMD_WIDTH == 8 +#if KNOB_ARCH==KNOB_ARCH_AVX + __m128i src = _mm_castps_si128(_mm256_castps256_ps128(in)); + __m128i resLo = _mm_cvtepu8_epi32(src); + __m128i resHi = _mm_shuffle_epi8(src, + _mm_set_epi32(0x80808007, 0x80808006, 0x80808005, 0x80808004)); + + __m256i result = _mm256_castsi128_si256(resLo); + result = _mm256_insertf128_si256(result, resHi, 1); + return _mm256_castsi256_ps(result); +#elif KNOB_ARCH==KNOB_ARCH_AVX2 + return _mm256_castsi256_ps(_mm256_cvtepu8_epi32(_mm_castps_si128(_mm256_castps256_ps128(in)))); +#endif +#else +#error Unsupported vector width +#endif + } + + static simdscalar pack(simdscalar &in) + { +#if KNOB_SIMD_WIDTH == 8 + simdscalari src = _simd_castps_si(in); + __m128i res16 = _mm_packus_epi32(_mm256_castsi256_si128(src), _mm256_extractf128_si256(src, 1)); + __m128i res8 = _mm_packus_epi16(res16, _mm_undefined_si128()); + return _mm256_castsi256_ps(_mm256_castsi128_si256(res8)); +#else +#error Unsupported vector width +#endif + } +}; + +////////////////////////////////////////////////////////////////////////// +/// PackTraits - Helpers for packing / unpacking 8 bit signed channels +////////////////////////////////////////////////////////////////////////// +template <> +struct PackTraits<8, true> +{ + static const uint32_t MyNumBits = 8; + + static simdscalar loadSOA(const BYTE *pSrc) + { +#if KNOB_SIMD_WIDTH == 8 + __m256 result = _mm256_setzero_ps(); + __m128 vLo = _mm_castpd_ps(_mm_load_sd((double*)pSrc)); + return _mm256_insertf128_ps(result, vLo, 0); +#else +#error Unsupported vector width +#endif + } + + static void storeSOA(BYTE *pDst, simdscalar src) + { + // store simd bytes +#if KNOB_SIMD_WIDTH == 8 + _mm_storel_pd((double*)pDst, _mm_castps_pd(_mm256_castps256_ps128(src))); +#else +#error Unsupported vector width +#endif + } + + static simdscalar unpack(simdscalar &in) + { +#if KNOB_SIMD_WIDTH == 8 +#if KNOB_ARCH==KNOB_ARCH_AVX + SWR_ASSERT(0); // I think this may be incorrect. + __m128i src = _mm_castps_si128(_mm256_castps256_ps128(in)); + __m128i resLo = _mm_cvtepi8_epi32(src); + __m128i resHi = _mm_shuffle_epi8(src, + _mm_set_epi32(0x80808007, 0x80808006, 0x80808005, 0x80808004)); + + __m256i result = _mm256_castsi128_si256(resLo); + result = _mm256_insertf128_si256(result, resHi, 1); + return _mm256_castsi256_ps(result); +#elif KNOB_ARCH==KNOB_ARCH_AVX2 + return _mm256_castsi256_ps(_mm256_cvtepi8_epi32(_mm_castps_si128(_mm256_castps256_ps128(in)))); +#endif +#else +#error Unsupported vector width +#endif + } + + static simdscalar pack(simdscalar &in) + { +#if KNOB_SIMD_WIDTH == 8 + simdscalari src = _simd_castps_si(in); + __m128i res16 = _mm_packs_epi32(_mm256_castsi256_si128(src), _mm256_extractf128_si256(src, 1)); + __m128i res8 = _mm_packs_epi16(res16, _mm_undefined_si128()); + return _mm256_castsi256_ps(_mm256_castsi128_si256(res8)); +#else +#error Unsupported vector width +#endif + } +}; + +////////////////////////////////////////////////////////////////////////// +/// PackTraits - Helpers for packing / unpacking 16 bit unsigned channels +////////////////////////////////////////////////////////////////////////// +template <> +struct PackTraits<16, false> +{ + static const uint32_t MyNumBits = 16; + + static simdscalar loadSOA(const BYTE *pSrc) + { +#if KNOB_SIMD_WIDTH == 8 + __m256 result = _mm256_setzero_ps(); + __m128 vLo = _mm_load_ps((const float*)pSrc); + return _mm256_insertf128_ps(result, vLo, 0); +#else +#error Unsupported vector width +#endif + } + + static void storeSOA(BYTE *pDst, simdscalar src) + { +#if KNOB_SIMD_WIDTH == 8 + // store 16B (2B * 8) + _mm_store_ps((float*)pDst, _mm256_castps256_ps128(src)); +#else +#error Unsupported vector width +#endif + } + + static simdscalar unpack(simdscalar &in) + { +#if KNOB_SIMD_WIDTH == 8 +#if KNOB_ARCH==KNOB_ARCH_AVX + __m128i src = _mm_castps_si128(_mm256_castps256_ps128(in)); + __m128i resLo = _mm_cvtepu16_epi32(src); + __m128i resHi = _mm_shuffle_epi8(src, + _mm_set_epi32(0x80800F0E, 0x80800D0C, 0x80800B0A, 0x80800908)); + + __m256i result = _mm256_castsi128_si256(resLo); + result = _mm256_insertf128_si256(result, resHi, 1); + return _mm256_castsi256_ps(result); +#elif KNOB_ARCH==KNOB_ARCH_AVX2 + return _mm256_castsi256_ps(_mm256_cvtepu16_epi32(_mm_castps_si128(_mm256_castps256_ps128(in)))); +#endif +#else +#error Unsupported vector width +#endif + } + + static simdscalar pack(simdscalar &in) + { +#if KNOB_SIMD_WIDTH == 8 + simdscalari src = _simd_castps_si(in); + __m256i res = _mm256_castsi128_si256(_mm_packus_epi32(_mm256_castsi256_si128(src), _mm256_extractf128_si256(src, 1))); + return _mm256_castsi256_ps(res); +#else +#error Unsupported vector width +#endif + } +}; + +////////////////////////////////////////////////////////////////////////// +/// PackTraits - Helpers for packing / unpacking 16 bit signed channels +////////////////////////////////////////////////////////////////////////// +template <> +struct PackTraits<16, true> +{ + static const uint32_t MyNumBits = 16; + + static simdscalar loadSOA(const BYTE *pSrc) + { +#if KNOB_SIMD_WIDTH == 8 + __m256 result = _mm256_setzero_ps(); + __m128 vLo = _mm_load_ps((const float*)pSrc); + return _mm256_insertf128_ps(result, vLo, 0); +#else +#error Unsupported vector width +#endif + } + + static void storeSOA(BYTE *pDst, simdscalar src) + { +#if KNOB_SIMD_WIDTH == 8 + // store 16B (2B * 8) + _mm_store_ps((float*)pDst, _mm256_castps256_ps128(src)); +#else +#error Unsupported vector width +#endif + } + + static simdscalar unpack(simdscalar &in) + { +#if KNOB_SIMD_WIDTH == 8 +#if KNOB_ARCH==KNOB_ARCH_AVX + SWR_ASSERT(0); // I think this is incorrectly implemented + __m128i src = _mm_castps_si128(_mm256_castps256_ps128(in)); + __m128i resLo = _mm_cvtepi16_epi32(src); + __m128i resHi = _mm_shuffle_epi8(src, + _mm_set_epi32(0x80800F0E, 0x80800D0C, 0x80800B0A, 0x80800908)); + + __m256i result = _mm256_castsi128_si256(resLo); + result = _mm256_insertf128_si256(result, resHi, 1); + return _mm256_castsi256_ps(result); +#elif KNOB_ARCH==KNOB_ARCH_AVX2 + return _mm256_castsi256_ps(_mm256_cvtepi16_epi32(_mm_castps_si128(_mm256_castps256_ps128(in)))); +#endif +#else +#error Unsupported vector width +#endif + } + + static simdscalar pack(simdscalar &in) + { +#if KNOB_SIMD_WIDTH == 8 + simdscalari src = _simd_castps_si(in); + __m256i res = _mm256_castsi128_si256(_mm_packs_epi32(_mm256_castsi256_si128(src), _mm256_extractf128_si256(src, 1))); + return _mm256_castsi256_ps(res); +#else +#error Unsupported vector width +#endif + } +}; + +////////////////////////////////////////////////////////////////////////// +/// PackTraits - Helpers for packing / unpacking 32 bit channels +////////////////////////////////////////////////////////////////////////// +template <> +struct PackTraits<32, false> +{ + static const uint32_t MyNumBits = 32; + + static simdscalar loadSOA(const BYTE *pSrc) { return _simd_load_ps((const float*)pSrc); } + static void storeSOA(BYTE *pDst, simdscalar src) { _simd_store_ps((float*)pDst, src); } + static simdscalar unpack(simdscalar &in) { return in; } + static simdscalar pack(simdscalar &in) { return in; } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits. +////////////////////////////////////////////////////////////////////////// +template +struct TypeTraits : PackTraits +{ + static const SWR_TYPE MyType = type; + static float toFloat() { return 0.0; } + static float fromFloat() { SWR_ASSERT(0); return 0.0; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for UINT8 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<8> +{ + static const SWR_TYPE MyType = SWR_TYPE_UINT; + static float toFloat() { return 0.0; } + static float fromFloat() { SWR_ASSERT(0); return 0.0; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for UINT8 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<8, true> +{ + static const SWR_TYPE MyType = SWR_TYPE_SINT; + static float toFloat() { return 0.0; } + static float fromFloat() { SWR_ASSERT(0); return 0.0; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for UINT16 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<16> +{ + static const SWR_TYPE MyType = SWR_TYPE_UINT; + static float toFloat() { return 0.0; } + static float fromFloat() { SWR_ASSERT(0); return 0.0; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for SINT16 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<16, true> +{ + static const SWR_TYPE MyType = SWR_TYPE_SINT; + static float toFloat() { return 0.0; } + static float fromFloat() { SWR_ASSERT(0); return 0.0; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for UINT32 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<32> +{ + static const SWR_TYPE MyType = SWR_TYPE_UINT; + static float toFloat() { return 0.0; } + static float fromFloat() { SWR_ASSERT(0); return 0.0; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for UINT32 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<32> +{ + static const SWR_TYPE MyType = SWR_TYPE_SINT; + static float toFloat() { return 0.0; } + static float fromFloat() { SWR_ASSERT(0); return 0.0; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for UNORM5 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<5> +{ + static const SWR_TYPE MyType = SWR_TYPE_UNORM; + static float toFloat() { return 1.0f / 31.0f; } + static float fromFloat() { return 31.0f; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for UNORM6 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<6> +{ + static const SWR_TYPE MyType = SWR_TYPE_UNORM; + static float toFloat() { return 1.0f / 63.0f; } + static float fromFloat() { return 63.0f; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for UNORM8 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<8> +{ + static const SWR_TYPE MyType = SWR_TYPE_UNORM; + static float toFloat() { return 1.0f / 255.0f; } + static float fromFloat() { return 255.0f; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for UNORM8 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<8, true> +{ + static const SWR_TYPE MyType = SWR_TYPE_SNORM; + static float toFloat() { return 1.0f / 127.0f; } + static float fromFloat() { return 127.0f; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for UNORM16 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<16> +{ + static const SWR_TYPE MyType = SWR_TYPE_UNORM; + static float toFloat() { return 1.0f / 65535.0f; } + static float fromFloat() { return 65535.0f; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for SNORM16 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<16, true> +{ + static const SWR_TYPE MyType = SWR_TYPE_UNORM; + static float toFloat() { return 1.0f / 32767.0f; } + static float fromFloat() { return 32767.0f; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for UNORM24 +////////////////////////////////////////////////////////////////////////// +template<> +struct TypeTraits < SWR_TYPE_UNORM, 24 > : PackTraits<32> +{ + static const SWR_TYPE MyType = SWR_TYPE_UNORM; + static float toFloat() { return 1.0f / 16777215.0f; } + static float fromFloat() { return 16777215.0f; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } +}; + +////////////////////////////////////////////////////////////////////////// +// FLOAT Specializations from here on... +////////////////////////////////////////////////////////////////////////// +#define TO_M128i(a) _mm_castps_si128(a) +#define TO_M128(a) _mm_castsi128_ps(a) + +#include "math.h" + +template< unsigned expnum, unsigned expden, unsigned coeffnum, unsigned coeffden > +inline static __m128 fastpow(__m128 arg) { + __m128 ret = arg; + + static const __m128 factor = _mm_set1_ps(exp2(127.0f * expden / expnum - 127.0f) + * powf(1.0f * coeffnum / coeffden, 1.0f * expden / expnum)); + + // Apply a constant pre-correction factor. + ret = _mm_mul_ps(ret, factor); + + // Reinterpret arg as integer to obtain logarithm. + //asm("cvtdq2ps %1, %0" : "=x" (ret) : "x" (ret)); + ret = _mm_cvtepi32_ps(_mm_castps_si128(ret)); + + // Multiply logarithm by power. + ret = _mm_mul_ps(ret, _mm_set1_ps(1.0f * expnum / expden)); + + // Convert back to "integer" to exponentiate. + //asm("cvtps2dq %1, %0" : "=x" (ret) : "x" (ret)); + ret = _mm_castsi128_ps(_mm_cvtps_epi32(ret)); + + return ret; +} + +inline static __m128 pow512_4(__m128 arg) { + // 5/12 is too small, so compute the 4th root of 20/12 instead. + // 20/12 = 5/3 = 1 + 2/3 = 2 - 1/3. 2/3 is a suitable argument for fastpow. + // weighting coefficient: a^-1/2 = 2 a; a = 2^-2/3 + __m128 xf = fastpow< 2, 3, int(0.629960524947437 * 1e9), int(1e9) >(arg); + __m128 xover = _mm_mul_ps(arg, xf); + + __m128 xfm1 = _mm_rsqrt_ps(xf); + __m128 x2 = _mm_mul_ps(arg, arg); + __m128 xunder = _mm_mul_ps(x2, xfm1); + + // sqrt2 * over + 2 * sqrt2 * under + __m128 xavg = _mm_mul_ps(_mm_set1_ps(1.0f / (3.0f * 0.629960524947437f) * 0.999852f), + _mm_add_ps(xover, xunder)); + + xavg = _mm_mul_ps(xavg, _mm_rsqrt_ps(xavg)); + xavg = _mm_mul_ps(xavg, _mm_rsqrt_ps(xavg)); + return xavg; +} + +inline static __m128 powf_wrapper(__m128 Base, float Exp) +{ + float *f = (float *)(&Base); + + return _mm_set_ps(powf(f[0], Exp), + powf(f[1], Exp), + powf(f[2], Exp), + powf(f[3], Exp)); +} + +static inline __m128 ConvertFloatToSRGB2(__m128& Src) +{ + // create a mask with 0xFFFFFFFF in the DWORDs where the source is <= the minimal SRGB float value + __m128i CmpToSRGBThresholdMask = TO_M128i(_mm_cmpnlt_ps(_mm_set1_ps(0.0031308f), Src)); + + // squeeze the mask down to 16 bits (4 bits per DWORD) + int CompareResult = _mm_movemask_epi8(CmpToSRGBThresholdMask); + + __m128 Result; + + // + if (CompareResult == 0xFFFF) + { + // all DWORDs are <= the threshold + Result = _mm_mul_ps(Src, _mm_set1_ps(12.92f)); + } + else if (CompareResult == 0x0) + { + // all DWORDs are > the threshold + __m128 fSrc_0RGB = Src; + + // --> 1.055f * c(1.0f/2.4f) - 0.055f +#if KNOB_USE_FAST_SRGB == TRUE + // 1.0f / 2.4f is 5.0f / 12.0f which is used for approximation. + __m128 f = pow512_4(fSrc_0RGB); +#else + __m128 f = powf_wrapper(fSrc_0RGB, 1.0f / 2.4f); +#endif + f = _mm_mul_ps(f, _mm_set1_ps(1.055f)); + Result = _mm_sub_ps(f, _mm_set1_ps(0.055f)); + } + else + { + // some DWORDs are <= the threshold and some are > threshold + __m128 Src_0RGB_mul_denorm = _mm_mul_ps(Src, _mm_set1_ps(12.92f)); + + __m128 fSrc_0RGB = Src; + + // --> 1.055f * c(1.0f/2.4f) - 0.055f +#if KNOB_USE_FAST_SRGB == TRUE + // 1.0f / 2.4f is 5.0f / 12.0f which is used for approximation. + __m128 f = pow512_4(fSrc_0RGB); +#else + __m128 f = powf_wrapper(fSrc_0RGB, 1.0f / 2.4f); +#endif + f = _mm_mul_ps(f, _mm_set1_ps(1.055f)); + f = _mm_sub_ps(f, _mm_set1_ps(0.055f)); + + // Clear the alpha (is garbage after the sub) + __m128i i = _mm_and_si128(TO_M128i(f), _mm_set_epi32(0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF, 0xFFFFFFFF)); + + __m128i LessThanPart = _mm_and_si128(CmpToSRGBThresholdMask, TO_M128i(Src_0RGB_mul_denorm)); + __m128i GreaterEqualPart = _mm_andnot_si128(CmpToSRGBThresholdMask, i); + __m128i CombinedParts = _mm_or_si128(LessThanPart, GreaterEqualPart); + + Result = TO_M128(CombinedParts); + } + + return Result; +} + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for FLOAT16 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<16> +{ + static const SWR_TYPE MyType = SWR_TYPE_FLOAT; + static float toFloat() { return 1.0f; } + static float fromFloat() { return 1.0f; } + static simdscalar convertSrgb(simdscalar &in) { SWR_ASSERT(0); return _simd_setzero_ps(); } + + static simdscalar pack(const simdscalar &in) + { +#if KNOB_SIMD_WIDTH == 8 +#if (KNOB_ARCH == KNOB_ARCH_AVX) + // input is 8 packed float32, output is 8 packed float16 + simdscalari src = _simd_castps_si(in); + + static const uint32_t FLOAT_EXP_BITS = 8; + static const uint32_t FLOAT_MANTISSA_BITS = 23; + static const uint32_t FLOAT_MANTISSA_MASK = (1U << FLOAT_MANTISSA_BITS) - 1; + static const uint32_t FLOAT_EXP_MASK = ((1U << FLOAT_EXP_BITS) - 1) << FLOAT_MANTISSA_BITS; + + static const uint32_t HALF_EXP_BITS = 5; + static const uint32_t HALF_MANTISSA_BITS = 10; + static const uint32_t HALF_MANTISSA_MASK = (1U << HALF_MANTISSA_BITS) - 1; + static const uint32_t HALF_EXP_MASK = ((1U << HALF_EXP_BITS) - 1) << HALF_MANTISSA_BITS; + + // minimum exponent required, exponents below this are flushed to 0. + static const int32_t HALF_EXP_MIN = -14; + static const int32_t FLOAT_EXP_BIAS = 127; + static const int32_t FLOAT_EXP_MIN = HALF_EXP_MIN + FLOAT_EXP_BIAS; + static const int32_t FLOAT_EXP_MIN_FTZ = FLOAT_EXP_MIN - (HALF_MANTISSA_BITS + 1); // +1 for the lack of implicit significand + + // maximum exponent required, exponents above this are set to infinity + static const int32_t HALF_EXP_MAX = 15; + static const int32_t FLOAT_EXP_MAX = HALF_EXP_MAX + FLOAT_EXP_BIAS; + + const simdscalari vSignMask = _simd_set1_epi32(0x80000000); + const simdscalari vExpMask = _simd_set1_epi32(FLOAT_EXP_MASK); + const simdscalari vManMask = _simd_set1_epi32(FLOAT_MANTISSA_MASK); + const simdscalari vExpMin = _simd_set1_epi32(FLOAT_EXP_MASK & uint32_t(FLOAT_EXP_MIN << FLOAT_MANTISSA_BITS)); + const simdscalari vExpMinFtz = _simd_set1_epi32(FLOAT_EXP_MASK & uint32_t(FLOAT_EXP_MIN_FTZ << FLOAT_MANTISSA_BITS)); + const simdscalari vExpMax = _simd_set1_epi32(FLOAT_EXP_MASK & uint32_t(FLOAT_EXP_MAX << FLOAT_MANTISSA_BITS)); + + simdscalari vSign = _simd_and_si(src, vSignMask); + simdscalari vExp = _simd_and_si(src, vExpMask); + simdscalari vMan = _simd_and_si(src, vManMask); + + simdscalari vFTZMask = _simd_cmplt_epi32(vExp, vExpMinFtz); + simdscalari vDenormMask = _simd_andnot_si(vFTZMask, _simd_cmplt_epi32(vExp, vExpMin)); + simdscalari vInfMask = _simd_cmpeq_epi32(vExpMask, vExp); + simdscalari vClampMask = _simd_andnot_si(vInfMask, _simd_cmplt_epi32(vExpMax, vExp)); + + simdscalari vHalfExp = _simd_add_epi32(_simd_sub_epi32(vExp, vExpMin), _simd_set1_epi32(1U << FLOAT_MANTISSA_BITS)); + + // pack output 16-bits into the lower 16-bits of each 32-bit channel + simdscalari vDst = _simd_and_si(_simd_srli_epi32(vHalfExp, 13), _simd_set1_epi32(HALF_EXP_MASK)); + vDst = _simd_or_si(vDst, _simd_srli_epi32(vMan, FLOAT_MANTISSA_BITS - HALF_MANTISSA_BITS)); + + // Flush To Zero + vDst = _simd_andnot_si(vFTZMask, vDst); + // Apply Infinites / NaN + vDst = _simd_or_si(vDst, _simd_and_si(vInfMask, _simd_set1_epi32(HALF_EXP_MASK))); + + // Apply clamps + vDst = _simd_andnot_si(vClampMask, vDst); + vDst = _simd_or_si(vDst, + _simd_and_si(vClampMask, _simd_set1_epi32(0x7BFF))); + + // Compute Denormals (subnormals) + if (!_mm256_testz_si256(vDenormMask, vDenormMask)) + { + uint32_t *pDenormMask = (uint32_t*)&vDenormMask; + uint32_t *pExp = (uint32_t*)&vExp; + uint32_t *pMan = (uint32_t*)&vMan; + uint32_t *pDst = (uint32_t*)&vDst; + for (uint32_t i = 0; i < KNOB_SIMD_WIDTH; ++i) + { + if (pDenormMask[i]) + { + // Need to compute subnormal value + uint32_t exponent = pExp[i] >> FLOAT_MANTISSA_BITS; + uint32_t mantissa = pMan[i] | + (1U << FLOAT_MANTISSA_BITS); // Denorms include no "implicit" 1s. Make it explicit + + pDst[i] = mantissa >> ((FLOAT_EXP_MIN - exponent) + (FLOAT_MANTISSA_BITS - HALF_MANTISSA_BITS)); + } + } + } + + // Add in sign bits + vDst = _simd_or_si(vDst, _simd_srli_epi32(vSign, 16)); + + // Pack to lower 128-bits + vDst = _mm256_castsi128_si256(_mm_packus_epi32(_mm256_castsi256_si128(vDst), _mm256_extractf128_si256(vDst, 1))); + +#if 0 +#if !defined(NDEBUG) + simdscalari vCheck = _mm256_castsi128_si256(_mm256_cvtps_ph(in, _MM_FROUND_TRUNC)); + + for (uint32_t i = 0; i < 4; ++i) + { + SWR_ASSERT(vCheck.m256i_i32[i] == vDst.m256i_i32[i]); + } +#endif +#endif + + return _simd_castsi_ps(vDst); + +#else + return _mm256_castsi256_ps(_mm256_castsi128_si256(_mm256_cvtps_ph(in, _MM_FROUND_TRUNC))); +#endif +#else +#error Unsupported vector width +#endif + } + + static simdscalar unpack(const simdscalar &in) + { + // input is 8 packed float16, output is 8 packed float32 + SWR_ASSERT(0); // @todo + return _simd_setzero_ps(); + } +}; + +////////////////////////////////////////////////////////////////////////// +/// TypeTraits - Format type traits specialization for FLOAT32 +////////////////////////////////////////////////////////////////////////// +template<> struct TypeTraits : PackTraits<32> +{ + static const SWR_TYPE MyType = SWR_TYPE_FLOAT; + static float toFloat() { return 1.0f; } + static float fromFloat() { return 1.0f; } + static inline simdscalar convertSrgb(simdscalar &in) + { +#if (KNOB_ARCH == KNOB_ARCH_AVX || KNOB_ARCH == KNOB_ARCH_AVX2) + __m128 srcLo = _mm256_extractf128_ps(in, 0); + __m128 srcHi = _mm256_extractf128_ps(in, 1); + + srcLo = ConvertFloatToSRGB2(srcLo); + srcHi = ConvertFloatToSRGB2(srcHi); + + in = _mm256_insertf128_ps(in, srcLo, 0); + in = _mm256_insertf128_ps(in, srcHi, 1); + +#endif + return in; + } +}; + +////////////////////////////////////////////////////////////////////////// +/// Format1 - Bitfield for single component formats. +////////////////////////////////////////////////////////////////////////// +template +struct Format1 +{ + union + { + uint32_t r : x; + + ///@ The following are here to provide full template needed in Formats. + uint32_t g : x; + uint32_t b : x; + uint32_t a : x; + }; +}; + +////////////////////////////////////////////////////////////////////////// +/// Format1 - Bitfield for single component formats - 8 bit specialization +////////////////////////////////////////////////////////////////////////// +template<> +struct Format1<8> +{ + union + { + uint8_t r; + + ///@ The following are here to provide full template needed in Formats. + uint8_t g; + uint8_t b; + uint8_t a; + }; +}; + +////////////////////////////////////////////////////////////////////////// +/// Format1 - Bitfield for single component formats - 16 bit specialization +////////////////////////////////////////////////////////////////////////// +template<> +struct Format1<16> +{ + union + { + uint16_t r; + + ///@ The following are here to provide full template needed in Formats. + uint16_t g; + uint16_t b; + uint16_t a; + }; +}; + +////////////////////////////////////////////////////////////////////////// +/// Format2 - Bitfield for 2 component formats. +////////////////////////////////////////////////////////////////////////// +template +union Format2 +{ + struct + { + uint32_t r : x; + uint32_t g : y; + }; + struct + { + ///@ The following are here to provide full template needed in Formats. + uint32_t b : x; + uint32_t a : y; + }; +}; + +////////////////////////////////////////////////////////////////////////// +/// Format2 - Bitfield for 2 component formats - 16 bit specialization +////////////////////////////////////////////////////////////////////////// +template<> +union Format2<8,8> +{ + struct + { + uint16_t r : 8; + uint16_t g : 8; + }; + struct + { + ///@ The following are here to provide full template needed in Formats. + uint16_t b : 8; + uint16_t a : 8; + }; +}; + +////////////////////////////////////////////////////////////////////////// +/// Format3 - Bitfield for 3 component formats. +////////////////////////////////////////////////////////////////////////// +template +union Format3 +{ + struct + { + uint32_t r : x; + uint32_t g : y; + uint32_t b : z; + }; + uint32_t a; ///@note This is here to provide full template needed in Formats. +}; + +////////////////////////////////////////////////////////////////////////// +/// Format3 - Bitfield for 3 component formats - 16 bit specialization +////////////////////////////////////////////////////////////////////////// +template<> +union Format3<5,6,5> +{ + struct + { + uint16_t r : 5; + uint16_t g : 6; + uint16_t b : 5; + }; + uint16_t a; ///@note This is here to provide full template needed in Formats. +}; + +////////////////////////////////////////////////////////////////////////// +/// Format4 - Bitfield for 4 component formats. +////////////////////////////////////////////////////////////////////////// +template +struct Format4 +{ + uint32_t r : x; + uint32_t g : y; + uint32_t b : z; + uint32_t a : w; +}; + +////////////////////////////////////////////////////////////////////////// +/// Format4 - Bitfield for 4 component formats - 16 bit specialization +////////////////////////////////////////////////////////////////////////// +template<> +struct Format4<5,5,5,1> +{ + uint16_t r : 5; + uint16_t g : 5; + uint16_t b : 5; + uint16_t a : 1; +}; + +////////////////////////////////////////////////////////////////////////// +/// Format4 - Bitfield for 4 component formats - 16 bit specialization +////////////////////////////////////////////////////////////////////////// +template<> +struct Format4<4,4,4,4> +{ + uint16_t r : 4; + uint16_t g : 4; + uint16_t b : 4; + uint16_t a : 4; +}; + +////////////////////////////////////////////////////////////////////////// +/// ComponentTraits - Default components +////////////////////////////////////////////////////////////////////////// +template +struct Defaults +{ + INLINE static uint32_t GetDefault(uint32_t comp) + { + static const uint32_t defaults[4]{ x, y, z, w }; + return defaults[comp]; + } +}; + +////////////////////////////////////////////////////////////////////////// +/// ComponentTraits - Component type traits. +////////////////////////////////////////////////////////////////////////// +template +struct ComponentTraits +{ + INLINE static SWR_TYPE GetType(uint32_t comp) + { + static const SWR_TYPE CompType[4]{ X, Y, Z, W }; + return CompType[comp]; + } + + INLINE static uint32_t GetBPC(uint32_t comp) + { + static const uint32_t MyBpc[4]{ NumBitsX, NumBitsY, NumBitsZ, NumBitsW }; + return MyBpc[comp]; + } + + INLINE static bool isNormalized(uint32_t comp) + { + switch (comp) + { + case 0: + return (X == SWR_TYPE_UNORM || X == SWR_TYPE_SNORM) ? true : false; + case 1: + return (Y == SWR_TYPE_UNORM || Y == SWR_TYPE_SNORM) ? true : false; + case 2: + return (Z == SWR_TYPE_UNORM || Z == SWR_TYPE_SNORM) ? true : false; + case 3: + return (W == SWR_TYPE_UNORM || W == SWR_TYPE_SNORM) ? true : false; + } + SWR_ASSERT(0); + return false; + } + + INLINE static float toFloat(uint32_t comp) + { + switch (comp) + { + case 0: + return TypeTraits::toFloat(); + case 1: + return TypeTraits::toFloat(); + case 2: + return TypeTraits::toFloat(); + case 3: + return TypeTraits::toFloat(); + } + SWR_ASSERT(0); + return TypeTraits::toFloat(); + + } + + INLINE static float fromFloat(uint32_t comp) + { + switch (comp) + { + case 0: + return TypeTraits::fromFloat(); + case 1: + return TypeTraits::fromFloat(); + case 2: + return TypeTraits::fromFloat(); + case 3: + return TypeTraits::fromFloat(); + } + SWR_ASSERT(0); + return TypeTraits::fromFloat(); + } + + INLINE static simdscalar loadSOA(uint32_t comp, const BYTE* pSrc) + { + switch (comp) + { + case 0: + return TypeTraits::loadSOA(pSrc); + case 1: + return TypeTraits::loadSOA(pSrc); + case 2: + return TypeTraits::loadSOA(pSrc); + case 3: + return TypeTraits::loadSOA(pSrc); + } + SWR_ASSERT(0); + return TypeTraits::loadSOA(pSrc); + } + + INLINE static void storeSOA(uint32_t comp, BYTE *pDst, simdscalar src) + { + switch (comp) + { + case 0: + TypeTraits::storeSOA(pDst, src); + return; + case 1: + TypeTraits::storeSOA(pDst, src); + return; + case 2: + TypeTraits::storeSOA(pDst, src); + return; + case 3: + TypeTraits::storeSOA(pDst, src); + return; + } + SWR_ASSERT(0); + TypeTraits::storeSOA(pDst, src); + } + + INLINE static simdscalar unpack(uint32_t comp, simdscalar &in) + { + switch (comp) + { + case 0: + return TypeTraits::unpack(in); + case 1: + return TypeTraits::unpack(in); + case 2: + return TypeTraits::unpack(in); + case 3: + return TypeTraits::unpack(in); + } + SWR_ASSERT(0); + return TypeTraits::unpack(in); + } + + INLINE static simdscalar pack(uint32_t comp, simdscalar &in) + { + switch (comp) + { + case 0: + return TypeTraits::pack(in); + case 1: + return TypeTraits::pack(in); + case 2: + return TypeTraits::pack(in); + case 3: + return TypeTraits::pack(in); + } + SWR_ASSERT(0); + return TypeTraits::pack(in); + } + + INLINE static simdscalar convertSrgb(uint32_t comp, simdscalar &in) + { + switch (comp) + { + case 0: + return TypeTraits::convertSrgb(in);; + case 1: + return TypeTraits::convertSrgb(in);; + case 2: + return TypeTraits::convertSrgb(in);; + case 3: + return TypeTraits::convertSrgb(in);; + } + SWR_ASSERT(0); + return TypeTraits::convertSrgb(in); + } +}; diff --git a/src/gallium/drivers/swr/rasterizer/core/frontend.cpp b/src/gallium/drivers/swr/rasterizer/core/frontend.cpp new file mode 100644 index 00000000000..f43a672bd82 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/frontend.cpp @@ -0,0 +1,2345 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file frontend.cpp +* +* @brief Implementation for Frontend which handles vertex processing, +* primitive assembly, clipping, binning, etc. +* +******************************************************************************/ + +#include "api.h" +#include "frontend.h" +#include "backend.h" +#include "context.h" +#include "rdtsc_core.h" +#include "rasterizer.h" +#include "utils.h" +#include "threads.h" +#include "pa.h" +#include "clip.h" +#include "tilemgr.h" +#include "tessellator.h" + +////////////////////////////////////////////////////////////////////////// +/// @brief Helper macro to generate a bitmask +static INLINE uint32_t GenMask(uint32_t numBits) +{ + SWR_ASSERT(numBits <= (sizeof(uint32_t) * 8), "Too many bits (%d) for %s", numBits, __FUNCTION__); + return ((1U << numBits) - 1); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Offsets added to post-viewport vertex positions based on +/// raster state. +static const simdscalar g_pixelOffsets[SWR_PIXEL_LOCATION_UL + 1] = +{ + _simd_set1_ps(0.0f), // SWR_PIXEL_LOCATION_CENTER + _simd_set1_ps(0.5f), // SWR_PIXEL_LOCATION_UL +}; + +////////////////////////////////////////////////////////////////////////// +/// @brief FE handler for SwrSync. +/// @param pContext - pointer to SWR context. +/// @param pDC - pointer to draw context. +/// @param workerId - thread's worker id. Even thread has a unique id. +/// @param pUserData - Pointer to user data passed back to sync callback. +/// @todo This should go away when we switch this to use compute threading. +void ProcessSync( + SWR_CONTEXT *pContext, + DRAW_CONTEXT *pDC, + uint32_t workerId, + void *pUserData) +{ + SYNC_DESC *pSync = (SYNC_DESC*)pUserData; + BE_WORK work; + work.type = SYNC; + work.pfnWork = ProcessSyncBE; + work.desc.sync = *pSync; + + MacroTileMgr *pTileMgr = pDC->pTileMgr; + pTileMgr->enqueue(0, 0, &work); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief FE handler for SwrGetStats. +/// @param pContext - pointer to SWR context. +/// @param pDC - pointer to draw context. +/// @param workerId - thread's worker id. Even thread has a unique id. +/// @param pUserData - Pointer to user data passed back to stats callback. +/// @todo This should go away when we switch this to use compute threading. +void ProcessQueryStats( + SWR_CONTEXT *pContext, + DRAW_CONTEXT *pDC, + uint32_t workerId, + void *pUserData) +{ + QUERY_DESC *pQueryStats = (QUERY_DESC*)pUserData; + BE_WORK work; + work.type = QUERYSTATS; + work.pfnWork = ProcessQueryStatsBE; + work.desc.queryStats = *pQueryStats; + + MacroTileMgr *pTileMgr = pDC->pTileMgr; + pTileMgr->enqueue(0, 0, &work); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief FE handler for SwrClearRenderTarget. +/// @param pContext - pointer to SWR context. +/// @param pDC - pointer to draw context. +/// @param workerId - thread's worker id. Even thread has a unique id. +/// @param pUserData - Pointer to user data passed back to clear callback. +/// @todo This should go away when we switch this to use compute threading. +void ProcessClear( + SWR_CONTEXT *pContext, + DRAW_CONTEXT *pDC, + uint32_t workerId, + void *pUserData) +{ + CLEAR_DESC *pClear = (CLEAR_DESC*)pUserData; + MacroTileMgr *pTileMgr = pDC->pTileMgr; + + const API_STATE& state = GetApiState(pDC); + + // queue a clear to each macro tile + // compute macro tile bounds for the current scissor/viewport + uint32_t macroTileLeft = state.scissorInFixedPoint.left / KNOB_MACROTILE_X_DIM_FIXED; + uint32_t macroTileRight = state.scissorInFixedPoint.right / KNOB_MACROTILE_X_DIM_FIXED; + uint32_t macroTileTop = state.scissorInFixedPoint.top / KNOB_MACROTILE_Y_DIM_FIXED; + uint32_t macroTileBottom = state.scissorInFixedPoint.bottom / KNOB_MACROTILE_Y_DIM_FIXED; + + BE_WORK work; + work.type = CLEAR; + work.pfnWork = ProcessClearBE; + work.desc.clear = *pClear; + + for (uint32_t y = macroTileTop; y <= macroTileBottom; ++y) + { + for (uint32_t x = macroTileLeft; x <= macroTileRight; ++x) + { + pTileMgr->enqueue(x, y, &work); + } + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief FE handler for SwrStoreTiles. +/// @param pContext - pointer to SWR context. +/// @param pDC - pointer to draw context. +/// @param workerId - thread's worker id. Even thread has a unique id. +/// @param pUserData - Pointer to user data passed back to callback. +/// @todo This should go away when we switch this to use compute threading. +void ProcessStoreTiles( + SWR_CONTEXT *pContext, + DRAW_CONTEXT *pDC, + uint32_t workerId, + void *pUserData) +{ + RDTSC_START(FEProcessStoreTiles); + STORE_TILES_DESC *pStore = (STORE_TILES_DESC*)pUserData; + MacroTileMgr *pTileMgr = pDC->pTileMgr; + + const API_STATE& state = GetApiState(pDC); + + // queue a store to each macro tile + // compute macro tile bounds for the current render target + const uint32_t macroWidth = KNOB_MACROTILE_X_DIM; + const uint32_t macroHeight = KNOB_MACROTILE_Y_DIM; + + uint32_t numMacroTilesX = ((uint32_t)state.vp[0].width + (uint32_t)state.vp[0].x + (macroWidth - 1)) / macroWidth; + uint32_t numMacroTilesY = ((uint32_t)state.vp[0].height + (uint32_t)state.vp[0].y + (macroHeight - 1)) / macroHeight; + + // store tiles + BE_WORK work; + work.type = STORETILES; + work.pfnWork = ProcessStoreTileBE; + work.desc.storeTiles = *pStore; + + for (uint32_t x = 0; x < numMacroTilesX; ++x) + { + for (uint32_t y = 0; y < numMacroTilesY; ++y) + { + pTileMgr->enqueue(x, y, &work); + } + } + + RDTSC_STOP(FEProcessStoreTiles, 0, pDC->drawId); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief FE handler for SwrInvalidateTiles. +/// @param pContext - pointer to SWR context. +/// @param pDC - pointer to draw context. +/// @param workerId - thread's worker id. Even thread has a unique id. +/// @param pUserData - Pointer to user data passed back to callback. +/// @todo This should go away when we switch this to use compute threading. +void ProcessInvalidateTiles( + SWR_CONTEXT *pContext, + DRAW_CONTEXT *pDC, + uint32_t workerId, + void *pUserData) +{ + RDTSC_START(FEProcessInvalidateTiles); + INVALIDATE_TILES_DESC *pInv = (INVALIDATE_TILES_DESC*)pUserData; + MacroTileMgr *pTileMgr = pDC->pTileMgr; + + const API_STATE& state = GetApiState(pDC); + + // queue a store to each macro tile + // compute macro tile bounds for the current render target + uint32_t macroWidth = KNOB_MACROTILE_X_DIM; + uint32_t macroHeight = KNOB_MACROTILE_Y_DIM; + + uint32_t numMacroTilesX = ((uint32_t)state.vp[0].width + (uint32_t)state.vp[0].x + (macroWidth - 1)) / macroWidth; + uint32_t numMacroTilesY = ((uint32_t)state.vp[0].height + (uint32_t)state.vp[0].y + (macroHeight - 1)) / macroHeight; + + // load tiles + BE_WORK work; + work.type = INVALIDATETILES; + work.pfnWork = ProcessInvalidateTilesBE; + work.desc.invalidateTiles = *pInv; + + for (uint32_t x = 0; x < numMacroTilesX; ++x) + { + for (uint32_t y = 0; y < numMacroTilesY; ++y) + { + pTileMgr->enqueue(x, y, &work); + } + } + + RDTSC_STOP(FEProcessInvalidateTiles, 0, pDC->drawId); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes the number of primitives given the number of verts. +/// @param mode - primitive topology for draw operation. +/// @param numPrims - number of vertices or indices for draw. +/// @todo Frontend needs to be refactored. This will go in appropriate place then. +uint32_t GetNumPrims( + PRIMITIVE_TOPOLOGY mode, + uint32_t numPrims) +{ + switch (mode) + { + case TOP_POINT_LIST: return numPrims; + case TOP_TRIANGLE_LIST: return numPrims / 3; + case TOP_TRIANGLE_STRIP: return numPrims < 3 ? 0 : numPrims - 2; + case TOP_TRIANGLE_FAN: return numPrims < 3 ? 0 : numPrims - 2; + case TOP_TRIANGLE_DISC: return numPrims < 2 ? 0 : numPrims - 1; + case TOP_QUAD_LIST: return numPrims / 4; + case TOP_QUAD_STRIP: return numPrims < 4 ? 0 : (numPrims - 2) / 2; + case TOP_LINE_STRIP: return numPrims < 2 ? 0 : numPrims - 1; + case TOP_LINE_LIST: return numPrims / 2; + case TOP_LINE_LOOP: return numPrims; + case TOP_RECT_LIST: return numPrims / 3; + case TOP_LINE_LIST_ADJ: return numPrims / 4; + case TOP_LISTSTRIP_ADJ: return numPrims < 3 ? 0 : numPrims - 3; + case TOP_TRI_LIST_ADJ: return numPrims / 6; + case TOP_TRI_STRIP_ADJ: return numPrims < 4 ? 0 : (numPrims / 2) - 2; + + case TOP_PATCHLIST_1: + case TOP_PATCHLIST_2: + case TOP_PATCHLIST_3: + case TOP_PATCHLIST_4: + case TOP_PATCHLIST_5: + case TOP_PATCHLIST_6: + case TOP_PATCHLIST_7: + case TOP_PATCHLIST_8: + case TOP_PATCHLIST_9: + case TOP_PATCHLIST_10: + case TOP_PATCHLIST_11: + case TOP_PATCHLIST_12: + case TOP_PATCHLIST_13: + case TOP_PATCHLIST_14: + case TOP_PATCHLIST_15: + case TOP_PATCHLIST_16: + case TOP_PATCHLIST_17: + case TOP_PATCHLIST_18: + case TOP_PATCHLIST_19: + case TOP_PATCHLIST_20: + case TOP_PATCHLIST_21: + case TOP_PATCHLIST_22: + case TOP_PATCHLIST_23: + case TOP_PATCHLIST_24: + case TOP_PATCHLIST_25: + case TOP_PATCHLIST_26: + case TOP_PATCHLIST_27: + case TOP_PATCHLIST_28: + case TOP_PATCHLIST_29: + case TOP_PATCHLIST_30: + case TOP_PATCHLIST_31: + case TOP_PATCHLIST_32: + return numPrims / (mode - TOP_PATCHLIST_BASE); + + case TOP_POLYGON: + case TOP_POINT_LIST_BF: + case TOP_LINE_STRIP_CONT: + case TOP_LINE_STRIP_BF: + case TOP_LINE_STRIP_CONT_BF: + case TOP_TRIANGLE_FAN_NOSTIPPLE: + case TOP_TRI_STRIP_REVERSE: + case TOP_PATCHLIST_BASE: + case TOP_UNKNOWN: + SWR_ASSERT(false, "Unsupported topology: %d", mode); + return 0; + } + + return 0; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes the number of verts given the number of primitives. +/// @param mode - primitive topology for draw operation. +/// @param numPrims - number of primitives for draw. +uint32_t GetNumVerts( + PRIMITIVE_TOPOLOGY mode, + uint32_t numPrims) +{ + switch (mode) + { + case TOP_POINT_LIST: return numPrims; + case TOP_TRIANGLE_LIST: return numPrims * 3; + case TOP_TRIANGLE_STRIP: return numPrims ? numPrims + 2 : 0; + case TOP_TRIANGLE_FAN: return numPrims ? numPrims + 2 : 0; + case TOP_TRIANGLE_DISC: return numPrims ? numPrims + 1 : 0; + case TOP_QUAD_LIST: return numPrims * 4; + case TOP_QUAD_STRIP: return numPrims ? numPrims * 2 + 2 : 0; + case TOP_LINE_STRIP: return numPrims ? numPrims + 1 : 0; + case TOP_LINE_LIST: return numPrims * 2; + case TOP_LINE_LOOP: return numPrims; + case TOP_RECT_LIST: return numPrims * 3; + case TOP_LINE_LIST_ADJ: return numPrims * 4; + case TOP_LISTSTRIP_ADJ: return numPrims ? numPrims + 3 : 0; + case TOP_TRI_LIST_ADJ: return numPrims * 6; + case TOP_TRI_STRIP_ADJ: return numPrims ? (numPrims + 2) * 2 : 0; + + case TOP_PATCHLIST_1: + case TOP_PATCHLIST_2: + case TOP_PATCHLIST_3: + case TOP_PATCHLIST_4: + case TOP_PATCHLIST_5: + case TOP_PATCHLIST_6: + case TOP_PATCHLIST_7: + case TOP_PATCHLIST_8: + case TOP_PATCHLIST_9: + case TOP_PATCHLIST_10: + case TOP_PATCHLIST_11: + case TOP_PATCHLIST_12: + case TOP_PATCHLIST_13: + case TOP_PATCHLIST_14: + case TOP_PATCHLIST_15: + case TOP_PATCHLIST_16: + case TOP_PATCHLIST_17: + case TOP_PATCHLIST_18: + case TOP_PATCHLIST_19: + case TOP_PATCHLIST_20: + case TOP_PATCHLIST_21: + case TOP_PATCHLIST_22: + case TOP_PATCHLIST_23: + case TOP_PATCHLIST_24: + case TOP_PATCHLIST_25: + case TOP_PATCHLIST_26: + case TOP_PATCHLIST_27: + case TOP_PATCHLIST_28: + case TOP_PATCHLIST_29: + case TOP_PATCHLIST_30: + case TOP_PATCHLIST_31: + case TOP_PATCHLIST_32: + return numPrims * (mode - TOP_PATCHLIST_BASE); + + case TOP_POLYGON: + case TOP_POINT_LIST_BF: + case TOP_LINE_STRIP_CONT: + case TOP_LINE_STRIP_BF: + case TOP_LINE_STRIP_CONT_BF: + case TOP_TRIANGLE_FAN_NOSTIPPLE: + case TOP_TRI_STRIP_REVERSE: + case TOP_PATCHLIST_BASE: + case TOP_UNKNOWN: + SWR_ASSERT(false, "Unsupported topology: %d", mode); + return 0; + } + + return 0; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Return number of verts per primitive. +/// @param topology - topology +/// @param includeAdjVerts - include adjacent verts in primitive vertices +INLINE uint32_t NumVertsPerPrim(PRIMITIVE_TOPOLOGY topology, bool includeAdjVerts) +{ + uint32_t numVerts = 0; + switch (topology) + { + case TOP_POINT_LIST: + case TOP_POINT_LIST_BF: + numVerts = 1; + break; + case TOP_LINE_LIST: + case TOP_LINE_STRIP: + case TOP_LINE_LIST_ADJ: + case TOP_LINE_LOOP: + case TOP_LINE_STRIP_CONT: + case TOP_LINE_STRIP_BF: + case TOP_LISTSTRIP_ADJ: + numVerts = 2; + break; + case TOP_TRIANGLE_LIST: + case TOP_TRIANGLE_STRIP: + case TOP_TRIANGLE_FAN: + case TOP_TRI_LIST_ADJ: + case TOP_TRI_STRIP_ADJ: + case TOP_TRI_STRIP_REVERSE: + case TOP_RECT_LIST: + numVerts = 3; + break; + case TOP_QUAD_LIST: + case TOP_QUAD_STRIP: + numVerts = 4; + break; + case TOP_PATCHLIST_1: + case TOP_PATCHLIST_2: + case TOP_PATCHLIST_3: + case TOP_PATCHLIST_4: + case TOP_PATCHLIST_5: + case TOP_PATCHLIST_6: + case TOP_PATCHLIST_7: + case TOP_PATCHLIST_8: + case TOP_PATCHLIST_9: + case TOP_PATCHLIST_10: + case TOP_PATCHLIST_11: + case TOP_PATCHLIST_12: + case TOP_PATCHLIST_13: + case TOP_PATCHLIST_14: + case TOP_PATCHLIST_15: + case TOP_PATCHLIST_16: + case TOP_PATCHLIST_17: + case TOP_PATCHLIST_18: + case TOP_PATCHLIST_19: + case TOP_PATCHLIST_20: + case TOP_PATCHLIST_21: + case TOP_PATCHLIST_22: + case TOP_PATCHLIST_23: + case TOP_PATCHLIST_24: + case TOP_PATCHLIST_25: + case TOP_PATCHLIST_26: + case TOP_PATCHLIST_27: + case TOP_PATCHLIST_28: + case TOP_PATCHLIST_29: + case TOP_PATCHLIST_30: + case TOP_PATCHLIST_31: + case TOP_PATCHLIST_32: + numVerts = topology - TOP_PATCHLIST_BASE; + break; + default: + SWR_ASSERT(false, "Unsupported topology: %d", topology); + break; + } + + if (includeAdjVerts) + { + switch (topology) + { + case TOP_LISTSTRIP_ADJ: + case TOP_LINE_LIST_ADJ: numVerts = 4; break; + case TOP_TRI_STRIP_ADJ: + case TOP_TRI_LIST_ADJ: numVerts = 6; break; + default: break; + } + } + + return numVerts; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Generate mask from remaining work. +/// @param numWorkItems - Number of items being worked on by a SIMD. +static INLINE simdscalari GenerateMask(uint32_t numItemsRemaining) +{ + uint32_t numActive = (numItemsRemaining >= KNOB_SIMD_WIDTH) ? KNOB_SIMD_WIDTH : numItemsRemaining; + uint32_t mask = (numActive > 0) ? ((1 << numActive) - 1) : 0; + return _simd_castps_si(vMask(mask)); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief StreamOut - Streams vertex data out to SO buffers. +/// Generally, we are only streaming out a SIMDs worth of triangles. +/// @param pDC - pointer to draw context. +/// @param workerId - thread's worker id. Even thread has a unique id. +/// @param numPrims - Number of prims to streamout (e.g. points, lines, tris) +static void StreamOut( + DRAW_CONTEXT* pDC, + PA_STATE& pa, + uint32_t workerId, + uint32_t* pPrimData, + uint32_t streamIndex) +{ + RDTSC_START(FEStreamout); + + SWR_CONTEXT* pContext = pDC->pContext; + + const API_STATE& state = GetApiState(pDC); + const SWR_STREAMOUT_STATE &soState = state.soState; + + uint32_t soVertsPerPrim = NumVertsPerPrim(pa.binTopology, false); + + // The pPrimData buffer is sparse in that we allocate memory for all 32 attributes for each vertex. + uint32_t primDataDwordVertexStride = (KNOB_NUM_ATTRIBUTES * sizeof(float) * 4) / sizeof(uint32_t); + + SWR_STREAMOUT_CONTEXT soContext = { 0 }; + + // Setup buffer state pointers. + for (uint32_t i = 0; i < 4; ++i) + { + soContext.pBuffer[i] = &state.soBuffer[i]; + } + + uint32_t numPrims = pa.NumPrims(); + for (uint32_t primIndex = 0; primIndex < numPrims; ++primIndex) + { + DWORD slot = 0; + uint32_t soMask = soState.streamMasks[streamIndex]; + + // Write all entries into primitive data buffer for SOS. + while (_BitScanForward(&slot, soMask)) + { + __m128 attrib[MAX_NUM_VERTS_PER_PRIM]; // prim attribs (always 4 wide) + uint32_t paSlot = slot + VERTEX_ATTRIB_START_SLOT; + pa.AssembleSingle(paSlot, primIndex, attrib); + + // Attribute offset is relative offset from start of vertex. + // Note that attributes start at slot 1 in the PA buffer. We need to write this + // to prim data starting at slot 0. Which is why we do (slot - 1). + // Also note: GL works slightly differently, and needs slot 0 + uint32_t primDataAttribOffset = slot * sizeof(float) * 4 / sizeof(uint32_t); + + // Store each vertex's attrib at appropriate locations in pPrimData buffer. + for (uint32_t v = 0; v < soVertsPerPrim; ++v) + { + uint32_t* pPrimDataAttrib = pPrimData + primDataAttribOffset + (v * primDataDwordVertexStride); + + _mm_store_ps((float*)pPrimDataAttrib, attrib[v]); + } + soMask &= ~(1 << slot); + } + + // Update pPrimData pointer + soContext.pPrimData = pPrimData; + + // Call SOS + SWR_ASSERT(state.pfnSoFunc[streamIndex] != nullptr, "Trying to execute uninitialized streamout jit function."); + state.pfnSoFunc[streamIndex](soContext); + } + + // Update SO write offset. The driver provides memory for the update. + for (uint32_t i = 0; i < 4; ++i) + { + if (state.soBuffer[i].pWriteOffset) + { + *state.soBuffer[i].pWriteOffset = soContext.pBuffer[i]->streamOffset * sizeof(uint32_t); + + // The SOS increments the existing write offset. So we don't want to increment + // the SoWriteOffset stat using an absolute offset instead of relative. + SET_STAT(SoWriteOffset[i], soContext.pBuffer[i]->streamOffset); + } + } + + UPDATE_STAT(SoPrimStorageNeeded[streamIndex], soContext.numPrimStorageNeeded); + UPDATE_STAT(SoNumPrimsWritten[streamIndex], soContext.numPrimsWritten); + + RDTSC_STOP(FEStreamout, 1, 0); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes number of invocations. The current index represents +/// the start of the SIMD. The max index represents how much work +/// items are remaining. If there is less then a SIMD's left of work +/// then return the remaining amount of work. +/// @param curIndex - The start index for the SIMD. +/// @param maxIndex - The last index for all work items. +static INLINE uint32_t GetNumInvocations( + uint32_t curIndex, + uint32_t maxIndex) +{ + uint32_t remainder = (maxIndex - curIndex); + return (remainder >= KNOB_SIMD_WIDTH) ? KNOB_SIMD_WIDTH : remainder; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Converts a streamId buffer to a cut buffer for the given stream id. +/// The geometry shader will loop over each active streamout buffer, assembling +/// primitives for the downstream stages. When multistream output is enabled, +/// the generated stream ID buffer from the GS needs to be converted to a cut +/// buffer for the primitive assembler. +/// @param stream - stream id to generate the cut buffer for +/// @param pStreamIdBase - pointer to the stream ID buffer +/// @param numEmittedVerts - Number of total verts emitted by the GS +/// @param pCutBuffer - output buffer to write cuts to +void ProcessStreamIdBuffer(uint32_t stream, uint8_t* pStreamIdBase, uint32_t numEmittedVerts, uint8_t *pCutBuffer) +{ + SWR_ASSERT(stream < MAX_SO_STREAMS); + + uint32_t numInputBytes = (numEmittedVerts * 2 + 7) / 8; + uint32_t numOutputBytes = std::max(numInputBytes / 2, 1U); + + for (uint32_t b = 0; b < numOutputBytes; ++b) + { + uint8_t curInputByte = pStreamIdBase[2*b]; + uint8_t outByte = 0; + for (uint32_t i = 0; i < 4; ++i) + { + if ((curInputByte & 0x3) != stream) + { + outByte |= (1 << i); + } + curInputByte >>= 2; + } + + curInputByte = pStreamIdBase[2 * b + 1]; + for (uint32_t i = 0; i < 4; ++i) + { + if ((curInputByte & 0x3) != stream) + { + outByte |= (1 << (i + 4)); + } + curInputByte >>= 2; + } + + *pCutBuffer++ = outByte; + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Implements GS stage. +/// @param pDC - pointer to draw context. +/// @param workerId - thread's worker id. Even thread has a unique id. +/// @param pa - The primitive assembly object. +/// @param pGsOut - output stream for GS +template < + bool HasStreamOutT, + bool HasRastT> +static void GeometryShaderStage( + DRAW_CONTEXT *pDC, + uint32_t workerId, + PA_STATE& pa, + void* pGsOut, + void* pCutBuffer, + void* pStreamCutBuffer, + uint32_t* pSoPrimData, + simdscalari primID) +{ + RDTSC_START(FEGeometryShader); + + SWR_GS_CONTEXT gsContext; + SWR_CONTEXT* pContext = pDC->pContext; + + const API_STATE& state = GetApiState(pDC); + const SWR_GS_STATE* pState = &state.gsState; + + SWR_ASSERT(pGsOut != nullptr, "GS output buffer should be initialized"); + SWR_ASSERT(pCutBuffer != nullptr, "GS output cut buffer should be initialized"); + + gsContext.pStream = (uint8_t*)pGsOut; + gsContext.pCutOrStreamIdBuffer = (uint8_t*)pCutBuffer; + gsContext.PrimitiveID = primID; + + uint32_t numVertsPerPrim = NumVertsPerPrim(pa.binTopology, true); + simdvector attrib[MAX_ATTRIBUTES]; + + // assemble all attributes for the input primitive + for (uint32_t slot = 0; slot < pState->numInputAttribs; ++slot) + { + uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + slot; + pa.Assemble(attribSlot, attrib); + + for (uint32_t i = 0; i < numVertsPerPrim; ++i) + { + gsContext.vert[i].attrib[attribSlot] = attrib[i]; + } + } + + // assemble position + pa.Assemble(VERTEX_POSITION_SLOT, attrib); + for (uint32_t i = 0; i < numVertsPerPrim; ++i) + { + gsContext.vert[i].attrib[VERTEX_POSITION_SLOT] = attrib[i]; + } + + const uint32_t vertexStride = sizeof(simdvertex); + const uint32_t numSimdBatches = (state.gsState.maxNumVerts + KNOB_SIMD_WIDTH - 1) / KNOB_SIMD_WIDTH; + const uint32_t inputPrimStride = numSimdBatches * vertexStride; + const uint32_t instanceStride = inputPrimStride * KNOB_SIMD_WIDTH; + uint32_t cutPrimStride; + uint32_t cutInstanceStride; + + if (pState->isSingleStream) + { + cutPrimStride = (state.gsState.maxNumVerts + 7) / 8; + cutInstanceStride = cutPrimStride * KNOB_SIMD_WIDTH; + } + else + { + cutPrimStride = AlignUp(state.gsState.maxNumVerts * 2 / 8, 4); + cutInstanceStride = cutPrimStride * KNOB_SIMD_WIDTH; + } + + // record valid prims from the frontend to avoid over binning the newly generated + // prims from the GS + uint32_t numInputPrims = pa.NumPrims(); + + for (uint32_t instance = 0; instance < pState->instanceCount; ++instance) + { + gsContext.InstanceID = instance; + gsContext.mask = GenerateMask(numInputPrims); + + // execute the geometry shader + state.pfnGsFunc(GetPrivateState(pDC), &gsContext); + + gsContext.pStream += instanceStride; + gsContext.pCutOrStreamIdBuffer += cutInstanceStride; + } + + // set up new binner and state for the GS output topology + PFN_PROCESS_PRIMS pfnClipFunc = nullptr; + if (HasRastT) + { + switch (pState->outputTopology) + { + case TOP_TRIANGLE_STRIP: pfnClipFunc = ClipTriangles; break; + case TOP_LINE_STRIP: pfnClipFunc = ClipLines; break; + case TOP_POINT_LIST: pfnClipFunc = ClipPoints; break; + default: SWR_ASSERT(false, "Unexpected GS output topology: %d", pState->outputTopology); + } + } + + // foreach input prim: + // - setup a new PA based on the emitted verts for that prim + // - loop over the new verts, calling PA to assemble each prim + uint32_t* pVertexCount = (uint32_t*)&gsContext.vertexCount; + uint32_t* pPrimitiveId = (uint32_t*)&primID; + + uint32_t totalPrimsGenerated = 0; + for (uint32_t inputPrim = 0; inputPrim < numInputPrims; ++inputPrim) + { + uint8_t* pInstanceBase = (uint8_t*)pGsOut + inputPrim * inputPrimStride; + uint8_t* pCutBufferBase = (uint8_t*)pCutBuffer + inputPrim * cutPrimStride; + for (uint32_t instance = 0; instance < pState->instanceCount; ++instance) + { + uint32_t numEmittedVerts = pVertexCount[inputPrim]; + if (numEmittedVerts == 0) + { + continue; + } + + uint8_t* pBase = pInstanceBase + instance * instanceStride; + uint8_t* pCutBase = pCutBufferBase + instance * cutInstanceStride; + + DWORD numAttribs; + _BitScanReverse(&numAttribs, state.feAttribMask); + numAttribs++; + + for (uint32_t stream = 0; stream < MAX_SO_STREAMS; ++stream) + { + bool processCutVerts = false; + + uint8_t* pCutBuffer = pCutBase; + + // assign default stream ID, only relevant when GS is outputting a single stream + uint32_t streamID = 0; + if (pState->isSingleStream) + { + processCutVerts = true; + streamID = pState->singleStreamID; + if (streamID != stream) continue; + } + else + { + // early exit if this stream is not enabled for streamout + if (HasStreamOutT && !state.soState.streamEnable[stream]) + { + continue; + } + + // multi-stream output, need to translate StreamID buffer to a cut buffer + ProcessStreamIdBuffer(stream, pCutBase, numEmittedVerts, (uint8_t*)pStreamCutBuffer); + pCutBuffer = (uint8_t*)pStreamCutBuffer; + processCutVerts = false; + } + + PA_STATE_CUT gsPa(pDC, pBase, numEmittedVerts, pCutBuffer, numEmittedVerts, numAttribs, pState->outputTopology, processCutVerts); + + while (gsPa.GetNextStreamOutput()) + { + do + { + bool assemble = gsPa.Assemble(VERTEX_POSITION_SLOT, attrib); + + if (assemble) + { + totalPrimsGenerated += gsPa.NumPrims(); + + if (HasStreamOutT) + { + StreamOut(pDC, gsPa, workerId, pSoPrimData, stream); + } + + if (HasRastT && state.soState.streamToRasterizer == stream) + { + simdscalari vPrimId; + // pull primitiveID from the GS output if available + if (state.gsState.emitsPrimitiveID) + { + simdvector primIdAttrib[3]; + gsPa.Assemble(VERTEX_PRIMID_SLOT, primIdAttrib); + vPrimId = _simd_castps_si(primIdAttrib[0].x); + } + else + { + vPrimId = _simd_set1_epi32(pPrimitiveId[inputPrim]); + } + + pfnClipFunc(pDC, gsPa, workerId, attrib, GenMask(gsPa.NumPrims()), vPrimId); + } + } + } while (gsPa.NextPrim()); + } + } + } + } + + // update GS pipeline stats + UPDATE_STAT(GsInvocations, numInputPrims * pState->instanceCount); + UPDATE_STAT(GsPrimitives, totalPrimsGenerated); + + RDTSC_STOP(FEGeometryShader, 1, 0); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Allocate GS buffers +/// @param pDC - pointer to draw context. +/// @param state - API state +/// @param ppGsOut - pointer to GS output buffer allocation +/// @param ppCutBuffer - pointer to GS output cut buffer allocation +static INLINE void AllocateGsBuffers(DRAW_CONTEXT* pDC, const API_STATE& state, void** ppGsOut, void** ppCutBuffer, + void **ppStreamCutBuffer) +{ + Arena* pArena = pDC->pArena; + SWR_ASSERT(pArena != nullptr); + SWR_ASSERT(state.gsState.gsEnable); + // allocate arena space to hold GS output verts + // @todo pack attribs + // @todo support multiple streams + const uint32_t vertexStride = sizeof(simdvertex); + const uint32_t numSimdBatches = (state.gsState.maxNumVerts + KNOB_SIMD_WIDTH - 1) / KNOB_SIMD_WIDTH; + uint32_t size = state.gsState.instanceCount * numSimdBatches * vertexStride * KNOB_SIMD_WIDTH; + *ppGsOut = pArena->AllocAligned(size, KNOB_SIMD_WIDTH * sizeof(float)); + + const uint32_t cutPrimStride = (state.gsState.maxNumVerts + 7) / 8; + const uint32_t streamIdPrimStride = AlignUp(state.gsState.maxNumVerts * 2 / 8, 4); + const uint32_t cutBufferSize = cutPrimStride * state.gsState.instanceCount * KNOB_SIMD_WIDTH; + const uint32_t streamIdSize = streamIdPrimStride * state.gsState.instanceCount * KNOB_SIMD_WIDTH; + + // allocate arena space to hold cut or streamid buffer, which is essentially a bitfield sized to the + // maximum vertex output as defined by the GS state, per SIMD lane, per GS instance + + // allocate space for temporary per-stream cut buffer if multi-stream is enabled + if (state.gsState.isSingleStream) + { + *ppCutBuffer = pArena->AllocAligned(cutBufferSize, KNOB_SIMD_WIDTH * sizeof(float)); + *ppStreamCutBuffer = nullptr; + } + else + { + *ppCutBuffer = pArena->AllocAligned(streamIdSize, KNOB_SIMD_WIDTH * sizeof(float)); + *ppStreamCutBuffer = pArena->AllocAligned(cutBufferSize, KNOB_SIMD_WIDTH * sizeof(float)); + } + +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Contains all data generated by the HS and passed to the +/// tessellator and DS. +struct TessellationThreadLocalData +{ + SWR_HS_CONTEXT hsContext; + ScalarPatch patchData[KNOB_SIMD_WIDTH]; + void* pTxCtx; + size_t tsCtxSize; + + simdscalar* pDSOutput; + size_t numDSOutputVectors; +}; + +THREAD TessellationThreadLocalData* gt_pTessellationThreadData = nullptr; + +////////////////////////////////////////////////////////////////////////// +/// @brief Allocate tessellation data for this worker thread. +INLINE +static void AllocateTessellationData(SWR_CONTEXT* pContext) +{ + /// @TODO - Don't use thread local storage. Use Worker local storage instead. + if (gt_pTessellationThreadData == nullptr) + { + gt_pTessellationThreadData = (TessellationThreadLocalData*) + _aligned_malloc(sizeof(TessellationThreadLocalData), 64); + memset(gt_pTessellationThreadData, 0, sizeof(*gt_pTessellationThreadData)); + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Implements Tessellation Stages. +/// @param pDC - pointer to draw context. +/// @param workerId - thread's worker id. Even thread has a unique id. +/// @param pa - The primitive assembly object. +/// @param pGsOut - output stream for GS +template < + bool HasGeometryShaderT, + bool HasStreamOutT, + bool HasRastT> +static void TessellationStages( + DRAW_CONTEXT *pDC, + uint32_t workerId, + PA_STATE& pa, + void* pGsOut, + void* pCutBuffer, + void* pCutStreamBuffer, + uint32_t* pSoPrimData, + simdscalari primID) +{ + const API_STATE& state = GetApiState(pDC); + const SWR_TS_STATE& tsState = state.tsState; + SWR_CONTEXT *pContext = pDC->pContext; // Needed for UPDATE_STATS macro + + SWR_ASSERT(gt_pTessellationThreadData); + + HANDLE tsCtx = TSInitCtx( + tsState.domain, + tsState.partitioning, + tsState.tsOutputTopology, + gt_pTessellationThreadData->pTxCtx, + gt_pTessellationThreadData->tsCtxSize); + if (tsCtx == nullptr) + { + gt_pTessellationThreadData->pTxCtx = _aligned_malloc(gt_pTessellationThreadData->tsCtxSize, 64); + tsCtx = TSInitCtx( + tsState.domain, + tsState.partitioning, + tsState.tsOutputTopology, + gt_pTessellationThreadData->pTxCtx, + gt_pTessellationThreadData->tsCtxSize); + } + SWR_ASSERT(tsCtx); + + PFN_PROCESS_PRIMS pfnClipFunc = nullptr; + if (HasRastT) + { + switch (tsState.postDSTopology) + { + case TOP_TRIANGLE_LIST: pfnClipFunc = ClipTriangles; break; + case TOP_LINE_LIST: pfnClipFunc = ClipLines; break; + case TOP_POINT_LIST: pfnClipFunc = ClipPoints; break; + default: SWR_ASSERT(false, "Unexpected DS output topology: %d", tsState.postDSTopology); + } + } + + SWR_HS_CONTEXT& hsContext = gt_pTessellationThreadData->hsContext; + hsContext.pCPout = gt_pTessellationThreadData->patchData; + hsContext.PrimitiveID = primID; + + uint32_t numVertsPerPrim = NumVertsPerPrim(pa.binTopology, false); + // Max storage for one attribute for an entire simdprimitive + simdvector simdattrib[MAX_NUM_VERTS_PER_PRIM]; + + // assemble all attributes for the input primitives + for (uint32_t slot = 0; slot < tsState.numHsInputAttribs; ++slot) + { + uint32_t attribSlot = VERTEX_ATTRIB_START_SLOT + slot; + pa.Assemble(attribSlot, simdattrib); + + for (uint32_t i = 0; i < numVertsPerPrim; ++i) + { + hsContext.vert[i].attrib[attribSlot] = simdattrib[i]; + } + } + +#if defined(_DEBUG) + memset(hsContext.pCPout, 0x90, sizeof(ScalarPatch) * KNOB_SIMD_WIDTH); +#endif + + uint32_t numPrims = pa.NumPrims(); + hsContext.mask = GenerateMask(numPrims); + + // Run the HS + RDTSC_START(FEHullShader); + state.pfnHsFunc(GetPrivateState(pDC), &hsContext); + RDTSC_STOP(FEHullShader, 0, 0); + + UPDATE_STAT(HsInvocations, numPrims); + + const uint32_t* pPrimId = (const uint32_t*)&primID; + + for (uint32_t p = 0; p < numPrims; ++p) + { + // Run Tessellator + SWR_TS_TESSELLATED_DATA tsData = { 0 }; + RDTSC_START(FETessellation); + TSTessellate(tsCtx, hsContext.pCPout[p].tessFactors, tsData); + RDTSC_STOP(FETessellation, 0, 0); + + if (tsData.NumPrimitives == 0) + { + continue; + } + SWR_ASSERT(tsData.NumDomainPoints); + + // Allocate DS Output memory + uint32_t requiredDSVectorInvocations = AlignUp(tsData.NumDomainPoints, KNOB_SIMD_WIDTH) / KNOB_SIMD_WIDTH; + size_t requiredDSOutputVectors = requiredDSVectorInvocations * tsState.numDsOutputAttribs; + size_t requiredAllocSize = sizeof(simdvector) * requiredDSOutputVectors; + if (requiredDSOutputVectors > gt_pTessellationThreadData->numDSOutputVectors) + { + _aligned_free(gt_pTessellationThreadData->pDSOutput); + gt_pTessellationThreadData->pDSOutput = (simdscalar*)_aligned_malloc(requiredAllocSize, 64); + gt_pTessellationThreadData->numDSOutputVectors = requiredDSOutputVectors; + } + SWR_ASSERT(gt_pTessellationThreadData->pDSOutput); + SWR_ASSERT(gt_pTessellationThreadData->numDSOutputVectors >= requiredDSOutputVectors); + +#if defined(_DEBUG) + memset(gt_pTessellationThreadData->pDSOutput, 0x90, requiredAllocSize); +#endif + + // Run Domain Shader + SWR_DS_CONTEXT dsContext; + dsContext.PrimitiveID = pPrimId[p]; + dsContext.pCpIn = &hsContext.pCPout[p]; + dsContext.pDomainU = (simdscalar*)tsData.pDomainPointsU; + dsContext.pDomainV = (simdscalar*)tsData.pDomainPointsV; + dsContext.pOutputData = gt_pTessellationThreadData->pDSOutput; + dsContext.vectorStride = requiredDSVectorInvocations; + + uint32_t dsInvocations = 0; + + for (dsContext.vectorOffset = 0; dsContext.vectorOffset < requiredDSVectorInvocations; ++dsContext.vectorOffset) + { + dsContext.mask = GenerateMask(tsData.NumDomainPoints - dsInvocations); + + RDTSC_START(FEDomainShader); + state.pfnDsFunc(GetPrivateState(pDC), &dsContext); + RDTSC_STOP(FEDomainShader, 0, 0); + + dsInvocations += KNOB_SIMD_WIDTH; + } + UPDATE_STAT(DsInvocations, tsData.NumDomainPoints); + + PA_TESS tessPa( + pDC, + dsContext.pOutputData, + dsContext.vectorStride, + tsState.numDsOutputAttribs, + tsData.ppIndices, + tsData.NumPrimitives, + tsState.postDSTopology); + + while (tessPa.HasWork()) + { + if (HasGeometryShaderT) + { + GeometryShaderStage( + pDC, workerId, tessPa, pGsOut, pCutBuffer, pCutStreamBuffer, pSoPrimData, + _simd_set1_epi32(dsContext.PrimitiveID)); + } + else + { + if (HasStreamOutT) + { + StreamOut(pDC, tessPa, workerId, pSoPrimData, 0); + } + + if (HasRastT) + { + simdvector prim[3]; // Only deal with triangles, lines, or points + RDTSC_START(FEPAAssemble); +#if SWR_ENABLE_ASSERTS + bool assemble = +#endif + tessPa.Assemble(VERTEX_POSITION_SLOT, prim); + RDTSC_STOP(FEPAAssemble, 1, 0); + SWR_ASSERT(assemble); + + SWR_ASSERT(pfnClipFunc); + pfnClipFunc(pDC, tessPa, workerId, prim, + GenMask(tessPa.NumPrims()), _simd_set1_epi32(dsContext.PrimitiveID)); + } + } + + tessPa.NextPrim(); + + } // while (tessPa.HasWork()) + } // for (uint32_t p = 0; p < numPrims; ++p) + + TSDestroyCtx(tsCtx); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief FE handler for SwrDraw. +/// @tparam IsIndexedT - Is indexed drawing enabled +/// @tparam HasTessellationT - Is tessellation enabled +/// @tparam HasGeometryShaderT - Is the geometry shader stage enabled +/// @tparam HasStreamOutT - Is stream-out enabled +/// @tparam HasRastT - Is rasterization enabled +/// @param pContext - pointer to SWR context. +/// @param pDC - pointer to draw context. +/// @param workerId - thread's worker id. +/// @param pUserData - Pointer to DRAW_WORK +template < + bool IsIndexedT, + bool HasTessellationT, + bool HasGeometryShaderT, + bool HasStreamOutT, + bool HasRastT> +void ProcessDraw( + SWR_CONTEXT *pContext, + DRAW_CONTEXT *pDC, + uint32_t workerId, + void *pUserData) +{ + +#if KNOB_ENABLE_TOSS_POINTS + if (KNOB_TOSS_QUEUE_FE) + { + return; + } +#endif + + RDTSC_START(FEProcessDraw); + + DRAW_WORK& work = *(DRAW_WORK*)pUserData; + const API_STATE& state = GetApiState(pDC); + __m256i vScale = _mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0); + SWR_VS_CONTEXT vsContext; + simdvertex vin; + + int indexSize = 0; + uint32_t endVertex = work.numVerts; + + const int32_t* pLastRequestedIndex = nullptr; + if (IsIndexedT) + { + switch (work.type) + { + case R32_UINT: + indexSize = sizeof(uint32_t); + pLastRequestedIndex = &(work.pIB[endVertex]); + break; + case R16_UINT: + indexSize = sizeof(uint16_t); + // nasty address offset to last index + pLastRequestedIndex = (int32_t*)(&(((uint16_t*)work.pIB)[endVertex])); + break; + case R8_UINT: + indexSize = sizeof(uint8_t); + // nasty address offset to last index + pLastRequestedIndex = (int32_t*)(&(((uint8_t*)work.pIB)[endVertex])); + break; + default: + SWR_ASSERT(0); + } + } + else + { + // No cuts, prune partial primitives. + endVertex = GetNumVerts(state.topology, GetNumPrims(state.topology, work.numVerts)); + } + + SWR_FETCH_CONTEXT fetchInfo = { 0 }; + fetchInfo.pStreams = &state.vertexBuffers[0]; + fetchInfo.StartInstance = work.startInstance; + fetchInfo.StartVertex = 0; + + vsContext.pVin = &vin; + + if (IsIndexedT) + { + fetchInfo.BaseVertex = work.baseVertex; + + // if the entire index buffer isn't being consumed, set the last index + // so that fetches < a SIMD wide will be masked off + fetchInfo.pLastIndex = (const int32_t*)(((BYTE*)state.indexBuffer.pIndices) + state.indexBuffer.size); + if (pLastRequestedIndex < fetchInfo.pLastIndex) + { + fetchInfo.pLastIndex = pLastRequestedIndex; + } + } + else + { + fetchInfo.StartVertex = work.startVertex; + } + +#ifdef KNOB_ENABLE_RDTSC + uint32_t numPrims = GetNumPrims(state.topology, work.numVerts); +#endif + + void* pGsOut = nullptr; + void* pCutBuffer = nullptr; + void* pStreamCutBuffer = nullptr; + if (HasGeometryShaderT) + { + AllocateGsBuffers(pDC, state, &pGsOut, &pCutBuffer, &pStreamCutBuffer); + } + + if (HasTessellationT) + { + SWR_ASSERT(state.tsState.tsEnable == true); + SWR_ASSERT(state.pfnHsFunc != nullptr); + SWR_ASSERT(state.pfnDsFunc != nullptr); + + AllocateTessellationData(pContext); + } + else + { + SWR_ASSERT(state.tsState.tsEnable == false); + SWR_ASSERT(state.pfnHsFunc == nullptr); + SWR_ASSERT(state.pfnDsFunc == nullptr); + } + + // allocate space for streamout input prim data + uint32_t* pSoPrimData = nullptr; + if (HasStreamOutT) + { + pSoPrimData = (uint32_t*)pDC->pArena->AllocAligned(4096, 16); + + // update the + for (uint32_t i = 0; i < 4; ++i) + { + SET_STAT(SoWriteOffset[i], state.soBuffer[i].streamOffset); + } + + } + + // choose primitive assembler + PA_FACTORY paFactory(pDC, state.topology, work.numVerts); + PA_STATE& pa = paFactory.GetPA(); + + /// @todo: temporarily move instance loop in the FE to ensure SO ordering + for (uint32_t instanceNum = 0; instanceNum < work.numInstances; instanceNum++) + { + simdscalari vIndex; + uint32_t i = 0; + + if (IsIndexedT) + { + fetchInfo.pIndices = work.pIB; + } + else + { + vIndex = _simd_add_epi32(_simd_set1_epi32(work.startVertexID), vScale); + fetchInfo.pIndices = (const int32_t*)&vIndex; + } + + fetchInfo.CurInstance = instanceNum; + vsContext.InstanceID = instanceNum; + + while (pa.HasWork()) + { + // PaGetNextVsOutput currently has the side effect of updating some PA state machine state. + // So we need to keep this outside of (i < endVertex) check. + simdmask* pvCutIndices = nullptr; + if (IsIndexedT) + { + pvCutIndices = &pa.GetNextVsIndices(); + } + + simdvertex& vout = pa.GetNextVsOutput(); + vsContext.pVout = &vout; + + if (i < endVertex) + { + + // 1. Execute FS/VS for a single SIMD. + RDTSC_START(FEFetchShader); + state.pfnFetchFunc(fetchInfo, vin); + RDTSC_STOP(FEFetchShader, 0, 0); + + // forward fetch generated vertex IDs to the vertex shader + vsContext.VertexID = fetchInfo.VertexID; + + // Setup active mask for vertex shader. + vsContext.mask = GenerateMask(endVertex - i); + + // forward cut mask to the PA + if (IsIndexedT) + { + *pvCutIndices = _simd_movemask_ps(_simd_castsi_ps(fetchInfo.CutMask)); + } + + UPDATE_STAT(IaVertices, GetNumInvocations(i, endVertex)); + +#if KNOB_ENABLE_TOSS_POINTS + if (!KNOB_TOSS_FETCH) +#endif + { + RDTSC_START(FEVertexShader); + state.pfnVertexFunc(GetPrivateState(pDC), &vsContext); + RDTSC_STOP(FEVertexShader, 0, 0); + + UPDATE_STAT(VsInvocations, GetNumInvocations(i, endVertex)); + } + } + + // 2. Assemble primitives given the last two SIMD. + do + { + simdvector prim[MAX_NUM_VERTS_PER_PRIM]; + // PaAssemble returns false if there is not enough verts to assemble. + RDTSC_START(FEPAAssemble); + bool assemble = pa.Assemble(VERTEX_POSITION_SLOT, prim); + RDTSC_STOP(FEPAAssemble, 1, 0); + +#if KNOB_ENABLE_TOSS_POINTS + if (!KNOB_TOSS_FETCH) +#endif + { +#if KNOB_ENABLE_TOSS_POINTS + if (!KNOB_TOSS_VS) +#endif + { + if (assemble) + { + UPDATE_STAT(IaPrimitives, pa.NumPrims()); + + if (HasTessellationT) + { + TessellationStages( + pDC, workerId, pa, pGsOut, pCutBuffer, pStreamCutBuffer, pSoPrimData, pa.GetPrimID(work.startPrimID)); + } + else if (HasGeometryShaderT) + { + GeometryShaderStage( + pDC, workerId, pa, pGsOut, pCutBuffer, pStreamCutBuffer, pSoPrimData, pa.GetPrimID(work.startPrimID)); + } + else + { + // If streamout is enabled then stream vertices out to memory. + if (HasStreamOutT) + { + StreamOut(pDC, pa, workerId, pSoPrimData, 0); + } + + if (HasRastT) + { + SWR_ASSERT(pDC->pState->pfnProcessPrims); + pDC->pState->pfnProcessPrims(pDC, pa, workerId, prim, + GenMask(pa.NumPrims()), pa.GetPrimID(work.startPrimID)); + } + } + } + } + } + } while (pa.NextPrim()); + + i += KNOB_SIMD_WIDTH; + if (IsIndexedT) + { + fetchInfo.pIndices = (int*)((BYTE*)fetchInfo.pIndices + KNOB_SIMD_WIDTH * indexSize); + } + else + { + vIndex = _simd_add_epi32(vIndex, _simd_set1_epi32(KNOB_SIMD_WIDTH)); + } + } + pa.Reset(); + } + + RDTSC_STOP(FEProcessDraw, numPrims * work.numInstances, pDC->drawId); +} +// Explicit Instantiation of all combinations +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +template void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); + + +////////////////////////////////////////////////////////////////////////// +/// @brief Processes attributes for the backend based on linkage mask and +/// linkage map. Essentially just doing an SOA->AOS conversion and pack. +/// @param pDC - Draw context +/// @param pa - Primitive Assembly state +/// @param linkageMask - Specifies which VS outputs are routed to PS. +/// @param pLinkageMap - maps VS attribute slot to PS slot +/// @param triIndex - Triangle to process attributes for +/// @param pBuffer - Output result +template +INLINE void ProcessAttributes( + DRAW_CONTEXT *pDC, + PA_STATE&pa, + uint32_t linkageMask, + const uint8_t* pLinkageMap, + uint32_t triIndex, + float *pBuffer) +{ + DWORD slot = 0; + uint32_t mapIdx = 0; + LONG constantInterpMask = pDC->pState->state.backendState.constantInterpolationMask; + const uint32_t provokingVertex = pDC->pState->state.frontendState.topologyProvokingVertex; + + while (_BitScanForward(&slot, linkageMask)) + { + linkageMask &= ~(1 << slot); // done with this bit. + + // compute absolute slot in vertex attrib array + uint32_t inputSlot = VERTEX_ATTRIB_START_SLOT + pLinkageMap[mapIdx]; + + __m128 attrib[3]; // triangle attribs (always 4 wide) + pa.AssembleSingle(inputSlot, triIndex, attrib); + + if (_bittest(&constantInterpMask, mapIdx)) + { + for (uint32_t i = 0; i < NumVerts; ++i) + { + _mm_store_ps(pBuffer, attrib[provokingVertex]); + pBuffer += 4; + } + } + else + { + for (uint32_t i = 0; i < NumVerts; ++i) + { + _mm_store_ps(pBuffer, attrib[i]); + pBuffer += 4; + } + } + + // pad out the attrib buffer to 3 verts to ensure the triangle + // interpolation code in the pixel shader works correctly for the + // 3 topologies - point, line, tri. This effectively zeros out the + // effect of the missing vertices in the triangle interpolation. + for (uint32_t i = NumVerts; i < 3; ++i) + { + _mm_store_ps(pBuffer, attrib[NumVerts - 1]); + pBuffer += 4; + } + + mapIdx++; + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Processes enabled user clip distances. Loads the active clip +/// distances from the PA, sets up barycentric equations, and +/// stores the results to the output buffer +/// @param pa - Primitive Assembly state +/// @param primIndex - primitive index to process +/// @param clipDistMask - mask of enabled clip distances +/// @param pUserClipBuffer - buffer to store results +template +void ProcessUserClipDist(PA_STATE& pa, uint32_t primIndex, uint8_t clipDistMask, float* pUserClipBuffer) +{ + DWORD clipDist; + while (_BitScanForward(&clipDist, clipDistMask)) + { + clipDistMask &= ~(1 << clipDist); + uint32_t clipSlot = clipDist >> 2; + uint32_t clipComp = clipDist & 0x3; + uint32_t clipAttribSlot = clipSlot == 0 ? + VERTEX_CLIPCULL_DIST_LO_SLOT : VERTEX_CLIPCULL_DIST_HI_SLOT; + + __m128 primClipDist[3]; + pa.AssembleSingle(clipAttribSlot, primIndex, primClipDist); + + float vertClipDist[NumVerts]; + for (uint32_t e = 0; e < NumVerts; ++e) + { + OSALIGNSIMD(float) aVertClipDist[4]; + _mm_store_ps(aVertClipDist, primClipDist[e]); + vertClipDist[e] = aVertClipDist[clipComp]; + }; + + // setup plane equations for barycentric interpolation in the backend + float baryCoeff[NumVerts]; + for (uint32_t e = 0; e < NumVerts - 1; ++e) + { + baryCoeff[e] = vertClipDist[e] - vertClipDist[NumVerts - 1]; + } + baryCoeff[NumVerts - 1] = vertClipDist[NumVerts - 1]; + + for (uint32_t e = 0; e < NumVerts; ++e) + { + *(pUserClipBuffer++) = baryCoeff[e]; + } + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Bin triangle primitives to macro tiles. Performs setup, clipping +/// culling, viewport transform, etc. +/// @param pDC - pointer to draw context. +/// @param pa - The primitive assembly object. +/// @param workerId - thread's worker id. Even thread has a unique id. +/// @param tri - Contains triangle position data for SIMDs worth of triangles. +/// @param primID - Primitive ID for each triangle. +void BinTriangles( + DRAW_CONTEXT *pDC, + PA_STATE& pa, + uint32_t workerId, + simdvector tri[3], + uint32_t triMask, + simdscalari primID) +{ + RDTSC_START(FEBinTriangles); + + const API_STATE& state = GetApiState(pDC); + const SWR_RASTSTATE& rastState = state.rastState; + const SWR_FRONTEND_STATE& feState = state.frontendState; + const SWR_GS_STATE& gsState = state.gsState; + + // Simple wireframe mode for debugging purposes only + + simdscalar vRecipW0 = _simd_set1_ps(1.0f); + simdscalar vRecipW1 = _simd_set1_ps(1.0f); + simdscalar vRecipW2 = _simd_set1_ps(1.0f); + + if (!feState.vpTransformDisable) + { + // perspective divide + vRecipW0 = _simd_div_ps(_simd_set1_ps(1.0f), tri[0].w); + vRecipW1 = _simd_div_ps(_simd_set1_ps(1.0f), tri[1].w); + vRecipW2 = _simd_div_ps(_simd_set1_ps(1.0f), tri[2].w); + + tri[0].v[0] = _simd_mul_ps(tri[0].v[0], vRecipW0); + tri[1].v[0] = _simd_mul_ps(tri[1].v[0], vRecipW1); + tri[2].v[0] = _simd_mul_ps(tri[2].v[0], vRecipW2); + + tri[0].v[1] = _simd_mul_ps(tri[0].v[1], vRecipW0); + tri[1].v[1] = _simd_mul_ps(tri[1].v[1], vRecipW1); + tri[2].v[1] = _simd_mul_ps(tri[2].v[1], vRecipW2); + + tri[0].v[2] = _simd_mul_ps(tri[0].v[2], vRecipW0); + tri[1].v[2] = _simd_mul_ps(tri[1].v[2], vRecipW1); + tri[2].v[2] = _simd_mul_ps(tri[2].v[2], vRecipW2); + + // viewport transform to screen coords + viewportTransform<3>(tri, state.vpMatrix[0]); + } + + // adjust for pixel center location + simdscalar offset = g_pixelOffsets[rastState.pixelLocation]; + tri[0].x = _simd_add_ps(tri[0].x, offset); + tri[0].y = _simd_add_ps(tri[0].y, offset); + + tri[1].x = _simd_add_ps(tri[1].x, offset); + tri[1].y = _simd_add_ps(tri[1].y, offset); + + tri[2].x = _simd_add_ps(tri[2].x, offset); + tri[2].y = _simd_add_ps(tri[2].y, offset); + + // convert to fixed point + simdscalari vXi[3], vYi[3]; + vXi[0] = fpToFixedPointVertical(tri[0].x); + vYi[0] = fpToFixedPointVertical(tri[0].y); + vXi[1] = fpToFixedPointVertical(tri[1].x); + vYi[1] = fpToFixedPointVertical(tri[1].y); + vXi[2] = fpToFixedPointVertical(tri[2].x); + vYi[2] = fpToFixedPointVertical(tri[2].y); + + // triangle setup + simdscalari vAi[3], vBi[3]; + triangleSetupABIntVertical(vXi, vYi, vAi, vBi); + + // determinant + simdscalari vDet[2]; + calcDeterminantIntVertical(vAi, vBi, vDet); + + // cull zero area + int maskLo = _simd_movemask_pd(_simd_castsi_pd(_simd_cmpeq_epi64(vDet[0], _simd_setzero_si()))); + int maskHi = _simd_movemask_pd(_simd_castsi_pd(_simd_cmpeq_epi64(vDet[1], _simd_setzero_si()))); + + int cullZeroAreaMask = maskLo | ((maskHi << KNOB_SIMD_WIDTH / 2)); + + uint32_t origTriMask = triMask; + triMask &= ~cullZeroAreaMask; + + // determine front winding tris + // CW +det + // CCW -det + maskLo = _simd_movemask_pd(_simd_castsi_pd(_simd_cmpgt_epi64(vDet[0], _simd_setzero_si()))); + maskHi = _simd_movemask_pd(_simd_castsi_pd(_simd_cmpgt_epi64(vDet[1], _simd_setzero_si()))); + int cwTriMask = maskLo | (maskHi << (KNOB_SIMD_WIDTH /2) ); + + uint32_t frontWindingTris; + if (rastState.frontWinding == SWR_FRONTWINDING_CW) + { + frontWindingTris = cwTriMask; + } + else + { + frontWindingTris = ~cwTriMask; + } + + // cull + uint32_t cullTris; + switch ((SWR_CULLMODE)rastState.cullMode) + { + case SWR_CULLMODE_BOTH: cullTris = 0xffffffff; break; + case SWR_CULLMODE_NONE: cullTris = 0x0; break; + case SWR_CULLMODE_FRONT: cullTris = frontWindingTris; break; + case SWR_CULLMODE_BACK: cullTris = ~frontWindingTris; break; + default: SWR_ASSERT(false, "Invalid cull mode: %d", rastState.cullMode); cullTris = 0x0; break; + } + + triMask &= ~cullTris; + + if (origTriMask ^ triMask) + { + RDTSC_EVENT(FECullZeroAreaAndBackface, _mm_popcnt_u32(origTriMask ^ triMask), 0); + } + + // compute per tri backface + uint32_t frontFaceMask = frontWindingTris; + + uint32_t *pPrimID = (uint32_t *)&primID; + DWORD triIndex = 0; + + if (!triMask) + { + goto endBinTriangles; + } + + // Calc bounding box of triangles + simdBBox bbox; + calcBoundingBoxIntVertical(vXi, vYi, bbox); + + // determine if triangle falls between pixel centers and discard + // only discard for non-MSAA case + // (left + 127) & ~255 + // (right + 128) & ~255 + + if(rastState.sampleCount == SWR_MULTISAMPLE_1X) + { + origTriMask = triMask; + + int cullCenterMask; + { + simdscalari left = _simd_add_epi32(bbox.left, _simd_set1_epi32(127)); + left = _simd_and_si(left, _simd_set1_epi32(~255)); + simdscalari right = _simd_add_epi32(bbox.right, _simd_set1_epi32(128)); + right = _simd_and_si(right, _simd_set1_epi32(~255)); + + simdscalari vMaskH = _simd_cmpeq_epi32(left, right); + + simdscalari top = _simd_add_epi32(bbox.top, _simd_set1_epi32(127)); + top = _simd_and_si(top, _simd_set1_epi32(~255)); + simdscalari bottom = _simd_add_epi32(bbox.bottom, _simd_set1_epi32(128)); + bottom = _simd_and_si(bottom, _simd_set1_epi32(~255)); + + simdscalari vMaskV = _simd_cmpeq_epi32(top, bottom); + vMaskV = _simd_or_si(vMaskH, vMaskV); + cullCenterMask = _simd_movemask_ps(_simd_castsi_ps(vMaskV)); + } + + triMask &= ~cullCenterMask; + + if(origTriMask ^ triMask) + { + RDTSC_EVENT(FECullBetweenCenters, _mm_popcnt_u32(origTriMask ^ triMask), 0); + } + } + + // Intersect with scissor/viewport. Subtract 1 ULP in x.8 fixed point since right/bottom edge is exclusive. + bbox.left = _simd_max_epi32(bbox.left, _simd_set1_epi32(state.scissorInFixedPoint.left)); + bbox.top = _simd_max_epi32(bbox.top, _simd_set1_epi32(state.scissorInFixedPoint.top)); + bbox.right = _simd_min_epi32(_simd_sub_epi32(bbox.right, _simd_set1_epi32(1)), _simd_set1_epi32(state.scissorInFixedPoint.right)); + bbox.bottom = _simd_min_epi32(_simd_sub_epi32(bbox.bottom, _simd_set1_epi32(1)), _simd_set1_epi32(state.scissorInFixedPoint.bottom)); + + // Cull tris completely outside scissor + { + simdscalari maskOutsideScissorX = _simd_cmpgt_epi32(bbox.left, bbox.right); + simdscalari maskOutsideScissorY = _simd_cmpgt_epi32(bbox.top, bbox.bottom); + simdscalari maskOutsideScissorXY = _simd_or_si(maskOutsideScissorX, maskOutsideScissorY); + uint32_t maskOutsideScissor = _simd_movemask_ps(_simd_castsi_ps(maskOutsideScissorXY)); + triMask = triMask & ~maskOutsideScissor; + } + + if (!triMask) + { + goto endBinTriangles; + } + + // Convert triangle bbox to macrotile units. + bbox.left = _simd_srai_epi32(bbox.left, KNOB_MACROTILE_X_DIM_FIXED_SHIFT); + bbox.top = _simd_srai_epi32(bbox.top, KNOB_MACROTILE_Y_DIM_FIXED_SHIFT); + bbox.right = _simd_srai_epi32(bbox.right, KNOB_MACROTILE_X_DIM_FIXED_SHIFT); + bbox.bottom = _simd_srai_epi32(bbox.bottom, KNOB_MACROTILE_Y_DIM_FIXED_SHIFT); + + OSALIGNSIMD(uint32_t) aMTLeft[KNOB_SIMD_WIDTH], aMTRight[KNOB_SIMD_WIDTH], aMTTop[KNOB_SIMD_WIDTH], aMTBottom[KNOB_SIMD_WIDTH]; + _simd_store_si((simdscalari*)aMTLeft, bbox.left); + _simd_store_si((simdscalari*)aMTRight, bbox.right); + _simd_store_si((simdscalari*)aMTTop, bbox.top); + _simd_store_si((simdscalari*)aMTBottom, bbox.bottom); + + // transpose verts needed for backend + /// @todo modify BE to take non-transformed verts + __m128 vHorizX[8], vHorizY[8], vHorizZ[8], vHorizW[8]; + vTranspose3x8(vHorizX, tri[0].x, tri[1].x, tri[2].x); + vTranspose3x8(vHorizY, tri[0].y, tri[1].y, tri[2].y); + vTranspose3x8(vHorizZ, tri[0].z, tri[1].z, tri[2].z); + vTranspose3x8(vHorizW, vRecipW0, vRecipW1, vRecipW2); + + // store render target array index + OSALIGNSIMD(uint32_t) aRTAI[KNOB_SIMD_WIDTH]; + if (gsState.gsEnable && gsState.emitsRenderTargetArrayIndex) + { + simdvector vRtai[3]; + pa.Assemble(VERTEX_RTAI_SLOT, vRtai); + simdscalari vRtaii; + vRtaii = _simd_castps_si(vRtai[0].x); + _simd_store_si((simdscalari*)aRTAI, vRtaii); + } + else + { + _simd_store_si((simdscalari*)aRTAI, _simd_setzero_si()); + } + + // scan remaining valid triangles and bin each separately + while (_BitScanForward(&triIndex, triMask)) + { + uint32_t linkageCount = state.linkageCount; + uint32_t linkageMask = state.linkageMask; + uint32_t numScalarAttribs = linkageCount * 4; + + BE_WORK work; + work.type = DRAW; + + TRIANGLE_WORK_DESC &desc = work.desc.tri; + + desc.triFlags.frontFacing = state.forceFront ? 1 : ((frontFaceMask >> triIndex) & 1); + desc.triFlags.primID = pPrimID[triIndex]; + desc.triFlags.renderTargetArrayIndex = aRTAI[triIndex]; + + if(rastState.samplePattern == SWR_MSAA_STANDARD_PATTERN) + { + work.pfnWork = gRasterizerTable[rastState.scissorEnable][rastState.sampleCount]; + } + else + { + // for center sample pattern, all samples are at pixel center; calculate coverage + // once at center and broadcast the results in the backend + work.pfnWork = gRasterizerTable[rastState.scissorEnable][SWR_MULTISAMPLE_1X]; + } + + Arena* pArena = pDC->pArena; + SWR_ASSERT(pArena != nullptr); + + // store active attribs + float *pAttribs = (float*)pArena->AllocAligned(numScalarAttribs * 3 * sizeof(float), 16); + desc.pAttribs = pAttribs; + desc.numAttribs = linkageCount; + ProcessAttributes<3>(pDC, pa, linkageMask, state.linkageMap, triIndex, desc.pAttribs); + + // store triangle vertex data + desc.pTriBuffer = (float*)pArena->AllocAligned(4 * 4 * sizeof(float), 16); + + _mm_store_ps(&desc.pTriBuffer[0], vHorizX[triIndex]); + _mm_store_ps(&desc.pTriBuffer[4], vHorizY[triIndex]); + _mm_store_ps(&desc.pTriBuffer[8], vHorizZ[triIndex]); + _mm_store_ps(&desc.pTriBuffer[12], vHorizW[triIndex]); + + // store user clip distances + if (rastState.clipDistanceMask) + { + uint32_t numClipDist = _mm_popcnt_u32(rastState.clipDistanceMask); + desc.pUserClipBuffer = (float*)pArena->Alloc(numClipDist * 3 * sizeof(float)); + ProcessUserClipDist<3>(pa, triIndex, rastState.clipDistanceMask, desc.pUserClipBuffer); + } + + MacroTileMgr *pTileMgr = pDC->pTileMgr; + for (uint32_t y = aMTTop[triIndex]; y <= aMTBottom[triIndex]; ++y) + { + for (uint32_t x = aMTLeft[triIndex]; x <= aMTRight[triIndex]; ++x) + { +#if KNOB_ENABLE_TOSS_POINTS + if (!KNOB_TOSS_SETUP_TRIS) +#endif + { + pTileMgr->enqueue(x, y, &work); + } + } + } + + triMask &= ~(1 << triIndex); + } + +endBinTriangles: + RDTSC_STOP(FEBinTriangles, 1, 0); +} + + + +////////////////////////////////////////////////////////////////////////// +/// @brief Bin SIMD points to the backend. Only supports point size of 1 +/// @param pDC - pointer to draw context. +/// @param pa - The primitive assembly object. +/// @param workerId - thread's worker id. Even thread has a unique id. +/// @param tri - Contains point position data for SIMDs worth of points. +/// @param primID - Primitive ID for each point. +void BinPoints( + DRAW_CONTEXT *pDC, + PA_STATE& pa, + uint32_t workerId, + simdvector prim[3], + uint32_t primMask, + simdscalari primID) +{ + RDTSC_START(FEBinPoints); + + simdvector& primVerts = prim[0]; + + const API_STATE& state = GetApiState(pDC); + const SWR_FRONTEND_STATE& feState = state.frontendState; + const SWR_GS_STATE& gsState = state.gsState; + const SWR_RASTSTATE& rastState = state.rastState; + + if (!feState.vpTransformDisable) + { + // perspective divide + simdscalar vRecipW0 = _simd_div_ps(_simd_set1_ps(1.0f), primVerts.w); + primVerts.x = _simd_mul_ps(primVerts.x, vRecipW0); + primVerts.y = _simd_mul_ps(primVerts.y, vRecipW0); + primVerts.z = _simd_mul_ps(primVerts.z, vRecipW0); + + // viewport transform to screen coords + viewportTransform<1>(&primVerts, state.vpMatrix[0]); + } + + // adjust for pixel center location + simdscalar offset = g_pixelOffsets[rastState.pixelLocation]; + primVerts.x = _simd_add_ps(primVerts.x, offset); + primVerts.y = _simd_add_ps(primVerts.y, offset); + + // convert to fixed point + simdscalari vXi, vYi; + vXi = fpToFixedPointVertical(primVerts.x); + vYi = fpToFixedPointVertical(primVerts.y); + + if (CanUseSimplePoints(pDC)) + { + // adjust for top-left rule + vXi = _simd_sub_epi32(vXi, _simd_set1_epi32(1)); + vYi = _simd_sub_epi32(vYi, _simd_set1_epi32(1)); + + // cull points off the top-left edge of the viewport + primMask &= ~_simd_movemask_ps(_simd_castsi_ps(vXi)); + primMask &= ~_simd_movemask_ps(_simd_castsi_ps(vYi)); + + // compute macro tile coordinates + simdscalari macroX = _simd_srai_epi32(vXi, KNOB_MACROTILE_X_DIM_FIXED_SHIFT); + simdscalari macroY = _simd_srai_epi32(vYi, KNOB_MACROTILE_Y_DIM_FIXED_SHIFT); + + OSALIGNSIMD(uint32_t) aMacroX[KNOB_SIMD_WIDTH], aMacroY[KNOB_SIMD_WIDTH]; + _simd_store_si((simdscalari*)aMacroX, macroX); + _simd_store_si((simdscalari*)aMacroY, macroY); + + // compute raster tile coordinates + simdscalari rasterX = _simd_srai_epi32(vXi, KNOB_TILE_X_DIM_SHIFT + FIXED_POINT_SHIFT); + simdscalari rasterY = _simd_srai_epi32(vYi, KNOB_TILE_Y_DIM_SHIFT + FIXED_POINT_SHIFT); + + // compute raster tile relative x,y for coverage mask + simdscalari tileAlignedX = _simd_slli_epi32(rasterX, KNOB_TILE_X_DIM_SHIFT); + simdscalari tileAlignedY = _simd_slli_epi32(rasterY, KNOB_TILE_Y_DIM_SHIFT); + + simdscalari tileRelativeX = _simd_sub_epi32(_simd_srai_epi32(vXi, FIXED_POINT_SHIFT), tileAlignedX); + simdscalari tileRelativeY = _simd_sub_epi32(_simd_srai_epi32(vYi, FIXED_POINT_SHIFT), tileAlignedY); + + OSALIGNSIMD(uint32_t) aTileRelativeX[KNOB_SIMD_WIDTH]; + OSALIGNSIMD(uint32_t) aTileRelativeY[KNOB_SIMD_WIDTH]; + _simd_store_si((simdscalari*)aTileRelativeX, tileRelativeX); + _simd_store_si((simdscalari*)aTileRelativeY, tileRelativeY); + + OSALIGNSIMD(uint32_t) aTileAlignedX[KNOB_SIMD_WIDTH]; + OSALIGNSIMD(uint32_t) aTileAlignedY[KNOB_SIMD_WIDTH]; + _simd_store_si((simdscalari*)aTileAlignedX, tileAlignedX); + _simd_store_si((simdscalari*)aTileAlignedY, tileAlignedY); + + OSALIGNSIMD(float) aZ[KNOB_SIMD_WIDTH]; + _simd_store_ps((float*)aZ, primVerts.z); + + // store render target array index + OSALIGNSIMD(uint32_t) aRTAI[KNOB_SIMD_WIDTH]; + if (gsState.gsEnable && gsState.emitsRenderTargetArrayIndex) + { + simdvector vRtai; + pa.Assemble(VERTEX_RTAI_SLOT, &vRtai); + simdscalari vRtaii = _simd_castps_si(vRtai.x); + _simd_store_si((simdscalari*)aRTAI, vRtaii); + } + else + { + _simd_store_si((simdscalari*)aRTAI, _simd_setzero_si()); + } + + uint32_t *pPrimID = (uint32_t *)&primID; + DWORD primIndex = 0; + // scan remaining valid triangles and bin each separately + while (_BitScanForward(&primIndex, primMask)) + { + uint32_t linkageCount = state.linkageCount; + uint32_t linkageMask = state.linkageMask; + + uint32_t numScalarAttribs = linkageCount * 4; + + BE_WORK work; + work.type = DRAW; + + TRIANGLE_WORK_DESC &desc = work.desc.tri; + + // points are always front facing + desc.triFlags.frontFacing = 1; + desc.triFlags.primID = pPrimID[primIndex]; + desc.triFlags.renderTargetArrayIndex = aRTAI[primIndex]; + + work.pfnWork = RasterizeSimplePoint; + + Arena* pArena = pDC->pArena; + SWR_ASSERT(pArena != nullptr); + + // store attributes + float *pAttribs = (float*)pArena->AllocAligned(3 * numScalarAttribs * sizeof(float), 16); + desc.pAttribs = pAttribs; + desc.numAttribs = linkageCount; + + ProcessAttributes<1>(pDC, pa, linkageMask, state.linkageMap, primIndex, pAttribs); + + // store raster tile aligned x, y, perspective correct z + float *pTriBuffer = (float*)pArena->AllocAligned(4 * sizeof(float), 16); + desc.pTriBuffer = pTriBuffer; + *(uint32_t*)pTriBuffer++ = aTileAlignedX[primIndex]; + *(uint32_t*)pTriBuffer++ = aTileAlignedY[primIndex]; + *pTriBuffer = aZ[primIndex]; + + uint32_t tX = aTileRelativeX[primIndex]; + uint32_t tY = aTileRelativeY[primIndex]; + + // pack the relative x,y into the coverageMask, the rasterizer will + // generate the true coverage mask from it + work.desc.tri.triFlags.coverageMask = tX | (tY << 4); + + // bin it + MacroTileMgr *pTileMgr = pDC->pTileMgr; +#if KNOB_ENABLE_TOSS_POINTS + if (!KNOB_TOSS_SETUP_TRIS) +#endif + { + pTileMgr->enqueue(aMacroX[primIndex], aMacroY[primIndex], &work); + } + primMask &= ~(1 << primIndex); + } + } + else + { + // non simple points need to be potentially binned to multiple macro tiles + simdscalar vPointSize; + if (rastState.pointParam) + { + simdvector size[3]; + pa.Assemble(VERTEX_POINT_SIZE_SLOT, size); + vPointSize = size[0].x; + } + else + { + vPointSize = _simd_set1_ps(rastState.pointSize); + } + + // bloat point to bbox + simdBBox bbox; + bbox.left = bbox.right = vXi; + bbox.top = bbox.bottom = vYi; + + simdscalar vHalfWidth = _simd_mul_ps(vPointSize, _simd_set1_ps(0.5f)); + simdscalari vHalfWidthi = fpToFixedPointVertical(vHalfWidth); + bbox.left = _simd_sub_epi32(bbox.left, vHalfWidthi); + bbox.right = _simd_add_epi32(bbox.right, vHalfWidthi); + bbox.top = _simd_sub_epi32(bbox.top, vHalfWidthi); + bbox.bottom = _simd_add_epi32(bbox.bottom, vHalfWidthi); + + // Intersect with scissor/viewport. Subtract 1 ULP in x.8 fixed point since right/bottom edge is exclusive. + bbox.left = _simd_max_epi32(bbox.left, _simd_set1_epi32(state.scissorInFixedPoint.left)); + bbox.top = _simd_max_epi32(bbox.top, _simd_set1_epi32(state.scissorInFixedPoint.top)); + bbox.right = _simd_min_epi32(_simd_sub_epi32(bbox.right, _simd_set1_epi32(1)), _simd_set1_epi32(state.scissorInFixedPoint.right)); + bbox.bottom = _simd_min_epi32(_simd_sub_epi32(bbox.bottom, _simd_set1_epi32(1)), _simd_set1_epi32(state.scissorInFixedPoint.bottom)); + + // Cull bloated points completely outside scissor + simdscalari maskOutsideScissorX = _simd_cmpgt_epi32(bbox.left, bbox.right); + simdscalari maskOutsideScissorY = _simd_cmpgt_epi32(bbox.top, bbox.bottom); + simdscalari maskOutsideScissorXY = _simd_or_si(maskOutsideScissorX, maskOutsideScissorY); + uint32_t maskOutsideScissor = _simd_movemask_ps(_simd_castsi_ps(maskOutsideScissorXY)); + primMask = primMask & ~maskOutsideScissor; + + // Convert bbox to macrotile units. + bbox.left = _simd_srai_epi32(bbox.left, KNOB_MACROTILE_X_DIM_FIXED_SHIFT); + bbox.top = _simd_srai_epi32(bbox.top, KNOB_MACROTILE_Y_DIM_FIXED_SHIFT); + bbox.right = _simd_srai_epi32(bbox.right, KNOB_MACROTILE_X_DIM_FIXED_SHIFT); + bbox.bottom = _simd_srai_epi32(bbox.bottom, KNOB_MACROTILE_Y_DIM_FIXED_SHIFT); + + OSALIGNSIMD(uint32_t) aMTLeft[KNOB_SIMD_WIDTH], aMTRight[KNOB_SIMD_WIDTH], aMTTop[KNOB_SIMD_WIDTH], aMTBottom[KNOB_SIMD_WIDTH]; + _simd_store_si((simdscalari*)aMTLeft, bbox.left); + _simd_store_si((simdscalari*)aMTRight, bbox.right); + _simd_store_si((simdscalari*)aMTTop, bbox.top); + _simd_store_si((simdscalari*)aMTBottom, bbox.bottom); + + // store render target array index + OSALIGNSIMD(uint32_t) aRTAI[KNOB_SIMD_WIDTH]; + if (gsState.gsEnable && gsState.emitsRenderTargetArrayIndex) + { + simdvector vRtai[2]; + pa.Assemble(VERTEX_RTAI_SLOT, vRtai); + simdscalari vRtaii = _simd_castps_si(vRtai[0].x); + _simd_store_si((simdscalari*)aRTAI, vRtaii); + } + else + { + _simd_store_si((simdscalari*)aRTAI, _simd_setzero_si()); + } + + OSALIGNSIMD(float) aPointSize[KNOB_SIMD_WIDTH]; + _simd_store_ps((float*)aPointSize, vPointSize); + + uint32_t *pPrimID = (uint32_t *)&primID; + + OSALIGNSIMD(float) aPrimVertsX[KNOB_SIMD_WIDTH]; + OSALIGNSIMD(float) aPrimVertsY[KNOB_SIMD_WIDTH]; + OSALIGNSIMD(float) aPrimVertsZ[KNOB_SIMD_WIDTH]; + + _simd_store_ps((float*)aPrimVertsX, primVerts.x); + _simd_store_ps((float*)aPrimVertsY, primVerts.y); + _simd_store_ps((float*)aPrimVertsZ, primVerts.z); + + // scan remaining valid prims and bin each separately + DWORD primIndex; + while (_BitScanForward(&primIndex, primMask)) + { + uint32_t linkageCount = state.linkageCount; + uint32_t linkageMask = state.linkageMask; + uint32_t numScalarAttribs = linkageCount * 4; + + BE_WORK work; + work.type = DRAW; + + TRIANGLE_WORK_DESC &desc = work.desc.tri; + + desc.triFlags.frontFacing = 1; + desc.triFlags.primID = pPrimID[primIndex]; + desc.triFlags.pointSize = aPointSize[primIndex]; + desc.triFlags.renderTargetArrayIndex = aRTAI[primIndex]; + + work.pfnWork = RasterizeTriPoint; + + Arena* pArena = pDC->pArena; + SWR_ASSERT(pArena != nullptr); + + // store active attribs + desc.pAttribs = (float*)pArena->AllocAligned(numScalarAttribs * 3 * sizeof(float), 16); + desc.numAttribs = linkageCount; + ProcessAttributes<1>(pDC, pa, linkageMask, state.linkageMap, primIndex, desc.pAttribs); + + // store point vertex data + float *pTriBuffer = (float*)pArena->AllocAligned(4 * sizeof(float), 16); + desc.pTriBuffer = pTriBuffer; + *pTriBuffer++ = aPrimVertsX[primIndex]; + *pTriBuffer++ = aPrimVertsY[primIndex]; + *pTriBuffer = aPrimVertsZ[primIndex]; + + // store user clip distances + if (rastState.clipDistanceMask) + { + uint32_t numClipDist = _mm_popcnt_u32(rastState.clipDistanceMask); + desc.pUserClipBuffer = (float*)pArena->Alloc(numClipDist * 2 * sizeof(float)); + ProcessUserClipDist<2>(pa, primIndex, rastState.clipDistanceMask, desc.pUserClipBuffer); + } + + MacroTileMgr *pTileMgr = pDC->pTileMgr; + for (uint32_t y = aMTTop[primIndex]; y <= aMTBottom[primIndex]; ++y) + { + for (uint32_t x = aMTLeft[primIndex]; x <= aMTRight[primIndex]; ++x) + { +#if KNOB_ENABLE_TOSS_POINTS + if (!KNOB_TOSS_SETUP_TRIS) +#endif + { + pTileMgr->enqueue(x, y, &work); + } + } + } + + primMask &= ~(1 << primIndex); + } + } + + + + + RDTSC_STOP(FEBinPoints, 1, 0); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Bin SIMD lines to the backend. +/// @param pDC - pointer to draw context. +/// @param pa - The primitive assembly object. +/// @param workerId - thread's worker id. Even thread has a unique id. +/// @param tri - Contains line position data for SIMDs worth of points. +/// @param primID - Primitive ID for each line. +void BinLines( + DRAW_CONTEXT *pDC, + PA_STATE& pa, + uint32_t workerId, + simdvector prim[], + uint32_t primMask, + simdscalari primID) +{ + RDTSC_START(FEBinLines); + + const API_STATE& state = GetApiState(pDC); + const SWR_RASTSTATE& rastState = state.rastState; + const SWR_FRONTEND_STATE& feState = state.frontendState; + const SWR_GS_STATE& gsState = state.gsState; + + simdscalar vRecipW0 = _simd_set1_ps(1.0f); + simdscalar vRecipW1 = _simd_set1_ps(1.0f); + + if (!feState.vpTransformDisable) + { + // perspective divide + vRecipW0 = _simd_div_ps(_simd_set1_ps(1.0f), prim[0].w); + vRecipW1 = _simd_div_ps(_simd_set1_ps(1.0f), prim[1].w); + + prim[0].v[0] = _simd_mul_ps(prim[0].v[0], vRecipW0); + prim[1].v[0] = _simd_mul_ps(prim[1].v[0], vRecipW1); + + prim[0].v[1] = _simd_mul_ps(prim[0].v[1], vRecipW0); + prim[1].v[1] = _simd_mul_ps(prim[1].v[1], vRecipW1); + + prim[0].v[2] = _simd_mul_ps(prim[0].v[2], vRecipW0); + prim[1].v[2] = _simd_mul_ps(prim[1].v[2], vRecipW1); + + // viewport transform to screen coords + viewportTransform<2>(prim, state.vpMatrix[0]); + } + + // adjust for pixel center location + simdscalar offset = g_pixelOffsets[rastState.pixelLocation]; + prim[0].x = _simd_add_ps(prim[0].x, offset); + prim[0].y = _simd_add_ps(prim[0].y, offset); + + prim[1].x = _simd_add_ps(prim[1].x, offset); + prim[1].y = _simd_add_ps(prim[1].y, offset); + + // convert to fixed point + simdscalari vXi[2], vYi[2]; + vXi[0] = fpToFixedPointVertical(prim[0].x); + vYi[0] = fpToFixedPointVertical(prim[0].y); + vXi[1] = fpToFixedPointVertical(prim[1].x); + vYi[1] = fpToFixedPointVertical(prim[1].y); + + // compute x-major vs y-major mask + simdscalari xLength = _simd_abs_epi32(_simd_sub_epi32(vXi[0], vXi[1])); + simdscalari yLength = _simd_abs_epi32(_simd_sub_epi32(vYi[0], vYi[1])); + simdscalar vYmajorMask = _simd_castsi_ps(_simd_cmpgt_epi32(yLength, xLength)); + uint32_t yMajorMask = _simd_movemask_ps(vYmajorMask); + + // cull zero-length lines + simdscalari vZeroLengthMask = _simd_cmpeq_epi32(xLength, _simd_setzero_si()); + vZeroLengthMask = _simd_and_si(vZeroLengthMask, _simd_cmpeq_epi32(yLength, _simd_setzero_si())); + + primMask &= ~_simd_movemask_ps(_simd_castsi_ps(vZeroLengthMask)); + + uint32_t *pPrimID = (uint32_t *)&primID; + + simdscalar vUnused = _simd_setzero_ps(); + + // Calc bounding box of lines + simdBBox bbox; + bbox.left = _simd_min_epi32(vXi[0], vXi[1]); + bbox.right = _simd_max_epi32(vXi[0], vXi[1]); + bbox.top = _simd_min_epi32(vYi[0], vYi[1]); + bbox.bottom = _simd_max_epi32(vYi[0], vYi[1]); + + // bloat bbox by line width along minor axis + simdscalar vHalfWidth = _simd_set1_ps(rastState.lineWidth / 2.0f); + simdscalari vHalfWidthi = fpToFixedPointVertical(vHalfWidth); + simdBBox bloatBox; + bloatBox.left = _simd_sub_epi32(bbox.left, vHalfWidthi); + bloatBox.right = _simd_add_epi32(bbox.right, vHalfWidthi); + bloatBox.top = _simd_sub_epi32(bbox.top, vHalfWidthi); + bloatBox.bottom = _simd_add_epi32(bbox.bottom, vHalfWidthi); + + bbox.left = _simd_blendv_epi32(bbox.left, bloatBox.left, vYmajorMask); + bbox.right = _simd_blendv_epi32(bbox.right, bloatBox.right, vYmajorMask); + bbox.top = _simd_blendv_epi32(bloatBox.top, bbox.top, vYmajorMask); + bbox.bottom = _simd_blendv_epi32(bloatBox.bottom, bbox.bottom, vYmajorMask); + + // Intersect with scissor/viewport. Subtract 1 ULP in x.8 fixed point since right/bottom edge is exclusive. + bbox.left = _simd_max_epi32(bbox.left, _simd_set1_epi32(state.scissorInFixedPoint.left)); + bbox.top = _simd_max_epi32(bbox.top, _simd_set1_epi32(state.scissorInFixedPoint.top)); + bbox.right = _simd_min_epi32(_simd_sub_epi32(bbox.right, _simd_set1_epi32(1)), _simd_set1_epi32(state.scissorInFixedPoint.right)); + bbox.bottom = _simd_min_epi32(_simd_sub_epi32(bbox.bottom, _simd_set1_epi32(1)), _simd_set1_epi32(state.scissorInFixedPoint.bottom)); + + // Cull prims completely outside scissor + { + simdscalari maskOutsideScissorX = _simd_cmpgt_epi32(bbox.left, bbox.right); + simdscalari maskOutsideScissorY = _simd_cmpgt_epi32(bbox.top, bbox.bottom); + simdscalari maskOutsideScissorXY = _simd_or_si(maskOutsideScissorX, maskOutsideScissorY); + uint32_t maskOutsideScissor = _simd_movemask_ps(_simd_castsi_ps(maskOutsideScissorXY)); + primMask = primMask & ~maskOutsideScissor; + } + + if (!primMask) + { + goto endBinLines; + } + + // Convert triangle bbox to macrotile units. + bbox.left = _simd_srai_epi32(bbox.left, KNOB_MACROTILE_X_DIM_FIXED_SHIFT); + bbox.top = _simd_srai_epi32(bbox.top, KNOB_MACROTILE_Y_DIM_FIXED_SHIFT); + bbox.right = _simd_srai_epi32(bbox.right, KNOB_MACROTILE_X_DIM_FIXED_SHIFT); + bbox.bottom = _simd_srai_epi32(bbox.bottom, KNOB_MACROTILE_Y_DIM_FIXED_SHIFT); + + OSALIGNSIMD(uint32_t) aMTLeft[KNOB_SIMD_WIDTH], aMTRight[KNOB_SIMD_WIDTH], aMTTop[KNOB_SIMD_WIDTH], aMTBottom[KNOB_SIMD_WIDTH]; + _simd_store_si((simdscalari*)aMTLeft, bbox.left); + _simd_store_si((simdscalari*)aMTRight, bbox.right); + _simd_store_si((simdscalari*)aMTTop, bbox.top); + _simd_store_si((simdscalari*)aMTBottom, bbox.bottom); + + // transpose verts needed for backend + /// @todo modify BE to take non-transformed verts + __m128 vHorizX[8], vHorizY[8], vHorizZ[8], vHorizW[8]; + vTranspose3x8(vHorizX, prim[0].x, prim[1].x, vUnused); + vTranspose3x8(vHorizY, prim[0].y, prim[1].y, vUnused); + vTranspose3x8(vHorizZ, prim[0].z, prim[1].z, vUnused); + vTranspose3x8(vHorizW, vRecipW0, vRecipW1, vUnused); + + // store render target array index + OSALIGNSIMD(uint32_t) aRTAI[KNOB_SIMD_WIDTH]; + if (gsState.gsEnable && gsState.emitsRenderTargetArrayIndex) + { + simdvector vRtai[2]; + pa.Assemble(VERTEX_RTAI_SLOT, vRtai); + simdscalari vRtaii = _simd_castps_si(vRtai[0].x); + _simd_store_si((simdscalari*)aRTAI, vRtaii); + } + else + { + _simd_store_si((simdscalari*)aRTAI, _simd_setzero_si()); + } + + // scan remaining valid prims and bin each separately + DWORD primIndex; + while (_BitScanForward(&primIndex, primMask)) + { + uint32_t linkageCount = state.linkageCount; + uint32_t linkageMask = state.linkageMask; + uint32_t numScalarAttribs = linkageCount * 4; + + BE_WORK work; + work.type = DRAW; + + TRIANGLE_WORK_DESC &desc = work.desc.tri; + + desc.triFlags.frontFacing = 1; + desc.triFlags.primID = pPrimID[primIndex]; + desc.triFlags.yMajor = (yMajorMask >> primIndex) & 1; + desc.triFlags.renderTargetArrayIndex = aRTAI[primIndex]; + + work.pfnWork = RasterizeLine; + + Arena* pArena = pDC->pArena; + SWR_ASSERT(pArena != nullptr); + + // store active attribs + desc.pAttribs = (float*)pArena->AllocAligned(numScalarAttribs * 3 * sizeof(float), 16); + desc.numAttribs = linkageCount; + ProcessAttributes<2>(pDC, pa, linkageMask, state.linkageMap, primIndex, desc.pAttribs); + + // store line vertex data + desc.pTriBuffer = (float*)pArena->AllocAligned(4 * 4 * sizeof(float), 16); + _mm_store_ps(&desc.pTriBuffer[0], vHorizX[primIndex]); + _mm_store_ps(&desc.pTriBuffer[4], vHorizY[primIndex]); + _mm_store_ps(&desc.pTriBuffer[8], vHorizZ[primIndex]); + _mm_store_ps(&desc.pTriBuffer[12], vHorizW[primIndex]); + + // store user clip distances + if (rastState.clipDistanceMask) + { + uint32_t numClipDist = _mm_popcnt_u32(rastState.clipDistanceMask); + desc.pUserClipBuffer = (float*)pArena->Alloc(numClipDist * 2 * sizeof(float)); + ProcessUserClipDist<2>(pa, primIndex, rastState.clipDistanceMask, desc.pUserClipBuffer); + } + + MacroTileMgr *pTileMgr = pDC->pTileMgr; + for (uint32_t y = aMTTop[primIndex]; y <= aMTBottom[primIndex]; ++y) + { + for (uint32_t x = aMTLeft[primIndex]; x <= aMTRight[primIndex]; ++x) + { +#if KNOB_ENABLE_TOSS_POINTS + if (!KNOB_TOSS_SETUP_TRIS) +#endif + { + pTileMgr->enqueue(x, y, &work); + } + } + } + + primMask &= ~(1 << primIndex); + } + +endBinLines: + + RDTSC_STOP(FEBinLines, 1, 0); +} diff --git a/src/gallium/drivers/swr/rasterizer/core/frontend.h b/src/gallium/drivers/swr/rasterizer/core/frontend.h new file mode 100644 index 00000000000..acb935fc251 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/frontend.h @@ -0,0 +1,327 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file frontend.h +* +* @brief Definitions for Frontend which handles vertex processing, +* primitive assembly, clipping, binning, etc. +* +******************************************************************************/ +#pragma once +#include "context.h" + +INLINE +__m128i fpToFixedPoint(const __m128 vIn) +{ + __m128 vFixed = _mm_mul_ps(vIn, _mm_set1_ps(FIXED_POINT_SCALE)); + return _mm_cvtps_epi32(vFixed); +} + +INLINE +simdscalari fpToFixedPointVertical(const simdscalar vIn) +{ + simdscalar vFixed = _simd_mul_ps(vIn, _simd_set1_ps(FIXED_POINT_SCALE)); + return _simd_cvtps_epi32(vFixed); +} + + +// Calculates the A and B coefficients for the 3 edges of the triangle +// +// maths for edge equations: +// standard form of a line in 2d +// Ax + By + C = 0 +// A = y0 - y1 +// B = x1 - x0 +// C = x0y1 - x1y0 +INLINE +void triangleSetupAB(const __m128 vX, const __m128 vY, __m128 & vA, __m128 & vB) +{ + // vYsub = y1 y2 y0 dc + __m128 vYsub = _mm_shuffle_ps(vY, vY, _MM_SHUFFLE(3, 0, 2, 1)); + // vY = y0 y1 y2 dc + vA = _mm_sub_ps(vY, vYsub); + + // Result: + // A[0] = y0 - y1 + // A[1] = y1 - y2 + // A[2] = y2 - y0 + + // vXsub = x1 x2 x0 dc + __m128 vXsub = _mm_shuffle_ps(vX, vX, _MM_SHUFFLE(3, 0, 2, 1)); + // vX = x0 x1 x2 dc + vB = _mm_sub_ps(vXsub, vX); + + // Result: + // B[0] = x1 - x0 + // B[1] = x2 - x1 + // B[2] = x0 - x2 +} + +INLINE +void triangleSetupABVertical(const simdscalar vX[3], const simdscalar vY[3], simdscalar (&vA)[3], simdscalar (&vB)[3]) +{ + // generate edge equations + // A = y0 - y1 + // B = x1 - x0 + vA[0] = _simd_sub_ps(vY[0], vY[1]); + vA[1] = _simd_sub_ps(vY[1], vY[2]); + vA[2] = _simd_sub_ps(vY[2], vY[0]); + + vB[0] = _simd_sub_ps(vX[1], vX[0]); + vB[1] = _simd_sub_ps(vX[2], vX[1]); + vB[2] = _simd_sub_ps(vX[0], vX[2]); +} + +INLINE +void triangleSetupABInt(const __m128i vX, const __m128i vY, __m128i & vA, __m128i & vB) +{ + // generate edge equations + // A = y0 - y1 + // B = x1 - x0 + // C = x0y1 - x1y0 + __m128i vYsub = _mm_shuffle_epi32(vY, _MM_SHUFFLE(3, 0, 2, 1)); + vA = _mm_sub_epi32(vY, vYsub); + + __m128i vXsub = _mm_shuffle_epi32(vX, _MM_SHUFFLE(3, 0, 2, 1)); + vB = _mm_sub_epi32(vXsub, vX); +} + +INLINE +void triangleSetupABIntVertical(const simdscalari vX[3], const simdscalari vY[3], simdscalari (&vA)[3], simdscalari (&vB)[3]) +{ + // A = y0 - y1 + // B = x1 - x0 + vA[0] = _simd_sub_epi32(vY[0], vY[1]); + vA[1] = _simd_sub_epi32(vY[1], vY[2]); + vA[2] = _simd_sub_epi32(vY[2], vY[0]); + + vB[0] = _simd_sub_epi32(vX[1], vX[0]); + vB[1] = _simd_sub_epi32(vX[2], vX[1]); + vB[2] = _simd_sub_epi32(vX[0], vX[2]); +} +// Calculate the determinant of the triangle +// 2 vectors between the 3 points: P, Q +// Px = x0-x2, Py = y0-y2 +// Qx = x1-x2, Qy = y1-y2 +// |Px Qx| +// det = | | = PxQy - PyQx +// |Py Qy| +// simplifies to : (x0-x2)*(y1-y2) - (y0-y2)*(x1-x2) +// try to reuse our A & B coef's already calculated. factor out a -1 from Py and Qx +// : B[2]*A[1] - (-(y2-y0))*(-(x2-x1)) +// : B[2]*A[1] - (-1)(-1)(y2-y0)*(x2-x1) +// : B[2]*A[1] - A[2]*B[1] +INLINE +float calcDeterminantInt(const __m128i vA, const __m128i vB) +{ + // vAShuf = [A1, A0, A2, A0] + __m128i vAShuf = _mm_shuffle_epi32(vA, _MM_SHUFFLE(0, 2, 0, 1)); + // vBShuf = [B2, B0, B1, B0] + __m128i vBShuf = _mm_shuffle_epi32(vB, _MM_SHUFFLE(0, 1, 0, 2)); + // vMul = [A1*B2, B1*A2] + __m128i vMul = _mm_mul_epi32(vAShuf, vBShuf); + + // shuffle upper to lower + // vMul2 = [B1*A2, B1*A2] + __m128i vMul2 = _mm_shuffle_epi32(vMul, _MM_SHUFFLE(3, 2, 3, 2)); + //vMul = [A1*B2 - B1*A2] + vMul = _mm_sub_epi64(vMul, vMul2); + + // According to emmintrin.h __mm_store1_pd(), address must be 16-byte aligned + OSALIGN(int64_t, 16) result; + _mm_store1_pd((double*)&result, _mm_castsi128_pd(vMul)); + + double fResult = (double)result; + fResult = fResult * (1.0 / FIXED_POINT16_SCALE); + + return (float)fResult; +} + +INLINE +void calcDeterminantIntVertical(const simdscalari vA[3], const simdscalari vB[3], simdscalari *pvDet) +{ + // refer to calcDeterminantInt comment for calculation explanation + // A1*B2 + simdscalari vA1Lo = _simd_unpacklo_epi32(vA[1], vA[1]); // 0 0 1 1 4 4 5 5 + simdscalari vA1Hi = _simd_unpackhi_epi32(vA[1], vA[1]); // 2 2 3 3 6 6 7 7 + + simdscalari vB2Lo = _simd_unpacklo_epi32(vB[2], vB[2]); + simdscalari vB2Hi = _simd_unpackhi_epi32(vB[2], vB[2]); + + simdscalari vA1B2Lo = _simd_mul_epi32(vA1Lo, vB2Lo); // 0 1 4 5 + simdscalari vA1B2Hi = _simd_mul_epi32(vA1Hi, vB2Hi); // 2 3 6 7 + + // B1*A2 + simdscalari vA2Lo = _simd_unpacklo_epi32(vA[2], vA[2]); + simdscalari vA2Hi = _simd_unpackhi_epi32(vA[2], vA[2]); + + simdscalari vB1Lo = _simd_unpacklo_epi32(vB[1], vB[1]); + simdscalari vB1Hi = _simd_unpackhi_epi32(vB[1], vB[1]); + + simdscalari vA2B1Lo = _simd_mul_epi32(vA2Lo, vB1Lo); + simdscalari vA2B1Hi = _simd_mul_epi32(vA2Hi, vB1Hi); + + // A1*B2 - A2*B1 + simdscalari detLo = _simd_sub_epi64(vA1B2Lo, vA2B1Lo); + simdscalari detHi = _simd_sub_epi64(vA1B2Hi, vA2B1Hi); + + // shuffle 0 1 4 5 -> 0 1 2 3 + simdscalari vResultLo = _mm256_permute2f128_si256(detLo, detHi, 0x20); + simdscalari vResultHi = _mm256_permute2f128_si256(detLo, detHi, 0x31); + + pvDet[0] = vResultLo; + pvDet[1] = vResultHi; +} + +INLINE +void triangleSetupC(const __m128 vX, const __m128 vY, const __m128 vA, const __m128 &vB, __m128 &vC) +{ + // C = -Ax - By + vC = _mm_mul_ps(vA, vX); + __m128 vCy = _mm_mul_ps(vB, vY); + vC = _mm_mul_ps(vC, _mm_set1_ps(-1.0f)); + vC = _mm_sub_ps(vC, vCy); +} + +INLINE +void viewportTransform(__m128 &vX, __m128 &vY, __m128 &vZ, const SWR_VIEWPORT_MATRIX &vpMatrix) +{ + vX = _mm_mul_ps(vX, _mm_set1_ps(vpMatrix.m00)); + vX = _mm_add_ps(vX, _mm_set1_ps(vpMatrix.m30)); + + vY = _mm_mul_ps(vY, _mm_set1_ps(vpMatrix.m11)); + vY = _mm_add_ps(vY, _mm_set1_ps(vpMatrix.m31)); + + vZ = _mm_mul_ps(vZ, _mm_set1_ps(vpMatrix.m22)); + vZ = _mm_add_ps(vZ, _mm_set1_ps(vpMatrix.m32)); +} + +template +INLINE +void viewportTransform(simdvector *v, const SWR_VIEWPORT_MATRIX & vpMatrix) +{ + simdscalar m00 = _simd_load1_ps(&vpMatrix.m00); + simdscalar m30 = _simd_load1_ps(&vpMatrix.m30); + simdscalar m11 = _simd_load1_ps(&vpMatrix.m11); + simdscalar m31 = _simd_load1_ps(&vpMatrix.m31); + simdscalar m22 = _simd_load1_ps(&vpMatrix.m22); + simdscalar m32 = _simd_load1_ps(&vpMatrix.m32); + + for (uint32_t i = 0; i < NumVerts; ++i) + { + v[i].x = _simd_fmadd_ps(v[i].x, m00, m30); + v[i].y = _simd_fmadd_ps(v[i].y, m11, m31); + v[i].z = _simd_fmadd_ps(v[i].z, m22, m32); + } +} + +INLINE +void calcBoundingBoxInt(const __m128i &vX, const __m128i &vY, BBOX &bbox) +{ + // Need horizontal fp min here + __m128i vX1 = _mm_shuffle_epi32(vX, _MM_SHUFFLE(3, 2, 0, 1)); + __m128i vX2 = _mm_shuffle_epi32(vX, _MM_SHUFFLE(3, 0, 1, 2)); + + __m128i vY1 = _mm_shuffle_epi32(vY, _MM_SHUFFLE(3, 2, 0, 1)); + __m128i vY2 = _mm_shuffle_epi32(vY, _MM_SHUFFLE(3, 0, 1, 2)); + + + __m128i vMinX = _mm_min_epi32(vX, vX1); + vMinX = _mm_min_epi32(vMinX, vX2); + + __m128i vMaxX = _mm_max_epi32(vX, vX1); + vMaxX = _mm_max_epi32(vMaxX, vX2); + + __m128i vMinY = _mm_min_epi32(vY, vY1); + vMinY = _mm_min_epi32(vMinY, vY2); + + __m128i vMaxY = _mm_max_epi32(vY, vY1); + vMaxY = _mm_max_epi32(vMaxY, vY2); + + bbox.left = _mm_extract_epi32(vMinX, 0); + bbox.right = _mm_extract_epi32(vMaxX, 0); + bbox.top = _mm_extract_epi32(vMinY, 0); + bbox.bottom = _mm_extract_epi32(vMaxY, 0); + +#if 0 + Jacob: A = _mm_shuffle_ps(X, Y, 0 0 0 0) +B = _mm_shuffle_ps(Z, W, 0 0 0 0) +A = _mm_shuffle_epi32(A, 3 0 3 0) +A = _mm_shuffle_ps(A, B, 1 0 1 0) +#endif + +} + +INLINE +void calcBoundingBoxIntVertical(const simdscalari (&vX)[3], const simdscalari (&vY)[3], simdBBox &bbox) +{ + simdscalari vMinX = vX[0]; + vMinX = _simd_min_epi32(vMinX, vX[1]); + vMinX = _simd_min_epi32(vMinX, vX[2]); + + simdscalari vMaxX = vX[0]; + vMaxX = _simd_max_epi32(vMaxX, vX[1]); + vMaxX = _simd_max_epi32(vMaxX, vX[2]); + + simdscalari vMinY = vY[0]; + vMinY = _simd_min_epi32(vMinY, vY[1]); + vMinY = _simd_min_epi32(vMinY, vY[2]); + + simdscalari vMaxY = vY[0]; + vMaxY = _simd_max_epi32(vMaxY, vY[1]); + vMaxY = _simd_max_epi32(vMaxY, vY[2]); + + bbox.left = vMinX; + bbox.right = vMaxX; + bbox.top = vMinY; + bbox.bottom = vMaxY; +} + +INLINE +bool CanUseSimplePoints(DRAW_CONTEXT *pDC) +{ + const API_STATE& state = GetApiState(pDC); + + return (state.rastState.sampleCount == SWR_MULTISAMPLE_1X && + state.rastState.pointSize == 1.0f && + !state.rastState.pointParam && + !state.rastState.pointSpriteEnable); +} + +uint32_t GetNumPrims(PRIMITIVE_TOPOLOGY mode, uint32_t numElements); +uint32_t NumVertsPerPrim(PRIMITIVE_TOPOLOGY topology, bool includeAdjVerts); + +// Templated Draw front-end function. All combinations of template parameter values are available +template +void ProcessDraw(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); + +void ProcessClear(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +void ProcessStoreTiles(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +void ProcessInvalidateTiles(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +void ProcessSync(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); +void ProcessQueryStats(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint32_t workerId, void *pUserData); + +struct PA_STATE_BASE; // forward decl +void BinTriangles(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector tri[3], uint32_t primMask, simdscalari primID); +void BinPoints(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[3], uint32_t primMask, simdscalari primID); +void BinLines(DRAW_CONTEXT *pDC, PA_STATE& pa, uint32_t workerId, simdvector prims[3], uint32_t primMask, simdscalari primID); + diff --git a/src/gallium/drivers/swr/rasterizer/core/knobs.h b/src/gallium/drivers/swr/rasterizer/core/knobs.h new file mode 100644 index 00000000000..d7feb86273d --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/knobs.h @@ -0,0 +1,142 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file knobs.h +* +* @brief Static (Compile-Time) Knobs for Core. +* +******************************************************************************/ +#pragma once + +#include +#include + +#define KNOB_ARCH_AVX 0 +#define KNOB_ARCH_AVX2 1 +#define KNOB_ARCH_AVX512 2 + +/////////////////////////////////////////////////////////////////////////////// +// Architecture validation +/////////////////////////////////////////////////////////////////////////////// +#if !defined(KNOB_ARCH) +#define KNOB_ARCH KNOB_ARCH_AVX +#endif + +#if (KNOB_ARCH == KNOB_ARCH_AVX) +#define KNOB_ARCH_ISA AVX +#define KNOB_ARCH_STR "AVX" +#define KNOB_SIMD_WIDTH 8 +#elif (KNOB_ARCH == KNOB_ARCH_AVX2) +#define KNOB_ARCH_ISA AVX2 +#define KNOB_ARCH_STR "AVX2" +#define KNOB_SIMD_WIDTH 8 +#elif (KNOB_ARCH == KNOB_ARCH_AVX512) +#define KNOB_ARCH_ISA AVX512F +#define KNOB_ARCH_STR "AVX512" +#define KNOB_SIMD_WIDTH 16 +#error "AVX512 not yet supported" +#else +#error "Unknown architecture" +#endif + +#define MAX_KNOB_ARCH_STR_LEN sizeof("AVX512_PLUS_PADDING") + +/////////////////////////////////////////////////////////////////////////////// +// Configuration knobs +/////////////////////////////////////////////////////////////////////////////// +#define KNOB_MAX_NUM_THREADS 256 // Supports up to dual-HSW-Xeon. + +// Maximum supported number of active vertex buffer streams +#define KNOB_NUM_STREAMS 32 + +// Maximum supported number of attributes per vertex +#define KNOB_NUM_ATTRIBUTES 38 + +// Maximum supported active viewports and scissors +#define KNOB_NUM_VIEWPORTS_SCISSORS 16 + +// Guardband range used by the clipper +#define KNOB_GUARDBAND_WIDTH 32768.0f +#define KNOB_GUARDBAND_HEIGHT 32768.0f + +/////////////////////////////// +// Macro tile configuration +/////////////////////////////// + +// raster tile dimensions +#define KNOB_TILE_X_DIM 8 +#define KNOB_TILE_X_DIM_SHIFT 3 +#define KNOB_TILE_Y_DIM 8 +#define KNOB_TILE_Y_DIM_SHIFT 3 + +// fixed macrotile pixel dimension for now, eventually will be +// dynamically set based on tile format and pixel size +#define KNOB_MACROTILE_X_DIM 64 +#define KNOB_MACROTILE_Y_DIM 64 +#define KNOB_MACROTILE_X_DIM_FIXED (KNOB_MACROTILE_X_DIM << 8) +#define KNOB_MACROTILE_Y_DIM_FIXED (KNOB_MACROTILE_Y_DIM << 8) +#define KNOB_MACROTILE_X_DIM_FIXED_SHIFT 14 +#define KNOB_MACROTILE_Y_DIM_FIXED_SHIFT 14 +#define KNOB_MACROTILE_X_DIM_IN_TILES (KNOB_MACROTILE_X_DIM >> KNOB_TILE_X_DIM_SHIFT) +#define KNOB_MACROTILE_Y_DIM_IN_TILES (KNOB_MACROTILE_Y_DIM >> KNOB_TILE_Y_DIM_SHIFT) + +// total # of hot tiles available. This should be enough to +// fully render a 16kx16k 128bpp render target +#define KNOB_NUM_HOT_TILES_X 256 +#define KNOB_NUM_HOT_TILES_Y 256 +#define KNOB_COLOR_HOT_TILE_FORMAT R32G32B32A32_FLOAT +#define KNOB_DEPTH_HOT_TILE_FORMAT R32_FLOAT +#define KNOB_STENCIL_HOT_TILE_FORMAT R8_UINT + +// Max scissor rectangle +#define KNOB_MAX_SCISSOR_X KNOB_NUM_HOT_TILES_X * KNOB_MACROTILE_X_DIM +#define KNOB_MAX_SCISSOR_Y KNOB_NUM_HOT_TILES_Y * KNOB_MACROTILE_Y_DIM + +#if KNOB_SIMD_WIDTH==8 && KNOB_TILE_X_DIM < 4 +#error "incompatible width/tile dimensions" +#endif + +#if KNOB_SIMD_WIDTH == 8 +#define SIMD_TILE_X_DIM 4 +#define SIMD_TILE_Y_DIM 2 +#else +#error "Invalid simd width" +#endif + +/////////////////////////////////////////////////////////////////////////////// +// Optimization knobs +/////////////////////////////////////////////////////////////////////////////// +#define KNOB_USE_FAST_SRGB TRUE + +// enables cut-aware primitive assembler +#define KNOB_ENABLE_CUT_AWARE_PA TRUE + +/////////////////////////////////////////////////////////////////////////////// +// Debug knobs +/////////////////////////////////////////////////////////////////////////////// +//#define KNOB_ENABLE_RDTSC + +// Set to 1 to use the dynamic KNOB_TOSS_XXXX knobs. +#if !defined(KNOB_ENABLE_TOSS_POINTS) +#define KNOB_ENABLE_TOSS_POINTS 0 +#endif + diff --git a/src/gallium/drivers/swr/rasterizer/core/knobs_init.h b/src/gallium/drivers/swr/rasterizer/core/knobs_init.h new file mode 100644 index 00000000000..3f19555557f --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/knobs_init.h @@ -0,0 +1,98 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file knobs_init.h +* +* @brief Dynamic Knobs Initialization for Core. +* +******************************************************************************/ +#pragma once + +#include +#include +#include +#include +#include + +// Assume the type is compatible with a 32-bit integer +template +static inline void ConvertEnvToKnob(const char* pOverride, T& knobValue) +{ + uint32_t value = 0; + if (sscanf(pOverride, "%u", &value)) + { + knobValue = static_cast(value); + } +} + +static inline void ConvertEnvToKnob(const char* pOverride, bool& knobValue) +{ + size_t len = strlen(pOverride); + if (len == 1) + { + auto c = tolower(pOverride[0]); + if (c == 'y' || c == 't' || c == '1') + { + knobValue = true; + return; + } + if (c == 'n' || c == 'f' || c == '0') + { + knobValue = false; + return; + } + } + + // Try converting to a number and casting to bool + uint32_t value = 0; + if (sscanf(pOverride, "%u", &value)) + { + knobValue = value != 0; + return; + } +} + +static inline void ConvertEnvToKnob(const char* pOverride, float& knobValue) +{ + float value = knobValue; + if (sscanf(pOverride, "%f", &value)) + { + knobValue = value; + } +} + +template +static inline void InitKnob(T& knob) +{ + + // TODO, read registry first + + // Second, read environment variables + const char* pOverride = getenv(knob.Name()); + + if (pOverride) + { + auto knobValue = knob.Value(); + ConvertEnvToKnob(pOverride, knobValue); + knob.Value(knobValue); + } +} diff --git a/src/gallium/drivers/swr/rasterizer/core/multisample.cpp b/src/gallium/drivers/swr/rasterizer/core/multisample.cpp new file mode 100644 index 00000000000..d51a546b063 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/multisample.cpp @@ -0,0 +1,51 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file multisample.cpp +* +******************************************************************************/ + +#include "multisample.h" + +const uint32_t MultisampleTraits::samplePosXi[2] {0xC0, 0x40}; +const uint32_t MultisampleTraits::samplePosYi[2] {0xC0, 0x40}; +const uint32_t MultisampleTraits::samplePosXi[4] {0x60, 0xE0, 0x20, 0xA0}; +const uint32_t MultisampleTraits::samplePosYi[4] {0x20, 0x60, 0xA0, 0xE0}; +const uint32_t MultisampleTraits::samplePosXi[8] {0x90, 0x70, 0xD0, 0x50, 0x30, 0x10, 0xB0, 0xF0}; +const uint32_t MultisampleTraits::samplePosYi[8] {0x50, 0xB0, 0x90, 0x30, 0xD0, 0x70, 0xF0, 0x10}; +const uint32_t MultisampleTraits::samplePosXi[16] +{0x90, 0x70, 0x50, 0xC0, 0x30, 0xA0, 0xD0, 0xB0, 0x60, 0x80, 0x40, 0x20, 0x00, 0xF0, 0xE0, 0x10}; +const uint32_t MultisampleTraits::samplePosYi[16] +{0x90, 0x50, 0xA0, 0x70, 0x60, 0xD0, 0xB0, 0x30, 0xE0, 0x10, 0x20, 0xC0, 0x80, 0x40, 0xF0, 0x00}; + +const float MultisampleTraits::samplePosX{0.5f}; +const float MultisampleTraits::samplePosY{0.5f}; +const float MultisampleTraits::samplePosX[2]{0.75f, 0.25f}; +const float MultisampleTraits::samplePosY[2]{0.75f, 0.25f}; +const float MultisampleTraits::samplePosX[4]{0.375f, 0.875, 0.125, 0.625}; +const float MultisampleTraits::samplePosY[4]{0.125, 0.375, 0.625, 0.875}; +const float MultisampleTraits::samplePosX[8]{0.5625, 0.4375, 0.8125, 0.3125, 0.1875, 0.0625, 0.6875, 0.9375}; +const float MultisampleTraits::samplePosY[8]{0.3125, 0.6875, 0.5625, 0.1875, 0.8125, 0.4375, 0.9375, 0.0625}; +const float MultisampleTraits::samplePosX[16] +{0.5625, 0.4375, 0.3125, 0.7500, 0.1875, 0.6250, 0.8125, 0.6875, 0.3750, 0.5000, 0.2500, 0.1250, 0.0000, 0.9375, 0.8750, 0.0625}; +const float MultisampleTraits::samplePosY[16] +{0.5625, 0.3125, 0.6250, 0.4375, 0.3750, 0.8125, 0.6875, 0.1875, 0.8750, 0.0625, 0.1250, 0.7500, 0.5000, 0.2500, 0.9375, 0.0000}; diff --git a/src/gallium/drivers/swr/rasterizer/core/multisample.h b/src/gallium/drivers/swr/rasterizer/core/multisample.h new file mode 100644 index 00000000000..4ae777e2fc5 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/multisample.h @@ -0,0 +1,620 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file multisample.h +* +******************************************************************************/ + +#pragma once + +#include "context.h" +#include "format_traits.h" + +INLINE +uint32_t GetNumSamples(SWR_MULTISAMPLE_COUNT sampleCount) +{ + static const uint32_t sampleCountLUT[SWR_MULTISAMPLE_TYPE_MAX] {1, 2, 4, 8, 16}; + assert(sampleCount < SWR_MULTISAMPLE_TYPE_MAX); + return sampleCountLUT[sampleCount]; +} + +INLINE +SWR_MULTISAMPLE_COUNT GetSampleCount(uint32_t numSamples) +{ + switch(numSamples) + { + case 1: return SWR_MULTISAMPLE_1X; + case 2: return SWR_MULTISAMPLE_2X; + case 4: return SWR_MULTISAMPLE_4X; + case 8: return SWR_MULTISAMPLE_8X; + case 16: return SWR_MULTISAMPLE_16X; + default: assert(0); return SWR_MULTISAMPLE_1X; + } +} + +// hardcoded offsets based on Direct3d standard multisample positions +// 8 x 8 pixel grid ranging from (0, 0) to (15, 15), with (0, 0) = UL pixel corner +// coords are 0.8 fixed point offsets from (0, 0) +template +struct MultisampleTraits +{ + INLINE static __m128i vXi(uint32_t sampleNum) = delete; + INLINE static __m128i vYi(uint32_t sampleNum) = delete; + INLINE static simdscalar vX(uint32_t sampleNum) = delete; + INLINE static simdscalar vY(uint32_t sampleNum) = delete; + INLINE static float X(uint32_t sampleNum) = delete; + INLINE static float Y(uint32_t sampleNum) = delete; + INLINE static __m128i TileSampleOffsetsX() = delete; + INLINE static __m128i TileSampleOffsetsY() = delete; + INLINE static uint32_t RasterTileColorOffset(uint32_t sampleNum) = delete; + INLINE static uint32_t RasterTileDepthOffset(uint32_t sampleNum) = delete; + INLINE static uint32_t RasterTileStencilOffset(uint32_t sampleNum) = delete; + INLINE static simdscalari FullSampleMask() = delete; + + static const uint32_t numSamples = 0; +}; + +template<> +struct MultisampleTraits +{ + INLINE static __m128i vXi(uint32_t sampleNum) + { + static const __m128i X = _mm_set1_epi32(samplePosXi); + return X; + } + + INLINE static __m128i vYi(uint32_t sampleNum) + { + static const __m128i Y = _mm_set1_epi32(samplePosYi); + return Y; + } + + INLINE static simdscalar vX(uint32_t sampleNum) + { + static const simdscalar X = _simd_set1_ps(0.5f); + return X; + } + + INLINE static simdscalar vY(uint32_t sampleNum) + { + static const simdscalar Y = _simd_set1_ps(0.5f); + return Y; + } + + INLINE static float X(uint32_t sampleNum) {return samplePosX;}; + INLINE static float Y(uint32_t sampleNum) {return samplePosY;}; + + INLINE static __m128i TileSampleOffsetsX() + { + static const uint32_t bboxLeftEdge = 0x80; + static const uint32_t bboxRightEdge = 0x80; + // BR, BL, UR, UL + static const __m128i tileSampleOffsetX = _mm_set_epi32(bboxRightEdge, bboxLeftEdge, bboxRightEdge, bboxLeftEdge); + return tileSampleOffsetX; + } + + INLINE static __m128i TileSampleOffsetsY() + { + static const uint32_t bboxTopEdge = 0x80; + static const uint32_t bboxBottomEdge = 0x80; + // BR, BL, UR, UL + static const __m128i tileSampleOffsetY = _mm_set_epi32(bboxBottomEdge, bboxBottomEdge, bboxTopEdge, bboxTopEdge); + return tileSampleOffsetY; + } + + INLINE static uint32_t RasterTileColorOffset(uint32_t sampleNum) + { + return 0; + } + + INLINE static uint32_t RasterTileDepthOffset(uint32_t sampleNum) + { + return 0; + } + + INLINE static uint32_t RasterTileStencilOffset(uint32_t sampleNum) + { + return 0; + } + + INLINE static simdscalari FullSampleMask(){return _simd_set1_epi32(0x1);}; + + static const uint32_t samplePosXi {0x80}; + static const uint32_t samplePosYi {0x80}; + static const float samplePosX; + static const float samplePosY; + static const uint32_t numSamples = 1; +}; + +template<> +struct MultisampleTraits +{ + INLINE static __m128i vXi(uint32_t sampleNum) + { + SWR_ASSERT(sampleNum < numSamples); + static const __m128i X[numSamples] {_mm_set1_epi32(samplePosXi[0]), _mm_set1_epi32(samplePosXi[1])}; + return X[sampleNum]; + } + + INLINE static __m128i vYi(uint32_t sampleNum) + { + SWR_ASSERT(sampleNum < numSamples); + static const __m128i Y[numSamples] {_mm_set1_epi32(samplePosYi[0]), _mm_set1_epi32(samplePosYi[1])}; + return Y[sampleNum]; + } + + INLINE static simdscalar vX(uint32_t sampleNum) + { + static const simdscalar X[numSamples] {_simd_set1_ps(0.75f), _simd_set1_ps(0.25f)}; + assert(sampleNum < numSamples); + return X[sampleNum]; + } + + INLINE static simdscalar vY(uint32_t sampleNum) + { + static const simdscalar Y[numSamples] {_simd_set1_ps(0.75f), _simd_set1_ps(0.25f)}; + assert(sampleNum < numSamples); + return Y[sampleNum]; + } + + INLINE static float X(uint32_t sampleNum) { SWR_ASSERT(sampleNum < numSamples); return samplePosX[sampleNum]; }; + INLINE static float Y(uint32_t sampleNum) { SWR_ASSERT(sampleNum < numSamples); return samplePosY[sampleNum]; }; + + INLINE static __m128i TileSampleOffsetsX() + { + static const uint32_t bboxLeftEdge = 0x40; + static const uint32_t bboxRightEdge = 0xC0; + // BR, BL, UR, UL + static const __m128i tileSampleOffsetX = _mm_set_epi32(bboxRightEdge, bboxLeftEdge, bboxRightEdge, bboxLeftEdge); + return tileSampleOffsetX; + } + + INLINE static __m128i TileSampleOffsetsY() + { + static const uint32_t bboxTopEdge = 0x40; + static const uint32_t bboxBottomEdge = 0xC0; + // BR, BL, UR, UL + static const __m128i tileSampleOffsetY = _mm_set_epi32(bboxBottomEdge, bboxBottomEdge, bboxTopEdge, bboxTopEdge); + return tileSampleOffsetY; + } + + INLINE static uint32_t RasterTileColorOffset(uint32_t sampleNum) + { + static const uint32_t RasterTileColorOffsets[numSamples] + { 0, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) + }; + assert(sampleNum < numSamples); + return RasterTileColorOffsets[sampleNum]; + } + + INLINE static uint32_t RasterTileDepthOffset(uint32_t sampleNum) + { + static const uint32_t RasterTileDepthOffsets[numSamples] + { 0, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) + }; + assert(sampleNum < numSamples); + return RasterTileDepthOffsets[sampleNum]; + } + + INLINE static uint32_t RasterTileStencilOffset(uint32_t sampleNum) + { + static const uint32_t RasterTileStencilOffsets[numSamples] + { 0, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) + }; + assert(sampleNum < numSamples); + return RasterTileStencilOffsets[sampleNum]; + } + + INLINE static simdscalari FullSampleMask() + { + static const simdscalari mask =_simd_set1_epi32(0x3); + return mask; + } + + static const uint32_t samplePosXi[2]; + static const uint32_t samplePosYi[2]; + static const float samplePosX[2]; + static const float samplePosY[2]; + static const uint32_t numSamples = 2; +}; + +template<> +struct MultisampleTraits +{ + INLINE static __m128i vXi(uint32_t sampleNum) + { + static const __m128i X[numSamples] + {_mm_set1_epi32(samplePosXi[0]), _mm_set1_epi32(samplePosXi[1]), _mm_set1_epi32(samplePosXi[2]), _mm_set1_epi32(samplePosXi[3])}; + SWR_ASSERT(sampleNum < numSamples); + return X[sampleNum]; + } + + INLINE static __m128i vYi(uint32_t sampleNum) + { + static const __m128i Y[numSamples] + {_mm_set1_epi32(samplePosYi[0]), _mm_set1_epi32(samplePosYi[1]), _mm_set1_epi32(samplePosYi[2]), _mm_set1_epi32(samplePosYi[3])}; + SWR_ASSERT(sampleNum < numSamples); + return Y[sampleNum]; + } + + INLINE static simdscalar vX(uint32_t sampleNum) + { + static const simdscalar X[numSamples] + {_simd_set1_ps(0.375f), _simd_set1_ps(0.875), _simd_set1_ps(0.125), _simd_set1_ps(0.625)}; + assert(sampleNum < numSamples); + return X[sampleNum]; + } + + INLINE static simdscalar vY(uint32_t sampleNum) + { + static const simdscalar Y[numSamples] + {_simd_set1_ps(0.125), _simd_set1_ps(0.375f), _simd_set1_ps(0.625), _simd_set1_ps(0.875)}; + assert(sampleNum < numSamples); + return Y[sampleNum]; + } + + INLINE static float X(uint32_t sampleNum) { SWR_ASSERT(sampleNum < numSamples); return samplePosX[sampleNum]; }; + INLINE static float Y(uint32_t sampleNum) { SWR_ASSERT(sampleNum < numSamples); return samplePosY[sampleNum]; }; + + INLINE static __m128i TileSampleOffsetsX() + { + static const uint32_t bboxLeftEdge = 0x20; + static const uint32_t bboxRightEdge = 0xE0; + // BR, BL, UR, UL + static const __m128i tileSampleOffsetX = _mm_set_epi32(bboxRightEdge, bboxLeftEdge, bboxRightEdge, bboxLeftEdge); + return tileSampleOffsetX; + } + + INLINE static __m128i TileSampleOffsetsY() + { + static const uint32_t bboxTopEdge = 0x20; + static const uint32_t bboxBottomEdge = 0xE0; + // BR, BL, UR, UL + static const __m128i tileSampleOffsetY = _mm_set_epi32(bboxBottomEdge, bboxBottomEdge, bboxTopEdge, bboxTopEdge); + return tileSampleOffsetY; + } + + INLINE static uint32_t RasterTileColorOffset(uint32_t sampleNum) + { + static const uint32_t RasterTileColorOffsets[numSamples] + { 0, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8), + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 2, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 3, + }; + assert(sampleNum < numSamples); + return RasterTileColorOffsets[sampleNum]; + } + + INLINE static uint32_t RasterTileDepthOffset(uint32_t sampleNum) + { + static const uint32_t RasterTileDepthOffsets[numSamples] + { 0, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8), + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 2, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 3, + }; + assert(sampleNum < numSamples); + return RasterTileDepthOffsets[sampleNum]; + } + + INLINE static uint32_t RasterTileStencilOffset(uint32_t sampleNum) + { + static const uint32_t RasterTileStencilOffsets[numSamples] + { 0, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8), + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 2, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 3, + }; + assert(sampleNum < numSamples); + return RasterTileStencilOffsets[sampleNum]; + } + + INLINE static simdscalari FullSampleMask() + { + static const simdscalari mask = _simd_set1_epi32(0xF); + return mask; + } + + static const uint32_t samplePosXi[4]; + static const uint32_t samplePosYi[4]; + static const float samplePosX[4]; + static const float samplePosY[4]; + static const uint32_t numSamples = 4; +}; + +template<> +struct MultisampleTraits +{ + INLINE static __m128i vXi(uint32_t sampleNum) + { + static const __m128i X[numSamples] + {_mm_set1_epi32(samplePosXi[0]), _mm_set1_epi32(samplePosXi[1]), _mm_set1_epi32(samplePosXi[2]), _mm_set1_epi32(samplePosXi[3]), + _mm_set1_epi32(samplePosXi[4]), _mm_set1_epi32(samplePosXi[5]), _mm_set1_epi32(samplePosXi[6]), _mm_set1_epi32(samplePosXi[7])}; + SWR_ASSERT(sampleNum < numSamples); + return X[sampleNum]; + } + + INLINE static __m128i vYi(uint32_t sampleNum) + { + static const __m128i Y[numSamples] + {_mm_set1_epi32(samplePosYi[0]), _mm_set1_epi32(samplePosYi[1]), _mm_set1_epi32(samplePosYi[2]), _mm_set1_epi32(samplePosYi[3]), + _mm_set1_epi32(samplePosYi[4]), _mm_set1_epi32(samplePosYi[5]), _mm_set1_epi32(samplePosYi[6]), _mm_set1_epi32(samplePosYi[7])}; + SWR_ASSERT(sampleNum < numSamples); + return Y[sampleNum]; + } + + INLINE static simdscalar vX(uint32_t sampleNum) + { + static const simdscalar X[numSamples] + {_simd_set1_ps(0.5625), _simd_set1_ps(0.4375), _simd_set1_ps(0.8125), _simd_set1_ps(0.3125), + _simd_set1_ps(0.1875), _simd_set1_ps(0.0625), _simd_set1_ps(0.6875), _simd_set1_ps(0.9375)}; + assert(sampleNum < numSamples); + return X[sampleNum]; + } + + INLINE static simdscalar vY(uint32_t sampleNum) + { + static const simdscalar Y[numSamples] + {_simd_set1_ps(0.3125), _simd_set1_ps(0.6875), _simd_set1_ps(0.5625), _simd_set1_ps(0.1875), + _simd_set1_ps(0.8125), _simd_set1_ps(0.4375), _simd_set1_ps(0.9375), _simd_set1_ps(0.0625)}; + assert(sampleNum < numSamples); + return Y[sampleNum]; + } + + INLINE static float X(uint32_t sampleNum) { SWR_ASSERT(sampleNum < numSamples); return samplePosX[sampleNum]; }; + INLINE static float Y(uint32_t sampleNum) { SWR_ASSERT(sampleNum < numSamples); return samplePosY[sampleNum]; }; + + INLINE static __m128i TileSampleOffsetsX() + { + static const uint32_t bboxLeftEdge = 0x10; + static const uint32_t bboxRightEdge = 0xF0; + // BR, BL, UR, UL + static const __m128i tileSampleOffsetX = _mm_set_epi32(bboxRightEdge, bboxLeftEdge, bboxRightEdge, bboxLeftEdge); + return tileSampleOffsetX; + } + + INLINE static __m128i TileSampleOffsetsY() + { + static const uint32_t bboxTopEdge = 0x10; + static const uint32_t bboxBottomEdge = 0xF0; + // BR, BL, UR, UL + static const __m128i tileSampleOffsetY = _mm_set_epi32(bboxBottomEdge, bboxBottomEdge, bboxTopEdge, bboxTopEdge); + return tileSampleOffsetY; + } + + INLINE static uint32_t RasterTileColorOffset(uint32_t sampleNum) + { + static const uint32_t RasterTileColorOffsets[numSamples] + { 0, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8), + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 2, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 3, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 4, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 5, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 6, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 7, + }; + assert(sampleNum < numSamples); + return RasterTileColorOffsets[sampleNum]; + } + + INLINE static uint32_t RasterTileDepthOffset(uint32_t sampleNum) + { + static const uint32_t RasterTileDepthOffsets[numSamples] + { 0, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8), + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 2, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 3, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 4, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 5, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 6, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 7, + }; + assert(sampleNum < numSamples); + return RasterTileDepthOffsets[sampleNum]; + } + + INLINE static uint32_t RasterTileStencilOffset(uint32_t sampleNum) + { + static const uint32_t RasterTileStencilOffsets[numSamples] + { 0, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8), + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 2, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 3, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 4, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 5, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 6, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 7, + }; + assert(sampleNum < numSamples); + return RasterTileStencilOffsets[sampleNum]; + } + + INLINE static simdscalari FullSampleMask() + { + static const simdscalari mask = _simd_set1_epi32(0xFF); + return mask; + } + + static const uint32_t samplePosXi[8]; + static const uint32_t samplePosYi[8]; + static const float samplePosX[8]; + static const float samplePosY[8]; + static const uint32_t numSamples = 8; +}; + +template<> +struct MultisampleTraits +{ + INLINE static __m128i vXi(uint32_t sampleNum) + { + static const __m128i X[numSamples] + {_mm_set1_epi32(samplePosXi[0]), _mm_set1_epi32(samplePosXi[1]), _mm_set1_epi32(samplePosXi[2]), _mm_set1_epi32(samplePosXi[3]), + _mm_set1_epi32(samplePosXi[4]), _mm_set1_epi32(samplePosXi[5]), _mm_set1_epi32(samplePosXi[6]), _mm_set1_epi32(samplePosXi[7]), + _mm_set1_epi32(samplePosXi[8]), _mm_set1_epi32(samplePosXi[9]), _mm_set1_epi32(samplePosXi[10]), _mm_set1_epi32(samplePosXi[11]), + _mm_set1_epi32(samplePosXi[12]), _mm_set1_epi32(samplePosXi[13]), _mm_set1_epi32(samplePosXi[14]), _mm_set1_epi32(samplePosXi[15])}; + SWR_ASSERT(sampleNum < numSamples); + return X[sampleNum]; + } + + INLINE static __m128i vYi(uint32_t sampleNum) + { + static const __m128i Y[numSamples] + {_mm_set1_epi32(samplePosYi[0]), _mm_set1_epi32(samplePosYi[1]), _mm_set1_epi32(samplePosYi[2]), _mm_set1_epi32(samplePosYi[3]), + _mm_set1_epi32(samplePosYi[4]), _mm_set1_epi32(samplePosYi[5]), _mm_set1_epi32(samplePosYi[6]), _mm_set1_epi32(samplePosYi[7]), + _mm_set1_epi32(samplePosYi[8]), _mm_set1_epi32(samplePosYi[9]), _mm_set1_epi32(samplePosYi[10]), _mm_set1_epi32(samplePosYi[11]), + _mm_set1_epi32(samplePosYi[12]), _mm_set1_epi32(samplePosYi[13]), _mm_set1_epi32(samplePosYi[14]), _mm_set1_epi32(samplePosYi[15])}; + SWR_ASSERT(sampleNum < numSamples); + return Y[sampleNum]; + } + + INLINE static simdscalar vX(uint32_t sampleNum) + { + static const simdscalar X[numSamples] + {_simd_set1_ps(0.5625), _simd_set1_ps(0.4375), _simd_set1_ps(0.3125), _simd_set1_ps(0.7500), + _simd_set1_ps(0.1875), _simd_set1_ps(0.6250), _simd_set1_ps(0.8125), _simd_set1_ps(0.6875), + _simd_set1_ps(0.3750), _simd_set1_ps(0.5000), _simd_set1_ps(0.2500), _simd_set1_ps(0.1250), + _simd_set1_ps(0.0000), _simd_set1_ps(0.9375), _simd_set1_ps(0.8750), _simd_set1_ps(0.0625)}; + assert(sampleNum < numSamples); + return X[sampleNum]; + } + + INLINE static simdscalar vY(uint32_t sampleNum) + { + static const simdscalar Y[numSamples] + {_simd_set1_ps(0.5625), _simd_set1_ps(0.3125), _simd_set1_ps(0.6250), _simd_set1_ps(0.4375), + _simd_set1_ps(0.3750), _simd_set1_ps(0.8125), _simd_set1_ps(0.6875), _simd_set1_ps(0.1875), + _simd_set1_ps(0.8750), _simd_set1_ps(0.0625), _simd_set1_ps(0.1250), _simd_set1_ps(0.7500), + _simd_set1_ps(0.5000), _simd_set1_ps(0.2500), _simd_set1_ps(0.9375), _simd_set1_ps(0.0000)}; + assert(sampleNum < numSamples); + return Y[sampleNum]; + } + + INLINE static float X(uint32_t sampleNum) { SWR_ASSERT(sampleNum < numSamples); return samplePosX[sampleNum]; }; + INLINE static float Y(uint32_t sampleNum) { SWR_ASSERT(sampleNum < numSamples); return samplePosY[sampleNum]; }; + + INLINE static __m128i TileSampleOffsetsX() + { + static const uint32_t bboxLeftEdge = 0x00; + static const uint32_t bboxRightEdge = 0xF0; + // BR, BL, UR, UL + static const __m128i tileSampleOffsetX = _mm_set_epi32(bboxRightEdge, bboxLeftEdge, bboxRightEdge, bboxLeftEdge); + return tileSampleOffsetX; + } + + INLINE static __m128i TileSampleOffsetsY() + { + static const uint32_t bboxTopEdge = 0x00; + static const uint32_t bboxBottomEdge = 0xF0; + // BR, BL, UR, UL + static const __m128i tileSampleOffsetY = _mm_set_epi32(bboxBottomEdge, bboxBottomEdge, bboxTopEdge, bboxTopEdge); + return tileSampleOffsetY; + } + + INLINE static uint32_t RasterTileColorOffset(uint32_t sampleNum) + { + static const uint32_t RasterTileColorOffsets[numSamples] + { 0, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8), + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 2, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 3, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 4, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 5, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 6, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 7, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 8, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 9, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 10, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 11, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 12, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 13, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 14, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 15, + }; + assert(sampleNum < numSamples); + return RasterTileColorOffsets[sampleNum]; + } + + INLINE static uint32_t RasterTileDepthOffset(uint32_t sampleNum) + { + static const uint32_t RasterTileDepthOffsets[numSamples] + { 0, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8), + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 2, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 3, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 4, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 5, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 6, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 7, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 8, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 9, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 10, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 11, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 12, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 13, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 14, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 15, + }; + assert(sampleNum < numSamples); + return RasterTileDepthOffsets[sampleNum]; + } + + INLINE static uint32_t RasterTileStencilOffset(uint32_t sampleNum) + { + static const uint32_t RasterTileStencilOffsets[numSamples] + { 0, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8), + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 2, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 3, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 4, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 5, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 6, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 7, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 8, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 9, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 10, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 11, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 12, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 13, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 14, + (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * FormatTraits::bpp / 8) * 15, + }; + assert(sampleNum < numSamples); + return RasterTileStencilOffsets[sampleNum]; + } + + INLINE static simdscalari FullSampleMask() + { + static const simdscalari mask = _simd_set1_epi32(0xFFFF); + return mask; + } + + static const uint32_t samplePosXi[16]; + static const uint32_t samplePosYi[16]; + static const float samplePosX[16]; + static const float samplePosY[16]; + static const uint32_t numSamples = 16; +}; diff --git a/src/gallium/drivers/swr/rasterizer/core/pa.h b/src/gallium/drivers/swr/rasterizer/core/pa.h new file mode 100644 index 00000000000..2028d9fbcfe --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/pa.h @@ -0,0 +1,1208 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file pa.h +* +* @brief Definitions for primitive assembly. +* N primitives are assembled at a time, where N is the SIMD width. +* A state machine, that is specific for a given topology, drives the +* assembly of vertices into triangles. +* +******************************************************************************/ +#pragma once + +#include "frontend.h" + +struct PA_STATE +{ + DRAW_CONTEXT *pDC; // draw context + uint8_t* pStreamBase; // vertex stream + uint32_t streamSizeInVerts; // total size of the input stream in verts + + // The topology the binner will use. In some cases the FE changes the topology from the api state. + PRIMITIVE_TOPOLOGY binTopology; + + PA_STATE() {} + PA_STATE(DRAW_CONTEXT *in_pDC, uint8_t* in_pStreamBase, uint32_t in_streamSizeInVerts) : + pDC(in_pDC), pStreamBase(in_pStreamBase), streamSizeInVerts(in_streamSizeInVerts) {} + + virtual bool HasWork() = 0; + virtual simdvector& GetSimdVector(uint32_t index, uint32_t slot) = 0; + virtual bool Assemble(uint32_t slot, simdvector verts[]) = 0; + virtual void AssembleSingle(uint32_t slot, uint32_t primIndex, __m128 verts[]) = 0; + virtual bool NextPrim() = 0; + virtual simdvertex& GetNextVsOutput() = 0; + virtual bool GetNextStreamOutput() = 0; + virtual simdmask& GetNextVsIndices() = 0; + virtual uint32_t NumPrims() = 0; + virtual void Reset() = 0; + virtual simdscalari GetPrimID(uint32_t startID) = 0; +}; + +// The Optimized PA is a state machine that assembles triangles from vertex shader simd +// output. Here is the sequence +// 1. Execute FS/VS to generate a simd vertex (4 vertices for SSE simd and 8 for AVX simd). +// 2. Execute PA function to assemble and bin triangles. +// a. The PA function is a set of functions that collectively make up the +// state machine for a given topology. +// 1. We use a state index to track which PA function to call. +// b. Often the PA function needs to 2 simd vertices in order to assemble the next triangle. +// 1. We call this the current and previous simd vertex. +// 2. The SSE simd is 4-wide which is not a multiple of 3 needed for triangles. In +// order to assemble the second triangle, for a triangle list, we'll need the +// last vertex from the previous simd and the first 2 vertices from the current simd. +// 3. At times the PA can assemble multiple triangles from the 2 simd vertices. +// +// This optimized PA is not cut aware, so only should be used by non-indexed draws or draws without +// cuts +struct PA_STATE_OPT : public PA_STATE +{ + simdvertex leadingVertex; // For tri-fan + uint32_t numPrims; // Total number of primitives for draw. + uint32_t numPrimsComplete; // Total number of complete primitives. + + uint32_t numSimdPrims; // Number of prims in current simd. + + uint32_t cur; // index to current VS output. + uint32_t prev; // index to prev VS output. Not really needed in the state. + uint32_t first; // index to first VS output. Used for trifan. + + uint32_t counter; // state counter + bool reset; // reset state + + uint32_t primIDIncr; // how much to increment for each vector (typically vector / {1, 2}) + simdscalari primID; + + typedef bool(*PFN_PA_FUNC)(PA_STATE_OPT& state, uint32_t slot, simdvector verts[]); + typedef void(*PFN_PA_SINGLE_FUNC)(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]); + + PFN_PA_FUNC pfnPaFunc; // PA state machine function for assembling 4 triangles. + PFN_PA_SINGLE_FUNC pfnPaSingleFunc; // PA state machine function for assembling single triangle. + PFN_PA_FUNC pfnPaFuncReset; // initial state to set on reset + + // state used to advance the PA when Next is called + PFN_PA_FUNC pfnPaNextFunc; + uint32_t nextNumSimdPrims; + uint32_t nextNumPrimsIncrement; + bool nextReset; + bool isStreaming; + + simdmask tmpIndices; // temporary index store for unused virtual function + + PA_STATE_OPT() {} + PA_STATE_OPT(DRAW_CONTEXT* pDC, uint32_t numPrims, uint8_t* pStream, uint32_t streamSizeInVerts, + bool in_isStreaming, PRIMITIVE_TOPOLOGY topo = TOP_UNKNOWN); + + bool HasWork() + { + return (this->numPrimsComplete < this->numPrims) ? true : false; + } + + simdvector& GetSimdVector(uint32_t index, uint32_t slot) + { + simdvertex* pVertex = (simdvertex*)pStreamBase; + return pVertex[index].attrib[slot]; + } + + // Assembles 4 triangles. Each simdvector is a single vertex from 4 + // triangles (xxxx yyyy zzzz wwww) and there are 3 verts per triangle. + bool Assemble(uint32_t slot, simdvector verts[]) + { + return this->pfnPaFunc(*this, slot, verts); + } + + // Assembles 1 primitive. Each simdscalar is a vertex (xyzw). + void AssembleSingle(uint32_t slot, uint32_t primIndex, __m128 verts[]) + { + return this->pfnPaSingleFunc(*this, slot, primIndex, verts); + } + + bool NextPrim() + { + this->pfnPaFunc = this->pfnPaNextFunc; + this->numSimdPrims = this->nextNumSimdPrims; + this->numPrimsComplete += this->nextNumPrimsIncrement; + this->reset = this->nextReset; + + if (this->isStreaming) + { + this->reset = false; + } + + bool morePrims = false; + + if (this->numSimdPrims > 0) + { + morePrims = true; + this->numSimdPrims--; + } + else + { + this->counter = (this->reset) ? 0 : (this->counter + 1); + this->reset = false; + } + + this->pfnPaFunc = this->pfnPaNextFunc; + + if (!HasWork()) + { + morePrims = false; // no more to do + } + + return morePrims; + } + + simdvertex& GetNextVsOutput() + { + // increment cur and prev indices + const uint32_t numSimdVerts = this->streamSizeInVerts / KNOB_SIMD_WIDTH; + this->prev = this->cur; // prev is undefined for first state. + this->cur = this->counter % numSimdVerts; + + simdvertex* pVertex = (simdvertex*)pStreamBase; + return pVertex[this->cur]; + } + + simdmask& GetNextVsIndices() + { + // unused in optimized PA, pass tmp buffer back + return tmpIndices; + } + + bool GetNextStreamOutput() + { + this->prev = this->cur; + this->cur = this->counter; + + return HasWork(); + } + + uint32_t NumPrims() + { + return (this->numPrimsComplete + this->nextNumPrimsIncrement > this->numPrims) ? + (KNOB_SIMD_WIDTH - (this->numPrimsComplete + this->nextNumPrimsIncrement - this->numPrims)) : KNOB_SIMD_WIDTH; + } + + void SetNextState(PA_STATE_OPT::PFN_PA_FUNC pfnPaNextFunc, + PA_STATE_OPT::PFN_PA_SINGLE_FUNC pfnPaNextSingleFunc, + uint32_t numSimdPrims = 0, + uint32_t numPrimsIncrement = 0, + bool reset = false) + { + this->pfnPaNextFunc = pfnPaNextFunc; + this->nextNumSimdPrims = numSimdPrims; + this->nextNumPrimsIncrement = numPrimsIncrement; + this->nextReset = reset; + + this->pfnPaSingleFunc = pfnPaNextSingleFunc; + } + + void Reset() + { + this->pfnPaFunc = this->pfnPaFuncReset; + this->numPrimsComplete = 0; + this->numSimdPrims = 0; + this->cur = 0; + this->prev = 0; + this->first = 0; + this->counter = 0; + this->reset = false; + } + + simdscalari GetPrimID(uint32_t startID) + { + return _simd_add_epi32(this->primID, + _simd_set1_epi32(startID + this->primIDIncr * (this->numPrimsComplete / KNOB_SIMD_WIDTH))); + } +}; + +// helper C wrappers to avoid having to rewrite all the PA topology state functions +INLINE void SetNextPaState(PA_STATE_OPT& pa, PA_STATE_OPT::PFN_PA_FUNC pfnPaNextFunc, + PA_STATE_OPT::PFN_PA_SINGLE_FUNC pfnPaNextSingleFunc, + uint32_t numSimdPrims = 0, + uint32_t numPrimsIncrement = 0, + bool reset = false) +{ + return pa.SetNextState(pfnPaNextFunc, pfnPaNextSingleFunc, numSimdPrims, numPrimsIncrement, reset); +} +INLINE simdvector& PaGetSimdVector(PA_STATE& pa, uint32_t index, uint32_t slot) +{ + return pa.GetSimdVector(index, slot); +} + +INLINE __m128 swizzleLane0(const simdvector &a) +{ + simdscalar tmp0 = _mm256_unpacklo_ps(a.x, a.z); + simdscalar tmp1 = _mm256_unpacklo_ps(a.y, a.w); + return _mm256_extractf128_ps(_mm256_unpacklo_ps(tmp0, tmp1), 0); +} + +INLINE __m128 swizzleLane1(const simdvector &a) +{ + simdscalar tmp0 = _mm256_unpacklo_ps(a.x, a.z); + simdscalar tmp1 = _mm256_unpacklo_ps(a.y, a.w); + return _mm256_extractf128_ps(_mm256_unpackhi_ps(tmp0, tmp1), 0); +} + +INLINE __m128 swizzleLane2(const simdvector &a) +{ + simdscalar tmp0 = _mm256_unpackhi_ps(a.x, a.z); + simdscalar tmp1 = _mm256_unpackhi_ps(a.y, a.w); + return _mm256_extractf128_ps(_mm256_unpacklo_ps(tmp0, tmp1), 0); +} + +INLINE __m128 swizzleLane3(const simdvector &a) +{ + simdscalar tmp0 = _mm256_unpackhi_ps(a.x, a.z); + simdscalar tmp1 = _mm256_unpackhi_ps(a.y, a.w); + return _mm256_extractf128_ps(_mm256_unpackhi_ps(tmp0, tmp1), 0); +} + +INLINE __m128 swizzleLane4(const simdvector &a) +{ + simdscalar tmp0 = _mm256_unpacklo_ps(a.x, a.z); + simdscalar tmp1 = _mm256_unpacklo_ps(a.y, a.w); + return _mm256_extractf128_ps(_mm256_unpacklo_ps(tmp0, tmp1), 1); + +} + +INLINE __m128 swizzleLane5(const simdvector &a) +{ + simdscalar tmp0 = _mm256_unpacklo_ps(a.x, a.z); + simdscalar tmp1 = _mm256_unpacklo_ps(a.y, a.w); + return _mm256_extractf128_ps(_mm256_unpackhi_ps(tmp0, tmp1), 1); +} + +INLINE __m128 swizzleLane6(const simdvector &a) +{ + simdscalar tmp0 = _mm256_unpackhi_ps(a.x, a.z); + simdscalar tmp1 = _mm256_unpackhi_ps(a.y, a.w); + return _mm256_extractf128_ps(_mm256_unpacklo_ps(tmp0, tmp1), 1); +} + +INLINE __m128 swizzleLane7(const simdvector &a) +{ + simdscalar tmp0 = _mm256_unpackhi_ps(a.x, a.z); + simdscalar tmp1 = _mm256_unpackhi_ps(a.y, a.w); + return _mm256_extractf128_ps(_mm256_unpackhi_ps(tmp0, tmp1), 1); +} + +INLINE __m128 swizzleLaneN(const simdvector &a, int lane) +{ + switch (lane) { + case 0: + return swizzleLane0(a); + case 1: + return swizzleLane1(a); + case 2: + return swizzleLane2(a); + case 3: + return swizzleLane3(a); + case 4: + return swizzleLane4(a); + case 5: + return swizzleLane5(a); + case 6: + return swizzleLane6(a); + case 7: + return swizzleLane7(a); + default: + return _mm_setzero_ps(); + } +} + +// Cut-aware primitive assembler. +struct PA_STATE_CUT : public PA_STATE +{ + simdmask* pCutIndices; // cut indices buffer, 1 bit per vertex + uint32_t numVerts; // number of vertices available in buffer store + uint32_t numAttribs; // number of attributes + int32_t numRemainingVerts; // number of verts remaining to be assembled + uint32_t numVertsToAssemble; // total number of verts to assemble for the draw + OSALIGNSIMD(uint32_t) indices[MAX_NUM_VERTS_PER_PRIM][KNOB_SIMD_WIDTH]; // current index buffer for gather + simdscalari vOffsets[MAX_NUM_VERTS_PER_PRIM]; // byte offsets for currently assembling simd + uint32_t numPrimsAssembled; // number of primitives that are fully assembled + uint32_t headVertex; // current unused vertex slot in vertex buffer store + uint32_t tailVertex; // beginning vertex currently assembling + uint32_t curVertex; // current unprocessed vertex + uint32_t startPrimId; // starting prim id + simdscalari vPrimId; // vector of prim ID + bool needOffsets; // need to compute gather offsets for current SIMD + uint32_t vertsPerPrim; + simdvertex tmpVertex; // temporary simdvertex for unimplemented API + bool processCutVerts; // vertex indices with cuts should be processed as normal, otherwise they + // are ignored. Fetch shader sends invalid verts on cuts that should be ignored + // while the GS sends valid verts for every index + // Topology state tracking + uint32_t vert[MAX_NUM_VERTS_PER_PRIM]; + uint32_t curIndex; + bool reverseWinding; // indicates reverse winding for strips + int32_t adjExtraVert; // extra vert uses for tristrip w/ adj + + typedef void(PA_STATE_CUT::* PFN_PA_FUNC)(uint32_t vert, bool finish); + PFN_PA_FUNC pfnPa; // per-topology function that processes a single vert + + PA_STATE_CUT() {} + PA_STATE_CUT(DRAW_CONTEXT* pDC, uint8_t* in_pStream, uint32_t in_streamSizeInVerts, simdmask* in_pIndices, uint32_t in_numVerts, + uint32_t in_numAttribs, PRIMITIVE_TOPOLOGY topo, bool in_processCutVerts) + : PA_STATE(pDC, in_pStream, in_streamSizeInVerts) + { + numVerts = in_streamSizeInVerts; + numAttribs = in_numAttribs; + binTopology = topo; + needOffsets = false; + processCutVerts = in_processCutVerts; + + numVertsToAssemble = numRemainingVerts = in_numVerts; + numPrimsAssembled = 0; + headVertex = tailVertex = curVertex = 0; + + curIndex = 0; + pCutIndices = in_pIndices; + memset(indices, 0, sizeof(indices)); + vPrimId = _mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0); + reverseWinding = false; + adjExtraVert = -1; + + bool gsEnabled = pDC->pState->state.gsState.gsEnable; + vertsPerPrim = NumVertsPerPrim(topo, gsEnabled); + + switch (topo) + { + case TOP_TRIANGLE_LIST: pfnPa = &PA_STATE_CUT::ProcessVertTriList; break; + case TOP_TRI_LIST_ADJ: pfnPa = gsEnabled ? &PA_STATE_CUT::ProcessVertTriListAdj : &PA_STATE_CUT::ProcessVertTriListAdjNoGs; break; + case TOP_TRIANGLE_STRIP: pfnPa = &PA_STATE_CUT::ProcessVertTriStrip; break; + case TOP_TRI_STRIP_ADJ: if (gsEnabled) + { + pfnPa = &PA_STATE_CUT::ProcessVertTriStripAdj < true > ; + } + else + { + pfnPa = &PA_STATE_CUT::ProcessVertTriStripAdj < false > ; + } + break; + + case TOP_POINT_LIST: pfnPa = &PA_STATE_CUT::ProcessVertPointList; break; + case TOP_LINE_LIST: pfnPa = &PA_STATE_CUT::ProcessVertLineList; break; + case TOP_LINE_LIST_ADJ: pfnPa = gsEnabled ? &PA_STATE_CUT::ProcessVertLineListAdj : &PA_STATE_CUT::ProcessVertLineListAdjNoGs; break; + case TOP_LINE_STRIP: pfnPa = &PA_STATE_CUT::ProcessVertLineStrip; break; + case TOP_LISTSTRIP_ADJ: pfnPa = gsEnabled ? &PA_STATE_CUT::ProcessVertLineStripAdj : &PA_STATE_CUT::ProcessVertLineStripAdjNoGs; break; + default: assert(0 && "Unimplemented topology"); + } + } + + simdvertex& GetNextVsOutput() + { + uint32_t vertexIndex = this->headVertex / KNOB_SIMD_WIDTH; + this->headVertex = (this->headVertex + KNOB_SIMD_WIDTH) % this->numVerts; + this->needOffsets = true; + return ((simdvertex*)pStreamBase)[vertexIndex]; + } + + simdmask& GetNextVsIndices() + { + uint32_t vertexIndex = this->headVertex / KNOB_SIMD_WIDTH; + simdmask* pCurCutIndex = this->pCutIndices + vertexIndex; + return *pCurCutIndex; + } + + simdvector& GetSimdVector(uint32_t index, uint32_t slot) + { + // unused + SWR_ASSERT(0 && "Not implemented"); + return this->tmpVertex.attrib[0]; + } + + bool GetNextStreamOutput() + { + this->headVertex += KNOB_SIMD_WIDTH; + this->needOffsets = true; + return HasWork(); + } + + simdscalari GetPrimID(uint32_t startID) + { + return _simd_add_epi32(_simd_set1_epi32(startID), this->vPrimId); + } + + void Reset() + { + this->numRemainingVerts = this->numVertsToAssemble; + this->numPrimsAssembled = 0; + this->curIndex = 0; + this->curVertex = 0; + this->tailVertex = 0; + this->headVertex = 0; + this->reverseWinding = false; + this->adjExtraVert = -1; + this->vPrimId = _mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0); + } + + bool HasWork() + { + return this->numRemainingVerts > 0 || this->adjExtraVert != -1; + } + + bool IsVertexStoreFull() + { + return ((this->headVertex + KNOB_SIMD_WIDTH) % this->numVerts) == this->tailVertex; + } + + void RestartTopology() + { + this->curIndex = 0; + this->reverseWinding = false; + this->adjExtraVert = -1; + } + + bool IsCutIndex(uint32_t vertex) + { + uint32_t vertexIndex = vertex / KNOB_SIMD_WIDTH; + uint32_t vertexOffset = vertex & (KNOB_SIMD_WIDTH - 1); + return _bittest((const LONG*)&this->pCutIndices[vertexIndex], vertexOffset) == 1; + } + + // iterates across the unprocessed verts until we hit the end or we + // have assembled SIMD prims + void ProcessVerts() + { + while (this->numPrimsAssembled != KNOB_SIMD_WIDTH && + this->numRemainingVerts > 0 && + this->curVertex != this->headVertex) + { + // if cut index, restart topology + if (IsCutIndex(this->curVertex)) + { + if (this->processCutVerts) + { + (this->*pfnPa)(this->curVertex, false); + } + // finish off tri strip w/ adj before restarting topo + if (this->adjExtraVert != -1) + { + (this->*pfnPa)(this->curVertex, true); + } + RestartTopology(); + } + else + { + (this->*pfnPa)(this->curVertex, false); + } + + this->curVertex = (this->curVertex + 1) % this->numVerts; + this->numRemainingVerts--; + } + + // special case last primitive for tri strip w/ adj + if (this->numPrimsAssembled != KNOB_SIMD_WIDTH && this->numRemainingVerts == 0 && this->adjExtraVert != -1) + { + (this->*pfnPa)(this->curVertex, true); + } + } + + void Advance() + { + // done with current batch + // advance tail to the current unsubmitted vertex + this->tailVertex = this->curVertex; + this->numPrimsAssembled = 0; + this->vPrimId = _simd_add_epi32(vPrimId, _simd_set1_epi32(KNOB_SIMD_WIDTH)); + } + + bool NextPrim() + { + // if we've assembled enough prims, we can advance to the next set of verts + if (this->numPrimsAssembled == KNOB_SIMD_WIDTH || this->numRemainingVerts <= 0) + { + Advance(); + } + return false; + } + + void ComputeOffsets() + { + for (uint32_t v = 0; v < this->vertsPerPrim; ++v) + { + simdscalari vIndices = *(simdscalari*)&this->indices[v][0]; + + // step to simdvertex batch + const uint32_t simdShift = 3; // @todo make knob + simdscalari vVertexBatch = _simd_srai_epi32(vIndices, simdShift); + this->vOffsets[v] = _simd_mullo_epi32(vVertexBatch, _simd_set1_epi32(sizeof(simdvertex))); + + // step to index + const uint32_t simdMask = 0x7; // @todo make knob + simdscalari vVertexIndex = _simd_and_si(vIndices, _simd_set1_epi32(simdMask)); + this->vOffsets[v] = _simd_add_epi32(this->vOffsets[v], _simd_mullo_epi32(vVertexIndex, _simd_set1_epi32(sizeof(float)))); + } + } + + bool Assemble(uint32_t slot, simdvector result[]) + { + // process any outstanding verts + ProcessVerts(); + + // return false if we don't have enough prims assembled + if (this->numPrimsAssembled != KNOB_SIMD_WIDTH && this->numRemainingVerts > 0) + { + return false; + } + + // cache off gather offsets given the current SIMD set of indices the first time we get an assemble + if (this->needOffsets) + { + ComputeOffsets(); + this->needOffsets = false; + } + + for (uint32_t v = 0; v < this->vertsPerPrim; ++v) + { + simdscalari offsets = this->vOffsets[v]; + + // step to attribute + offsets = _simd_add_epi32(offsets, _simd_set1_epi32(slot * sizeof(simdvector))); + + float* pBase = (float*)this->pStreamBase; + for (uint32_t c = 0; c < 4; ++c) + { + result[v].v[c] = _simd_i32gather_ps(pBase, offsets, 1); + + // move base to next component + pBase += KNOB_SIMD_WIDTH; + } + } + + return true; + } + + void AssembleSingle(uint32_t slot, uint32_t triIndex, __m128 tri[3]) + { + // move to slot + for (uint32_t v = 0; v < this->vertsPerPrim; ++v) + { + uint32_t* pOffset = (uint32_t*)&this->vOffsets[v]; + uint32_t offset = pOffset[triIndex]; + offset += sizeof(simdvector) * slot; + float* pVert = (float*)&tri[v]; + for (uint32_t c = 0; c < 4; ++c) + { + float* pComponent = (float*)(this->pStreamBase + offset); + pVert[c] = *pComponent; + offset += KNOB_SIMD_WIDTH * sizeof(float); + } + } + } + + uint32_t NumPrims() + { + return this->numPrimsAssembled; + } + + // Per-topology functions + void ProcessVertTriStrip(uint32_t index, bool finish) + { + this->vert[this->curIndex] = index; + this->curIndex++; + if (this->curIndex == 3) + { + // assembled enough verts for prim, add to gather indices + this->indices[0][this->numPrimsAssembled] = this->vert[0]; + if (reverseWinding) + { + this->indices[1][this->numPrimsAssembled] = this->vert[2]; + this->indices[2][this->numPrimsAssembled] = this->vert[1]; + } + else + { + this->indices[1][this->numPrimsAssembled] = this->vert[1]; + this->indices[2][this->numPrimsAssembled] = this->vert[2]; + } + + // increment numPrimsAssembled + this->numPrimsAssembled++; + + // set up next prim state + this->vert[0] = this->vert[1]; + this->vert[1] = this->vert[2]; + this->curIndex = 2; + this->reverseWinding ^= 1; + } + } + + template + void AssembleTriStripAdj() + { + if (!gsEnabled) + { + this->vert[1] = this->vert[2]; + this->vert[2] = this->vert[4]; + + this->indices[0][this->numPrimsAssembled] = this->vert[0]; + this->indices[1][this->numPrimsAssembled] = this->vert[1]; + this->indices[2][this->numPrimsAssembled] = this->vert[2]; + + this->vert[4] = this->vert[2]; + this->vert[2] = this->vert[1]; + } + else + { + this->indices[0][this->numPrimsAssembled] = this->vert[0]; + this->indices[1][this->numPrimsAssembled] = this->vert[1]; + this->indices[2][this->numPrimsAssembled] = this->vert[2]; + this->indices[3][this->numPrimsAssembled] = this->vert[3]; + this->indices[4][this->numPrimsAssembled] = this->vert[4]; + this->indices[5][this->numPrimsAssembled] = this->vert[5]; + } + this->numPrimsAssembled++; + } + + + template + void ProcessVertTriStripAdj(uint32_t index, bool finish) + { + // handle last primitive of tristrip + if (finish && this->adjExtraVert != -1) + { + this->vert[3] = this->adjExtraVert; + AssembleTriStripAdj(); + this->adjExtraVert = -1; + return; + } + + switch (this->curIndex) + { + case 0: + case 1: + case 2: + case 4: + this->vert[this->curIndex] = index; + this->curIndex++; + break; + case 3: + this->vert[5] = index; + this->curIndex++; + break; + case 5: + if (this->adjExtraVert == -1) + { + this->adjExtraVert = index; + } + else + { + this->vert[3] = index; + if (!gsEnabled) + { + AssembleTriStripAdj(); + + uint32_t nextTri[6]; + if (this->reverseWinding) + { + nextTri[0] = this->vert[4]; + nextTri[1] = this->vert[0]; + nextTri[2] = this->vert[2]; + nextTri[4] = this->vert[3]; + nextTri[5] = this->adjExtraVert; + } + else + { + nextTri[0] = this->vert[2]; + nextTri[1] = this->adjExtraVert; + nextTri[2] = this->vert[3]; + nextTri[4] = this->vert[4]; + nextTri[5] = this->vert[0]; + } + for (uint32_t i = 0; i < 6; ++i) + { + this->vert[i] = nextTri[i]; + } + + this->adjExtraVert = -1; + this->reverseWinding ^= 1; + } + else + { + this->curIndex++; + } + } + break; + case 6: + SWR_ASSERT(this->adjExtraVert != -1, "Algorith failure!"); + AssembleTriStripAdj(); + + uint32_t nextTri[6]; + if (this->reverseWinding) + { + nextTri[0] = this->vert[4]; + nextTri[1] = this->vert[0]; + nextTri[2] = this->vert[2]; + nextTri[4] = this->vert[3]; + nextTri[5] = this->adjExtraVert; + } + else + { + nextTri[0] = this->vert[2]; + nextTri[1] = this->adjExtraVert; + nextTri[2] = this->vert[3]; + nextTri[4] = this->vert[4]; + nextTri[5] = this->vert[0]; + } + for (uint32_t i = 0; i < 6; ++i) + { + this->vert[i] = nextTri[i]; + } + this->reverseWinding ^= 1; + this->adjExtraVert = index; + this->curIndex--; + break; + } + } + + void ProcessVertTriList(uint32_t index, bool finish) + { + this->vert[this->curIndex] = index; + this->curIndex++; + if (this->curIndex == 3) + { + // assembled enough verts for prim, add to gather indices + this->indices[0][this->numPrimsAssembled] = this->vert[0]; + this->indices[1][this->numPrimsAssembled] = this->vert[1]; + this->indices[2][this->numPrimsAssembled] = this->vert[2]; + + // increment numPrimsAssembled + this->numPrimsAssembled++; + + // set up next prim state + this->curIndex = 0; + } + } + + void ProcessVertTriListAdj(uint32_t index, bool finish) + { + this->vert[this->curIndex] = index; + this->curIndex++; + if (this->curIndex == 6) + { + // assembled enough verts for prim, add to gather indices + this->indices[0][this->numPrimsAssembled] = this->vert[0]; + this->indices[1][this->numPrimsAssembled] = this->vert[1]; + this->indices[2][this->numPrimsAssembled] = this->vert[2]; + this->indices[3][this->numPrimsAssembled] = this->vert[3]; + this->indices[4][this->numPrimsAssembled] = this->vert[4]; + this->indices[5][this->numPrimsAssembled] = this->vert[5]; + + // increment numPrimsAssembled + this->numPrimsAssembled++; + + // set up next prim state + this->curIndex = 0; + } + } + + void ProcessVertTriListAdjNoGs(uint32_t index, bool finish) + { + this->vert[this->curIndex] = index; + this->curIndex++; + if (this->curIndex == 6) + { + // assembled enough verts for prim, add to gather indices + this->indices[0][this->numPrimsAssembled] = this->vert[0]; + this->indices[1][this->numPrimsAssembled] = this->vert[2]; + this->indices[2][this->numPrimsAssembled] = this->vert[4]; + + // increment numPrimsAssembled + this->numPrimsAssembled++; + + // set up next prim state + this->curIndex = 0; + } + } + + + void ProcessVertLineList(uint32_t index, bool finish) + { + this->vert[this->curIndex] = index; + this->curIndex++; + if (this->curIndex == 2) + { + this->indices[0][this->numPrimsAssembled] = this->vert[0]; + this->indices[1][this->numPrimsAssembled] = this->vert[1]; + + this->numPrimsAssembled++; + this->curIndex = 0; + } + } + + void ProcessVertLineStrip(uint32_t index, bool finish) + { + this->vert[this->curIndex] = index; + this->curIndex++; + if (this->curIndex == 2) + { + // assembled enough verts for prim, add to gather indices + this->indices[0][this->numPrimsAssembled] = this->vert[0]; + this->indices[1][this->numPrimsAssembled] = this->vert[1]; + + // increment numPrimsAssembled + this->numPrimsAssembled++; + + // set up next prim state + this->vert[0] = this->vert[1]; + this->curIndex = 1; + } + } + + void ProcessVertLineStripAdj(uint32_t index, bool finish) + { + this->vert[this->curIndex] = index; + this->curIndex++; + if (this->curIndex == 4) + { + // assembled enough verts for prim, add to gather indices + this->indices[0][this->numPrimsAssembled] = this->vert[0]; + this->indices[1][this->numPrimsAssembled] = this->vert[1]; + this->indices[2][this->numPrimsAssembled] = this->vert[2]; + this->indices[3][this->numPrimsAssembled] = this->vert[3]; + + // increment numPrimsAssembled + this->numPrimsAssembled++; + + // set up next prim state + this->vert[0] = this->vert[1]; + this->vert[1] = this->vert[2]; + this->vert[2] = this->vert[3]; + this->curIndex = 3; + } + } + + void ProcessVertLineStripAdjNoGs(uint32_t index, bool finish) + { + this->vert[this->curIndex] = index; + this->curIndex++; + if (this->curIndex == 4) + { + // assembled enough verts for prim, add to gather indices + this->indices[0][this->numPrimsAssembled] = this->vert[1]; + this->indices[1][this->numPrimsAssembled] = this->vert[2]; + + // increment numPrimsAssembled + this->numPrimsAssembled++; + + // set up next prim state + this->vert[0] = this->vert[1]; + this->vert[1] = this->vert[2]; + this->vert[2] = this->vert[3]; + this->curIndex = 3; + } + } + + void ProcessVertLineListAdj(uint32_t index, bool finish) + { + this->vert[this->curIndex] = index; + this->curIndex++; + if (this->curIndex == 4) + { + this->indices[0][this->numPrimsAssembled] = this->vert[0]; + this->indices[1][this->numPrimsAssembled] = this->vert[1]; + this->indices[2][this->numPrimsAssembled] = this->vert[2]; + this->indices[3][this->numPrimsAssembled] = this->vert[3]; + + this->numPrimsAssembled++; + this->curIndex = 0; + } + } + + void ProcessVertLineListAdjNoGs(uint32_t index, bool finish) + { + this->vert[this->curIndex] = index; + this->curIndex++; + if (this->curIndex == 4) + { + this->indices[0][this->numPrimsAssembled] = this->vert[1]; + this->indices[1][this->numPrimsAssembled] = this->vert[2]; + + this->numPrimsAssembled++; + this->curIndex = 0; + } + } + + void ProcessVertPointList(uint32_t index, bool finish) + { + this->vert[this->curIndex] = index; + this->curIndex++; + if (this->curIndex == 1) + { + this->indices[0][this->numPrimsAssembled] = this->vert[0]; + this->numPrimsAssembled++; + this->curIndex = 0; + } + } +}; + +// Primitive Assembly for data output from the DomainShader. +struct PA_TESS : PA_STATE +{ + PA_TESS( + DRAW_CONTEXT *in_pDC, + const simdscalar* in_pVertData, + uint32_t in_attributeStrideInVectors, + uint32_t in_numAttributes, + uint32_t* (&in_ppIndices)[3], + uint32_t in_numPrims, + PRIMITIVE_TOPOLOGY in_binTopology) : + + PA_STATE(in_pDC, nullptr, 0), + m_pVertexData(in_pVertData), + m_attributeStrideInVectors(in_attributeStrideInVectors), + m_numAttributes(in_numAttributes), + m_numPrims(in_numPrims) + { + m_vPrimId = _simd_setzero_si(); + binTopology = in_binTopology; + m_ppIndices[0] = in_ppIndices[0]; + m_ppIndices[1] = in_ppIndices[1]; + m_ppIndices[2] = in_ppIndices[2]; + + switch (binTopology) + { + case TOP_POINT_LIST: + m_numVertsPerPrim = 1; + break; + + case TOP_LINE_LIST: + m_numVertsPerPrim = 2; + break; + + case TOP_TRIANGLE_LIST: + m_numVertsPerPrim = 3; + break; + + default: + SWR_ASSERT(0, "Invalid binTopology (%d) for %s", binTopology, __FUNCTION__); + break; + } + } + + bool HasWork() + { + return m_numPrims != 0; + } + + simdvector& GetSimdVector(uint32_t index, uint32_t slot) + { + SWR_ASSERT(0, "%s NOT IMPLEMENTED", __FUNCTION__); + static simdvector junk = { 0 }; + return junk; + } + + static simdscalari GenPrimMask(uint32_t numPrims) + { + SWR_ASSERT(numPrims <= KNOB_SIMD_WIDTH); +#if KNOB_SIMD_WIDTH == 8 + static const OSALIGN(int32_t, 64) maskGen[KNOB_SIMD_WIDTH * 2] = + { + -1, -1, -1, -1, -1, -1, -1, -1, + 0, 0, 0, 0, 0, 0, 0, 0 + }; +#elif KNOB_SIMD_WIDTH == 16 + static const OSALIGN(int32_t, 128) maskGen[KNOB_SIMD_WIDTH * 2] = + { + -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, -1, + 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0 + }; +#else +#error "Help, help, I can't get up!" +#endif + + return _simd_loadu_si((const simdscalari*)&maskGen[KNOB_SIMD_WIDTH - numPrims]); + } + + bool Assemble(uint32_t slot, simdvector verts[]) + { + static_assert(KNOB_SIMD_WIDTH == 8, "Need to revisit this when AVX512 is implemented"); + SWR_ASSERT(slot < m_numAttributes); + + uint32_t numPrimsToAssemble = PA_TESS::NumPrims(); + if (0 == numPrimsToAssemble) + { + return false; + } + + simdscalari mask = GenPrimMask(numPrimsToAssemble); + + const float* pBaseAttrib = (const float*)&m_pVertexData[slot * m_attributeStrideInVectors * 4]; + for (uint32_t i = 0; i < m_numVertsPerPrim; ++i) + { + simdscalari indices = _simd_load_si((const simdscalari*)m_ppIndices[i]); + + const float* pBase = pBaseAttrib; + for (uint32_t c = 0; c < 4; ++c) + { + verts[i].v[c] = _simd_mask_i32gather_ps( + _simd_setzero_ps(), + pBase, + indices, + _simd_castsi_ps(mask), + 4 /* gcc doesn't like sizeof(float) */); + pBase += m_attributeStrideInVectors * KNOB_SIMD_WIDTH; + } + } + + return true; + } + + void AssembleSingle(uint32_t slot, uint32_t primIndex, __m128 verts[]) + { + SWR_ASSERT(slot < m_numAttributes); + SWR_ASSERT(primIndex < PA_TESS::NumPrims()); + + const float* pVertDataBase = (const float*)&m_pVertexData[slot * m_attributeStrideInVectors * 4]; + for (uint32_t i = 0; i < m_numVertsPerPrim; ++i) + { + uint32_t index = m_ppIndices[i][primIndex]; + const float* pVertData = pVertDataBase; + float* pVert = (float*)&verts[i]; + + for (uint32_t c = 0; c < 4; ++c) + { + pVert[c] = pVertData[index]; + pVertData += m_attributeStrideInVectors * KNOB_SIMD_WIDTH; + } + } + } + + bool NextPrim() + { + uint32_t numPrims = PA_TESS::NumPrims(); + m_numPrims -= numPrims; + m_ppIndices[0] += numPrims; + m_ppIndices[1] += numPrims; + m_ppIndices[2] += numPrims; + + return HasWork(); + } + + simdvertex& GetNextVsOutput() + { + SWR_ASSERT(0, "%s", __FUNCTION__); + static simdvertex junk; + return junk; + } + + bool GetNextStreamOutput() + { + SWR_ASSERT(0, "%s", __FUNCTION__); + return false; + } + + simdmask& GetNextVsIndices() + { + SWR_ASSERT(0, "%s", __FUNCTION__); + static simdmask junk; + return junk; + } + + uint32_t NumPrims() + { + return std::min(m_numPrims, KNOB_SIMD_WIDTH); + } + + void Reset() { SWR_ASSERT(0); }; + + simdscalari GetPrimID(uint32_t startID) + { + return _simd_add_epi32(_simd_set1_epi32(startID), m_vPrimId); + } + +private: + const simdscalar* m_pVertexData = nullptr; + uint32_t m_attributeStrideInVectors = 0; + uint32_t m_numAttributes = 0; + uint32_t m_numPrims = 0; + uint32_t* m_ppIndices[3]; + + uint32_t m_numVertsPerPrim = 0; + + simdscalari m_vPrimId; +}; + +// Primitive Assembler factory class, responsible for creating and initializing the correct assembler +// based on state. +template +struct PA_FACTORY +{ + PA_FACTORY(DRAW_CONTEXT* pDC, PRIMITIVE_TOPOLOGY in_topo, uint32_t numVerts) : topo(in_topo) + { +#if KNOB_ENABLE_CUT_AWARE_PA == TRUE + const API_STATE& state = GetApiState(pDC); + if ((IsIndexedT && ( + topo == TOP_TRIANGLE_STRIP || topo == TOP_POINT_LIST || + topo == TOP_LINE_LIST || topo == TOP_LINE_STRIP || + topo == TOP_TRIANGLE_LIST || topo == TOP_LINE_LIST_ADJ || + topo == TOP_LISTSTRIP_ADJ || topo == TOP_TRI_LIST_ADJ || + topo == TOP_TRI_STRIP_ADJ)) || + + // non-indexed draws with adjacency topologies must use cut-aware PA until we add support + // for them in the optimized PA + (!IsIndexedT && ( + topo == TOP_LINE_LIST_ADJ || topo == TOP_LISTSTRIP_ADJ || topo == TOP_TRI_LIST_ADJ || topo == TOP_TRI_STRIP_ADJ))) + { + memset(&indexStore, 0, sizeof(indexStore)); + DWORD numAttribs; + _BitScanReverse(&numAttribs, state.feAttribMask); + numAttribs++; + new (&this->paCut) PA_STATE_CUT(pDC, (uint8_t*)&this->vertexStore[0], MAX_NUM_VERTS_PER_PRIM * KNOB_SIMD_WIDTH, + &this->indexStore[0], numVerts, numAttribs, state.topology, false); + cutPA = true; + } + else +#endif + { + uint32_t numPrims = GetNumPrims(in_topo, numVerts); + new (&this->paOpt) PA_STATE_OPT(pDC, numPrims, (uint8_t*)&this->vertexStore[0], MAX_NUM_VERTS_PER_PRIM * KNOB_SIMD_WIDTH, false); + cutPA = false; + } + + } + + PA_STATE& GetPA() + { +#if KNOB_ENABLE_CUT_AWARE_PA == TRUE + if (cutPA) + { + return this->paCut; + } + else +#endif + { + return this->paOpt; + } + } + + PA_STATE_OPT paOpt; + PA_STATE_CUT paCut; + bool cutPA; + + PRIMITIVE_TOPOLOGY topo; + + simdvertex vertexStore[MAX_NUM_VERTS_PER_PRIM]; + simdmask indexStore[MAX_NUM_VERTS_PER_PRIM]; +}; diff --git a/src/gallium/drivers/swr/rasterizer/core/pa_avx.cpp b/src/gallium/drivers/swr/rasterizer/core/pa_avx.cpp new file mode 100644 index 00000000000..9850b436e39 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/pa_avx.cpp @@ -0,0 +1,1177 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file pa_avx.cpp +* +* @brief AVX implementation for primitive assembly. +* N primitives are assembled at a time, where N is the SIMD width. +* A state machine, that is specific for a given topology, drives the +* assembly of vertices into triangles. +* +******************************************************************************/ +#include "context.h" +#include "pa.h" +#include "frontend.h" + +#if (KNOB_SIMD_WIDTH == 8) + +bool PaTriList0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +bool PaTriList1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +bool PaTriList2(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +void PaTriListSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]); + +bool PaTriStrip0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +bool PaTriStrip1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +void PaTriStripSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]); + +bool PaTriFan0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +bool PaTriFan1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +void PaTriFanSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]); + +bool PaQuadList0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +bool PaQuadList1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +void PaQuadListSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]); + +bool PaLineLoop0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +bool PaLineLoop1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); + +bool PaLineList0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +bool PaLineList1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +void PaLineListSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t index, __m128 verts[]); + +bool PaLineStrip0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +bool PaLineStrip1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +void PaLineStripSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 lineverts[]); + +bool PaPoints0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +void PaPointsSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]); + +bool PaRectList0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +bool PaRectList1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +bool PaRectList2(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]); +void PaRectListSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]); + +template +void PaPatchListSingle(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]) +{ + // We have an input of KNOB_SIMD_WIDTH * TotalControlPoints and we output + // KNOB_SIMD_WIDTH * 1 patch. This function is called once per attribute. + // Each attribute has 4 components. + + /// @todo Optimize this + + float* pOutVec = (float*)verts; + + for (uint32_t cp = 0; cp < TotalControlPoints; ++cp) + { + uint32_t input_cp = primIndex * TotalControlPoints + cp; + uint32_t input_vec = input_cp / KNOB_SIMD_WIDTH; + uint32_t input_lane = input_cp % KNOB_SIMD_WIDTH; + + // Loop over all components of the attribute + for (uint32_t i = 0; i < 4; ++i) + { + const float* pInputVec = (const float*)(&PaGetSimdVector(pa, input_vec, slot)[i]); + pOutVec[cp * 4 + i] = pInputVec[input_lane]; + } + } +} + +template +static bool PaPatchList(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + SetNextPaState( + pa, + PaPatchList, + PaPatchListSingle); + + return false; +} + +template +static bool PaPatchListTerm(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + // We have an input of KNOB_SIMD_WIDTH * TotalControlPoints and we output + // KNOB_SIMD_WIDTH * 1 patch. This function is called once per attribute. + // Each attribute has 4 components. + + /// @todo Optimize this + + // Loop over all components of the attribute + for (uint32_t i = 0; i < 4; ++i) + { + for (uint32_t cp = 0; cp < TotalControlPoints; ++cp) + { + float vec[KNOB_SIMD_WIDTH]; + for (uint32_t lane = 0; lane < KNOB_SIMD_WIDTH; ++lane) + { + uint32_t input_cp = lane * TotalControlPoints + cp; + uint32_t input_vec = input_cp / KNOB_SIMD_WIDTH; + uint32_t input_lane = input_cp % KNOB_SIMD_WIDTH; + + const float* pInputVec = (const float*)(&PaGetSimdVector(pa, input_vec, slot)[i]); + vec[lane] = pInputVec[input_lane]; + } + verts[cp][i] = _simd_loadu_ps(vec); + } + } + + SetNextPaState( + pa, + PaPatchList, + PaPatchListSingle, + 0, + KNOB_SIMD_WIDTH, + true); + + return true; +} + +#define PA_PATCH_LIST_TERMINATOR(N) \ + template<> bool PaPatchList(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[])\ + { return PaPatchListTerm(pa, slot, verts); } +PA_PATCH_LIST_TERMINATOR(1) +PA_PATCH_LIST_TERMINATOR(2) +PA_PATCH_LIST_TERMINATOR(3) +PA_PATCH_LIST_TERMINATOR(4) +PA_PATCH_LIST_TERMINATOR(5) +PA_PATCH_LIST_TERMINATOR(6) +PA_PATCH_LIST_TERMINATOR(7) +PA_PATCH_LIST_TERMINATOR(8) +PA_PATCH_LIST_TERMINATOR(9) +PA_PATCH_LIST_TERMINATOR(10) +PA_PATCH_LIST_TERMINATOR(11) +PA_PATCH_LIST_TERMINATOR(12) +PA_PATCH_LIST_TERMINATOR(13) +PA_PATCH_LIST_TERMINATOR(14) +PA_PATCH_LIST_TERMINATOR(15) +PA_PATCH_LIST_TERMINATOR(16) +PA_PATCH_LIST_TERMINATOR(17) +PA_PATCH_LIST_TERMINATOR(18) +PA_PATCH_LIST_TERMINATOR(19) +PA_PATCH_LIST_TERMINATOR(20) +PA_PATCH_LIST_TERMINATOR(21) +PA_PATCH_LIST_TERMINATOR(22) +PA_PATCH_LIST_TERMINATOR(23) +PA_PATCH_LIST_TERMINATOR(24) +PA_PATCH_LIST_TERMINATOR(25) +PA_PATCH_LIST_TERMINATOR(26) +PA_PATCH_LIST_TERMINATOR(27) +PA_PATCH_LIST_TERMINATOR(28) +PA_PATCH_LIST_TERMINATOR(29) +PA_PATCH_LIST_TERMINATOR(30) +PA_PATCH_LIST_TERMINATOR(31) +PA_PATCH_LIST_TERMINATOR(32) +#undef PA_PATCH_LIST_TERMINATOR + +bool PaTriList0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + SetNextPaState(pa, PaTriList1, PaTriListSingle0); + return false; // Not enough vertices to assemble 4 or 8 triangles. +} + +bool PaTriList1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + SetNextPaState(pa, PaTriList2, PaTriListSingle0); + return false; // Not enough vertices to assemble 8 triangles. +} + +bool PaTriList2(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + simdvector& a = PaGetSimdVector(pa, 0, slot); + simdvector& b = PaGetSimdVector(pa, 1, slot); + simdvector& c = PaGetSimdVector(pa, 2, slot); + simdscalar s; + + // Tri Pattern - provoking vertex is always v0 + // v0 -> 0 3 6 9 12 15 18 21 + // v1 -> 1 4 7 10 13 16 19 22 + // v2 -> 2 5 8 11 14 17 20 23 + + for(int i = 0; i < 4; ++i) + { + simdvector& v0 = verts[0]; + v0[i] = _simd_blend_ps(a[i], b[i], 0x92); + v0[i] = _simd_blend_ps(v0[i], c[i], 0x24); + v0[i] = _mm256_permute_ps(v0[i], 0x6C); + s = _mm256_permute2f128_ps(v0[i], v0[i], 0x21); + v0[i] = _simd_blend_ps(v0[i], s, 0x44); + + simdvector& v1 = verts[1]; + v1[i] = _simd_blend_ps(a[i], b[i], 0x24); + v1[i] = _simd_blend_ps(v1[i], c[i], 0x49); + v1[i] = _mm256_permute_ps(v1[i], 0xB1); + s = _mm256_permute2f128_ps(v1[i], v1[i], 0x21); + v1[i] = _simd_blend_ps(v1[i], s, 0x66); + + simdvector& v2 = verts[2]; + v2[i] = _simd_blend_ps(a[i], b[i], 0x49); + v2[i] = _simd_blend_ps(v2[i], c[i], 0x92); + v2[i] = _mm256_permute_ps(v2[i], 0xC6); + s = _mm256_permute2f128_ps(v2[i], v2[i], 0x21); + v2[i] = _simd_blend_ps(v2[i], s, 0x22); + } + + SetNextPaState(pa, PaTriList0, PaTriListSingle0, 0, KNOB_SIMD_WIDTH, true); + return true; +} + +void PaTriListSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]) +{ + // We have 12 simdscalars contained within 3 simdvectors which + // hold at least 8 triangles worth of data. We want to assemble a single + // triangle with data in horizontal form. + simdvector& a = PaGetSimdVector(pa, 0, slot); + simdvector& b = PaGetSimdVector(pa, 1, slot); + simdvector& c = PaGetSimdVector(pa, 2, slot); + + // Convert from vertical to horizontal. + // Tri Pattern - provoking vertex is always v0 + // v0 -> 0 3 6 9 12 15 18 21 + // v1 -> 1 4 7 10 13 16 19 22 + // v2 -> 2 5 8 11 14 17 20 23 + switch(primIndex) + { + case 0: + verts[0] = swizzleLane0(a); + verts[1] = swizzleLane1(a); + verts[2] = swizzleLane2(a); + break; + case 1: + verts[0] = swizzleLane3(a); + verts[1] = swizzleLane4(a); + verts[2] = swizzleLane5(a); + break; + case 2: + verts[0] = swizzleLane6(a); + verts[1] = swizzleLane7(a); + verts[2] = swizzleLane0(b); + break; + case 3: + verts[0] = swizzleLane1(b); + verts[1] = swizzleLane2(b); + verts[2] = swizzleLane3(b); + break; + case 4: + verts[0] = swizzleLane4(b); + verts[1] = swizzleLane5(b); + verts[2] = swizzleLane6(b); + break; + case 5: + verts[0] = swizzleLane7(b); + verts[1] = swizzleLane0(c); + verts[2] = swizzleLane1(c); + break; + case 6: + verts[0] = swizzleLane2(c); + verts[1] = swizzleLane3(c); + verts[2] = swizzleLane4(c); + break; + case 7: + verts[0] = swizzleLane5(c); + verts[1] = swizzleLane6(c); + verts[2] = swizzleLane7(c); + break; + }; +} + +bool PaTriStrip0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + SetNextPaState(pa, PaTriStrip1, PaTriStripSingle0); + return false; // Not enough vertices to assemble 8 triangles. +} + +bool PaTriStrip1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + simdvector& a = PaGetSimdVector(pa, pa.prev, slot); + simdvector& b = PaGetSimdVector(pa, pa.cur, slot); + simdscalar s; + + for(int i = 0; i < 4; ++i) + { + simdscalar a0 = a[i]; + simdscalar b0 = b[i]; + + // Tri Pattern - provoking vertex is always v0 + // v0 -> 01234567 + // v1 -> 13355779 + // v2 -> 22446688 + simdvector& v0 = verts[0]; + v0[i] = a0; + + // s -> 4567891011 + s = _mm256_permute2f128_ps(a0, b0, 0x21); + // s -> 23456789 + s = _simd_shuffle_ps(a0, s, _MM_SHUFFLE(1, 0, 3, 2)); + + simdvector& v1 = verts[1]; + // v1 -> 13355779 + v1[i] = _simd_shuffle_ps(a0, s, _MM_SHUFFLE(3, 1, 3, 1)); + + simdvector& v2 = verts[2]; + // v2 -> 22446688 + v2[i] = _simd_shuffle_ps(a0, s, _MM_SHUFFLE(2, 2, 2, 2)); + } + + SetNextPaState(pa, PaTriStrip1, PaTriStripSingle0, 0, KNOB_SIMD_WIDTH); + return true; +} + +void PaTriStripSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]) +{ + simdvector& a = PaGetSimdVector(pa, pa.prev, slot); + simdvector& b = PaGetSimdVector(pa, pa.cur, slot); + + // Convert from vertical to horizontal. + // Tri Pattern - provoking vertex is always v0 + // v0 -> 01234567 + // v1 -> 13355779 + // v2 -> 22446688 + switch(primIndex) + { + case 0: + verts[0] = swizzleLane0(a); + verts[1] = swizzleLane1(a); + verts[2] = swizzleLane2(a); + break; + case 1: + verts[0] = swizzleLane1(a); + verts[1] = swizzleLane3(a); + verts[2] = swizzleLane2(a); + break; + case 2: + verts[0] = swizzleLane2(a); + verts[1] = swizzleLane3(a); + verts[2] = swizzleLane4(a); + break; + case 3: + verts[0] = swizzleLane3(a); + verts[1] = swizzleLane5(a); + verts[2] = swizzleLane4(a); + break; + case 4: + verts[0] = swizzleLane4(a); + verts[1] = swizzleLane5(a); + verts[2] = swizzleLane6(a); + break; + case 5: + verts[0] = swizzleLane5(a); + verts[1] = swizzleLane7(a); + verts[2] = swizzleLane6(a); + break; + case 6: + verts[0] = swizzleLane6(a); + verts[1] = swizzleLane7(a); + verts[2] = swizzleLane0(b); + break; + case 7: + verts[0] = swizzleLane7(a); + verts[1] = swizzleLane1(b); + verts[2] = swizzleLane0(b); + break; + }; +} + +bool PaTriFan0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + simdvector& a = PaGetSimdVector(pa, pa.cur, slot); + + // Extract vertex 0 to every lane of first vector + for(int i = 0; i < 4; ++i) + { + __m256 a0 = a[i]; + simdvector& v0 = verts[0]; + v0[i] = _simd_shuffle_ps(a0, a0, _MM_SHUFFLE(0, 0, 0, 0)); + v0[i] = _mm256_permute2f128_ps(v0[i], a0, 0x00); + } + + // store off leading vertex for attributes + simdvertex* pVertex = (simdvertex*)pa.pStreamBase; + pa.leadingVertex = pVertex[pa.cur]; + + SetNextPaState(pa, PaTriFan1, PaTriFanSingle0); + return false; // Not enough vertices to assemble 8 triangles. +} + +bool PaTriFan1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + simdvector& leadVert = pa.leadingVertex.attrib[slot]; + simdvector& a = PaGetSimdVector(pa, pa.prev, slot); + simdvector& b = PaGetSimdVector(pa, pa.cur, slot); + simdscalar s; + + // need to fill vectors 1/2 with new verts, and v0 with anchor vert. + for(int i = 0; i < 4; ++i) + { + simdscalar a0 = a[i]; + simdscalar b0 = b[i]; + + __m256 comp = leadVert[i]; + simdvector& v0 = verts[0]; + v0[i] = _simd_shuffle_ps(comp, comp, _MM_SHUFFLE(0, 0, 0, 0)); + v0[i] = _mm256_permute2f128_ps(v0[i], comp, 0x00); + + simdvector& v2 = verts[2]; + s = _mm256_permute2f128_ps(a0, b0, 0x21); + v2[i] = _simd_shuffle_ps(a0, s, _MM_SHUFFLE(1, 0, 3, 2)); + + simdvector& v1 = verts[1]; + v1[i] = _simd_shuffle_ps(a0, v2[i], _MM_SHUFFLE(2, 1, 2, 1)); + } + + SetNextPaState(pa, PaTriFan1, PaTriFanSingle0, 0, KNOB_SIMD_WIDTH); + return true; +} + +void PaTriFanSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]) +{ + // vert 0 from leading vertex + simdvector& lead = pa.leadingVertex.attrib[slot]; + verts[0] = swizzleLane0(lead); + + simdvector& a = PaGetSimdVector(pa, pa.prev, slot); + simdvector& b = PaGetSimdVector(pa, pa.cur, slot); + + // vert 1 + if (primIndex < 7) + { + verts[1] = swizzleLaneN(a, primIndex + 1); + } + else + { + verts[1] = swizzleLane0(b); + } + + // vert 2 + if (primIndex < 6) + { + verts[2] = swizzleLaneN(a, primIndex + 2); + } + else + { + verts[2] = swizzleLaneN(b, primIndex - 6); + } +} + +bool PaQuadList0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + SetNextPaState(pa, PaQuadList1, PaQuadListSingle0); + return false; // Not enough vertices to assemble 8 triangles. +} + +bool PaQuadList1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + simdvector& a = PaGetSimdVector(pa, 0, slot); + simdvector& b = PaGetSimdVector(pa, 1, slot); + simdscalar s1, s2; + + for(int i = 0; i < 4; ++i) + { + simdscalar a0 = a[i]; + simdscalar b0 = b[i]; + + s1 = _mm256_permute2f128_ps(a0, b0, 0x20); + s2 = _mm256_permute2f128_ps(a0, b0, 0x31); + + simdvector& v0 = verts[0]; + v0[i] = _simd_shuffle_ps(s1, s2, _MM_SHUFFLE(0, 0, 0, 0)); + + simdvector& v1 = verts[1]; + v1[i] = _simd_shuffle_ps(s1, s2, _MM_SHUFFLE(2, 1, 2, 1)); + + simdvector& v2 = verts[2]; + v2[i] = _simd_shuffle_ps(s1, s2, _MM_SHUFFLE(3, 2, 3, 2)); + } + + SetNextPaState(pa, PaQuadList0, PaQuadListSingle0, 0, KNOB_SIMD_WIDTH, true); + return true; +} + +void PaQuadListSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]) +{ + simdvector& a = PaGetSimdVector(pa, 0, slot); + simdvector& b = PaGetSimdVector(pa, 1, slot); + + switch (primIndex) + { + case 0: + // triangle 0 - 0 1 2 + verts[0] = swizzleLane0(a); + verts[1] = swizzleLane1(a); + verts[2] = swizzleLane2(a); + break; + + case 1: + // triangle 1 - 0 2 3 + verts[0] = swizzleLane0(a); + verts[1] = swizzleLane2(a); + verts[2] = swizzleLane3(a); + break; + + case 2: + // triangle 2 - 4 5 6 + verts[0] = swizzleLane4(a); + verts[1] = swizzleLane5(a); + verts[2] = swizzleLane6(a); + break; + + case 3: + // triangle 3 - 4 6 7 + verts[0] = swizzleLane4(a); + verts[1] = swizzleLane6(a); + verts[2] = swizzleLane7(a); + break; + + case 4: + // triangle 4 - 8 9 10 (0 1 2) + verts[0] = swizzleLane0(b); + verts[1] = swizzleLane1(b); + verts[2] = swizzleLane2(b); + break; + + case 5: + // triangle 1 - 0 2 3 + verts[0] = swizzleLane0(b); + verts[1] = swizzleLane2(b); + verts[2] = swizzleLane3(b); + break; + + case 6: + // triangle 2 - 4 5 6 + verts[0] = swizzleLane4(b); + verts[1] = swizzleLane5(b); + verts[2] = swizzleLane6(b); + break; + + case 7: + // triangle 3 - 4 6 7 + verts[0] = swizzleLane4(b); + verts[1] = swizzleLane6(b); + verts[2] = swizzleLane7(b); + break; + } +} + +void PaLineLoopSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t lineIndex, __m128 verts[]) +{ + PaLineStripSingle0(pa, slot, lineIndex, verts); + + if (pa.numPrimsComplete + lineIndex == pa.numPrims - 1) { + simdvector &start = PaGetSimdVector(pa, pa.first, slot); + verts[1] = swizzleLane0(start); + } +} + +bool PaLineLoop0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + SetNextPaState(pa, PaLineLoop1, PaLineLoopSingle0); + return false; +} + +bool PaLineLoop1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + PaLineStrip1(pa, slot, verts); + + if (pa.numPrimsComplete + KNOB_SIMD_WIDTH > pa.numPrims - 1) { + // loop reconnect now + int lane = pa.numPrims - pa.numPrimsComplete - 1; + simdvector &start = PaGetSimdVector(pa, pa.first, slot); + for (int i = 0; i < 4; i++) { + float *startVtx = (float *)&(start[i]); + float *targetVtx = (float *)&(verts[1][i]); + targetVtx[lane] = startVtx[0]; + } + } + + SetNextPaState(pa, PaLineLoop1, PaLineLoopSingle0, 0, KNOB_SIMD_WIDTH); + return true; +} + + +bool PaLineList0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + SetNextPaState(pa, PaLineList1, PaLineListSingle0); + return false; // Not enough vertices to assemble 8 lines +} + +bool PaLineList1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + simdvector& a = PaGetSimdVector(pa, 0, slot); + simdvector& b = PaGetSimdVector(pa, 1, slot); + /// @todo: verify provoking vertex is correct + // Line list 0 1 2 3 4 5 6 7 + // 8 9 10 11 12 13 14 15 + + // shuffle: + // 0 2 4 6 8 10 12 14 + // 1 3 5 7 9 11 13 15 + + for (uint32_t i = 0; i < 4; ++i) + { + // 0 1 2 3 8 9 10 11 + __m256 vALowBLow = _mm256_permute2f128_ps(a.v[i], b.v[i], 0x20); + // 4 5 6 7 12 13 14 15 + __m256 vAHighBHigh = _mm256_permute2f128_ps(a.v[i], b.v[i], 0x31); + + // 0 2 4 6 8 10 12 14 + verts[0].v[i] = _mm256_shuffle_ps(vALowBLow, vAHighBHigh, _MM_SHUFFLE(2, 0, 2, 0)); + // 1 3 5 7 9 11 13 15 + verts[1].v[i] = _mm256_shuffle_ps(vALowBLow, vAHighBHigh, _MM_SHUFFLE(3, 1, 3, 1)); + } + + SetNextPaState(pa, PaLineList0, PaLineListSingle0, 0, KNOB_SIMD_WIDTH, true); + return true; +} + +void PaLineListSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]) +{ + simdvector &a = PaGetSimdVector(pa, pa.prev, slot); + simdvector &b = PaGetSimdVector(pa, pa.cur, slot); + + switch (primIndex) + { + case 0: + verts[0] = swizzleLane0(a); + verts[1] = swizzleLane1(a); + break; + case 1: + verts[0] = swizzleLane2(a); + verts[1] = swizzleLane3(a); + break; + case 2: + verts[0] = swizzleLane4(a); + verts[1] = swizzleLane5(a); + break; + case 3: + verts[0] = swizzleLane6(a); + verts[1] = swizzleLane7(a); + break; + case 4: + verts[0] = swizzleLane0(b); + verts[1] = swizzleLane1(b); + break; + case 5: + verts[0] = swizzleLane2(b); + verts[1] = swizzleLane3(b); + break; + case 6: + verts[0] = swizzleLane4(b); + verts[1] = swizzleLane5(b); + break; + case 7: + verts[0] = swizzleLane6(b); + verts[1] = swizzleLane7(b); + break; + } +} + +bool PaLineStrip0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + SetNextPaState(pa, PaLineStrip1, PaLineStripSingle0); + return false; // Not enough vertices to assemble 8 lines +} + +bool PaLineStrip1(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + simdvector& a = PaGetSimdVector(pa, pa.prev, slot); + simdvector& b = PaGetSimdVector(pa, pa.cur, slot); + + /// @todo: verify provoking vertex is correct + // Line list 0 1 2 3 4 5 6 7 + // 8 9 10 11 12 13 14 15 + + // shuffle: + // 0 1 2 3 4 5 6 7 + // 1 2 3 4 5 6 7 8 + + verts[0] = a; + + for(uint32_t i = 0; i < 4; ++i) + { + // 1 2 3 x 5 6 7 x + __m256 vPermA = _mm256_permute_ps(a.v[i], 0x39); // indices hi->low 00 11 10 01 (0 3 2 1) + // 4 5 6 7 8 9 10 11 + __m256 vAHighBLow = _mm256_permute2f128_ps(a.v[i], b.v[i], 0x21); + + // x x x 4 x x x 8 + __m256 vPermB = _mm256_permute_ps(vAHighBLow, 0); // indices hi->low (0 0 0 0) + + verts[1].v[i] = _mm256_blend_ps(vPermA, vPermB, 0x88); + } + + SetNextPaState(pa, PaLineStrip1, PaLineStripSingle0, 0, KNOB_SIMD_WIDTH); + return true; +} + +void PaLineStripSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t lineIndex, __m128 verts[]) +{ + simdvector& a = PaGetSimdVector(pa, pa.prev, slot); + simdvector& b = PaGetSimdVector(pa, pa.cur, slot); + + switch (lineIndex) + { + case 0: + verts[0] = swizzleLane0(a); + verts[1] = swizzleLane1(a); + break; + case 1: + verts[0] = swizzleLane1(a); + verts[1] = swizzleLane2(a); + break; + case 2: + verts[0] = swizzleLane2(a); + verts[1] = swizzleLane3(a); + break; + case 3: + verts[0] = swizzleLane3(a); + verts[1] = swizzleLane4(a); + break; + case 4: + verts[0] = swizzleLane4(a); + verts[1] = swizzleLane5(a); + break; + case 5: + verts[0] = swizzleLane5(a); + verts[1] = swizzleLane6(a); + break; + case 6: + verts[0] = swizzleLane6(a); + verts[1] = swizzleLane7(a); + break; + case 7: + verts[0] = swizzleLane7(a); + verts[1] = swizzleLane0(b); + break; + } +} + +bool PaPoints0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + simdvector& a = PaGetSimdVector(pa, pa.cur, slot); + + verts[0] = a; // points only have 1 vertex. + + SetNextPaState(pa, PaPoints0, PaPointsSingle0, 0, KNOB_SIMD_WIDTH, true); + return true; +} + +void PaPointsSingle0(PA_STATE_OPT& pa, uint32_t slot, uint32_t primIndex, __m128 verts[]) +{ + simdvector &a = PaGetSimdVector(pa, pa.cur, slot); + switch(primIndex) + { + case 0: + verts[0] = swizzleLane0(a); + break; + case 1: + verts[0] = swizzleLane1(a); + break; + case 2: + verts[0] = swizzleLane2(a); + break; + case 3: + verts[0] = swizzleLane3(a); + break; + case 4: + verts[0] = swizzleLane4(a); + break; + case 5: + verts[0] = swizzleLane5(a); + break; + case 6: + verts[0] = swizzleLane6(a); + break; + case 7: + verts[0] = swizzleLane7(a); + break; + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief State 1 for RECT_LIST topology. +/// There is not enough to assemble 8 triangles. +bool PaRectList0(PA_STATE_OPT& pa, uint32_t slot, simdvector verts[]) +{ + SetNextPaState(pa, PaRectList1, PaRectListSingle0); + return false; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief State 1 for RECT_LIST topology. +/// Rect lists has the following format. +/// w x y z +/// v2 o---o v5 o---o v8 o---o v11 o---o +/// | \ | | \ | | \ | | \ | +/// v1 o---o v4 o---o v7 o---o v10 o---o +/// v0 v3 v6 v9 +/// +/// Only 3 vertices of the rectangle are supplied. The 4th vertex is implied. +/// +/// tri0 = { v0, v1, v2 } tri1 = { v0, v2, w } <-- w = v0 - v1 + v2 +/// tri2 = { v3, v4, v5 } tri3 = { v3, v5, x } <-- x = v3 - v4 + v5 +/// etc. +/// +/// PA outputs 3 simdvectors for each of the triangle vertices v0, v1, v2 +/// where v0 contains all the first vertices for 8 triangles. +/// +/// Result: +/// verts[0] = { v0, v0, v3, v3, v6, v6, v9, v9 } +/// verts[1] = { v1, v2, v4, v5, v7, v8, v10, v11 } +/// verts[2] = { v2, w, v5, x, v8, y, v11, z } +/// +/// @param pa - State for PA state machine. +/// @param slot - Index into VS output which is either a position (slot 0) or attribute. +/// @param verts - triangle output for binner. SOA - Array of v0 for 8 triangles, followed by v1, etc. +bool PaRectList1( + PA_STATE_OPT& pa, + uint32_t slot, + simdvector verts[]) +{ + // SIMD vectors a and b are the last two vertical outputs from the vertex shader. + simdvector& a = PaGetSimdVector(pa, 0, slot); // a[] = { v0, v1, v2, v3, v4, v5, v6, v7 } + simdvector& b = PaGetSimdVector(pa, 1, slot); // b[] = { v8, v9, v10, v11, v12, v13, v14, v15 } + + __m256 tmp0, tmp1, tmp2; + + // Loop over each component in the simdvector. + for(int i = 0; i < 4; ++i) + { + simdvector& v0 = verts[0]; // verts[0] needs to be { v0, v0, v3, v3, v6, v6, v9, v9 } + tmp0 = _mm256_permute2f128_ps(b[i], b[i], 0x01); // tmp0 = { v12, v13, v14, v15, v8, v9, v10, v11 } + v0[i] = _mm256_blend_ps(a[i], tmp0, 0x20); // v0 = { v0, *, *, v3, *, v9, v6, * } where * is don't care. + tmp1 = _mm256_permute_ps(v0[i], 0xF0); // tmp1 = { v0, v0, v3, v3, *, *, *, * } + v0[i] = _mm256_permute_ps(v0[i], 0x5A); // v0 = { *, *, *, *, v6, v6, v9, v9 } + v0[i] = _mm256_blend_ps(tmp1, v0[i], 0xF0); // v0 = { v0, v0, v3, v3, v6, v6, v9, v9 } + + /// NOTE This is a bit expensive due to conflicts between vertices in 'a' and 'b'. + /// AVX2 should make this much cheaper. + simdvector& v1 = verts[1]; // verts[1] needs to be { v1, v2, v4, v5, v7, v8, v10, v11 } + v1[i] = _mm256_permute_ps(a[i], 0x09); // v1 = { v1, v2, *, *, *, *, *, * } + tmp1 = _mm256_permute_ps(a[i], 0x43); // tmp1 = { *, *, *, *, v7, *, v4, v5 } + tmp2 = _mm256_blend_ps(v1[i], tmp1, 0xF0); // tmp2 = { v1, v2, *, *, v7, *, v4, v5 } + tmp1 = _mm256_permute2f128_ps(tmp2, tmp2, 0x1); // tmp1 = { v7, *, v4, v5, * *, *, * } + v1[i] = _mm256_permute_ps(tmp0, 0xE0); // v1 = { *, *, *, *, *, v8, v10, v11 } + v1[i] = _mm256_blend_ps(tmp2, v1[i], 0xE0); // v1 = { v1, v2, *, *, v7, v8, v10, v11 } + v1[i] = _mm256_blend_ps(v1[i], tmp1, 0x0C); // v1 = { v1, v2, v4, v5, v7, v8, v10, v11 } + + // verts[2] = { v2, w, v5, x, v8, y, v11, z } + simdvector& v2 = verts[2]; // verts[2] needs to be { v2, w, v5, x, v8, y, v11, z } + v2[i] = _mm256_permute_ps(tmp0, 0x30); // v2 = { *, *, *, *, v8, *, v11, * } + tmp1 = _mm256_permute_ps(tmp2, 0x31); // tmp1 = { v2, *, v5, *, *, *, *, * } + v2[i] = _mm256_blend_ps(tmp1, v2[i], 0xF0); + + // Need to compute 4th implied vertex for the rectangle. + tmp2 = _mm256_sub_ps(v0[i], v1[i]); + tmp2 = _mm256_add_ps(tmp2, v2[i]); // tmp2 = { w, *, x, *, y, *, z, * } + tmp2 = _mm256_permute_ps(tmp2, 0xA0); // tmp2 = { *, w, *, x, *, y, *, z } + v2[i] = _mm256_blend_ps(v2[i], tmp2, 0xAA); // v2 = { v2, w, v5, x, v8, y, v11, z } + } + + SetNextPaState(pa, PaRectList1, PaRectListSingle0, 0, KNOB_SIMD_WIDTH, true); + return true; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief State 2 for RECT_LIST topology. +/// Not implemented unless there is a use case for more then 8 rects. +/// @param pa - State for PA state machine. +/// @param slot - Index into VS output which is either a position (slot 0) or attribute. +/// @param verts - triangle output for binner. SOA - Array of v0 for 8 triangles, followed by v1, etc. +bool PaRectList2( + PA_STATE_OPT& pa, + uint32_t slot, + simdvector verts[]) +{ + SWR_ASSERT(0); // Is rect list used for anything other then clears? + SetNextPaState(pa, PaRectList0, PaRectListSingle0, 0, KNOB_SIMD_WIDTH, true); + return true; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief This procedure is called by the Binner to assemble the attributes. +/// Unlike position, which is stored vertically, the attributes are +/// stored horizontally. The outputs from the VS, labeled as 'a' and +/// 'b' are vertical. This function needs to transpose the lanes +/// containing the vertical attribute data into horizontal form. +/// @param pa - State for PA state machine. +/// @param slot - Index into VS output for a given attribute. +/// @param primIndex - Binner processes each triangle individually. +/// @param verts - triangle output for binner. SOA - Array of v0 for 8 triangles, followed by v1, etc. +void PaRectListSingle0( + PA_STATE_OPT& pa, + uint32_t slot, + uint32_t primIndex, + __m128 verts[]) +{ + // We have 12 simdscalars contained within 3 simdvectors which + // hold at least 8 triangles worth of data. We want to assemble a single + // triangle with data in horizontal form. + simdvector& a = PaGetSimdVector(pa, 0, slot); + + // Convert from vertical to horizontal. + switch(primIndex) + { + case 0: + verts[0] = swizzleLane0(a); + verts[1] = swizzleLane1(a); + verts[2] = swizzleLane2(a); + break; + case 1: + verts[0] = swizzleLane0(a); + verts[1] = swizzleLane2(a); + verts[2] = _mm_blend_ps(verts[0], verts[1], 0x2); + break; + case 2: + case 3: + case 4: + case 5: + case 6: + case 7: + SWR_ASSERT(0); + break; + }; +} + +PA_STATE_OPT::PA_STATE_OPT(DRAW_CONTEXT *in_pDC, uint32_t in_numPrims, uint8_t* pStream, uint32_t in_streamSizeInVerts, + bool in_isStreaming, PRIMITIVE_TOPOLOGY topo) : PA_STATE(in_pDC, pStream, in_streamSizeInVerts), numPrims(in_numPrims), numPrimsComplete(0), numSimdPrims(0), + cur(0), prev(0), first(0), counter(0), reset(false), pfnPaFunc(nullptr), isStreaming(in_isStreaming) +{ + const API_STATE& state = GetApiState(pDC); + + this->binTopology = topo == TOP_UNKNOWN ? state.topology : topo; + + switch (this->binTopology) + { + case TOP_TRIANGLE_LIST: + this->pfnPaFunc = PaTriList0; + break; + case TOP_TRIANGLE_STRIP: + this->pfnPaFunc = PaTriStrip0; + break; + case TOP_TRIANGLE_FAN: + this->pfnPaFunc = PaTriFan0; + break; + case TOP_QUAD_LIST: + this->pfnPaFunc = PaQuadList0; + this->numPrims = in_numPrims * 2; // Convert quad primitives into triangles + break; + case TOP_QUAD_STRIP: + // quad strip pattern when decomposed into triangles is the same as verts strips + this->pfnPaFunc = PaTriStrip0; + this->numPrims = in_numPrims * 2; // Convert quad primitives into triangles + break; + case TOP_LINE_LIST: + this->pfnPaFunc = PaLineList0; + this->numPrims = in_numPrims; + break; + case TOP_LINE_STRIP: + this->pfnPaFunc = PaLineStrip0; + this->numPrims = in_numPrims; + break; + case TOP_LINE_LOOP: + this->pfnPaFunc = PaLineLoop0; + this->numPrims = in_numPrims; + break; + case TOP_POINT_LIST: + // use point binner and rasterizer if supported + this->pfnPaFunc = PaPoints0; + this->numPrims = in_numPrims; + break; + case TOP_RECT_LIST: + this->pfnPaFunc = PaRectList0; + this->numPrims = in_numPrims * 2; + break; + + case TOP_PATCHLIST_1: + this->pfnPaFunc = PaPatchList<1>; + break; + case TOP_PATCHLIST_2: + this->pfnPaFunc = PaPatchList<2>; + break; + case TOP_PATCHLIST_3: + this->pfnPaFunc = PaPatchList<3>; + break; + case TOP_PATCHLIST_4: + this->pfnPaFunc = PaPatchList<4>; + break; + case TOP_PATCHLIST_5: + this->pfnPaFunc = PaPatchList<5>; + break; + case TOP_PATCHLIST_6: + this->pfnPaFunc = PaPatchList<6>; + break; + case TOP_PATCHLIST_7: + this->pfnPaFunc = PaPatchList<7>; + break; + case TOP_PATCHLIST_8: + this->pfnPaFunc = PaPatchList<8>; + break; + case TOP_PATCHLIST_9: + this->pfnPaFunc = PaPatchList<9>; + break; + case TOP_PATCHLIST_10: + this->pfnPaFunc = PaPatchList<10>; + break; + case TOP_PATCHLIST_11: + this->pfnPaFunc = PaPatchList<11>; + break; + case TOP_PATCHLIST_12: + this->pfnPaFunc = PaPatchList<12>; + break; + case TOP_PATCHLIST_13: + this->pfnPaFunc = PaPatchList<13>; + break; + case TOP_PATCHLIST_14: + this->pfnPaFunc = PaPatchList<14>; + break; + case TOP_PATCHLIST_15: + this->pfnPaFunc = PaPatchList<15>; + break; + case TOP_PATCHLIST_16: + this->pfnPaFunc = PaPatchList<16>; + break; + case TOP_PATCHLIST_17: + this->pfnPaFunc = PaPatchList<17>; + break; + case TOP_PATCHLIST_18: + this->pfnPaFunc = PaPatchList<18>; + break; + case TOP_PATCHLIST_19: + this->pfnPaFunc = PaPatchList<19>; + break; + case TOP_PATCHLIST_20: + this->pfnPaFunc = PaPatchList<20>; + break; + case TOP_PATCHLIST_21: + this->pfnPaFunc = PaPatchList<21>; + break; + case TOP_PATCHLIST_22: + this->pfnPaFunc = PaPatchList<22>; + break; + case TOP_PATCHLIST_23: + this->pfnPaFunc = PaPatchList<23>; + break; + case TOP_PATCHLIST_24: + this->pfnPaFunc = PaPatchList<24>; + break; + case TOP_PATCHLIST_25: + this->pfnPaFunc = PaPatchList<25>; + break; + case TOP_PATCHLIST_26: + this->pfnPaFunc = PaPatchList<26>; + break; + case TOP_PATCHLIST_27: + this->pfnPaFunc = PaPatchList<27>; + break; + case TOP_PATCHLIST_28: + this->pfnPaFunc = PaPatchList<28>; + break; + case TOP_PATCHLIST_29: + this->pfnPaFunc = PaPatchList<29>; + break; + case TOP_PATCHLIST_30: + this->pfnPaFunc = PaPatchList<30>; + break; + case TOP_PATCHLIST_31: + this->pfnPaFunc = PaPatchList<31>; + break; + case TOP_PATCHLIST_32: + this->pfnPaFunc = PaPatchList<32>; + break; + + default: + SWR_ASSERT(0); + break; + }; + + this->pfnPaFuncReset = this->pfnPaFunc; + + // simdscalari id8 = _mm256_set_epi32(0, 1, 2, 3, 4, 5, 6, 7); + // simdscalari id4 = _mm256_set_epi32(0, 0, 1, 1, 2, 2, 3, 3); + simdscalari id8 = _mm256_set_epi32(7, 6, 5, 4, 3, 2, 1, 0); + simdscalari id4 = _mm256_set_epi32(3, 3, 2, 2, 1, 1, 0, 0); + + switch(this->binTopology) + { + case TOP_TRIANGLE_LIST: + case TOP_TRIANGLE_STRIP: + case TOP_TRIANGLE_FAN: + case TOP_LINE_STRIP: + case TOP_LINE_LIST: + case TOP_LINE_LOOP: + this->primIDIncr = 8; + this->primID = id8; + break; + case TOP_QUAD_LIST: + case TOP_QUAD_STRIP: + case TOP_RECT_LIST: + this->primIDIncr = 4; + this->primID = id4; + break; + case TOP_POINT_LIST: + this->primIDIncr = 8; + this->primID = id8; + break; + case TOP_PATCHLIST_1: + case TOP_PATCHLIST_2: + case TOP_PATCHLIST_3: + case TOP_PATCHLIST_4: + case TOP_PATCHLIST_5: + case TOP_PATCHLIST_6: + case TOP_PATCHLIST_7: + case TOP_PATCHLIST_8: + case TOP_PATCHLIST_9: + case TOP_PATCHLIST_10: + case TOP_PATCHLIST_11: + case TOP_PATCHLIST_12: + case TOP_PATCHLIST_13: + case TOP_PATCHLIST_14: + case TOP_PATCHLIST_15: + case TOP_PATCHLIST_16: + case TOP_PATCHLIST_17: + case TOP_PATCHLIST_18: + case TOP_PATCHLIST_19: + case TOP_PATCHLIST_20: + case TOP_PATCHLIST_21: + case TOP_PATCHLIST_22: + case TOP_PATCHLIST_23: + case TOP_PATCHLIST_24: + case TOP_PATCHLIST_25: + case TOP_PATCHLIST_26: + case TOP_PATCHLIST_27: + case TOP_PATCHLIST_28: + case TOP_PATCHLIST_29: + case TOP_PATCHLIST_30: + case TOP_PATCHLIST_31: + case TOP_PATCHLIST_32: + // Always run KNOB_SIMD_WIDTH number of patches at a time. + this->primIDIncr = 8; + this->primID = id8; + break; + + default: + SWR_ASSERT(0); + break; + }; + +} +#endif diff --git a/src/gallium/drivers/swr/rasterizer/core/rasterizer.cpp b/src/gallium/drivers/swr/rasterizer/core/rasterizer.cpp new file mode 100644 index 00000000000..587e336d87d --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/rasterizer.cpp @@ -0,0 +1,1393 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file rasterizer.cpp +* +* @brief Implementation for the rasterizer. +* +******************************************************************************/ + +#include +#include + +#include "rasterizer.h" +#include "multisample.h" +#include "rdtsc_core.h" +#include "backend.h" +#include "utils.h" +#include "frontend.h" +#include "tilemgr.h" +#include "memory/tilingtraits.h" + +void GetRenderHotTiles(DRAW_CONTEXT *pDC, uint32_t macroID, uint32_t x, uint32_t y, RenderOutputBuffers &renderBuffers, + uint32_t numSamples, uint32_t renderTargetArrayIndex); +void StepRasterTileX(uint32_t MaxRT, RenderOutputBuffers &buffers, uint32_t colorTileStep, uint32_t depthTileStep, uint32_t stencilTileStep); +void StepRasterTileY(uint32_t MaxRT, RenderOutputBuffers &buffers, RenderOutputBuffers &startBufferRow, + uint32_t colorRowStep, uint32_t depthRowStep, uint32_t stencilRowStep); + +#define MASKTOVEC(i3,i2,i1,i0) {-i0,-i1,-i2,-i3} +const __m128 gMaskToVec[] = { + MASKTOVEC(0,0,0,0), + MASKTOVEC(0,0,0,1), + MASKTOVEC(0,0,1,0), + MASKTOVEC(0,0,1,1), + MASKTOVEC(0,1,0,0), + MASKTOVEC(0,1,0,1), + MASKTOVEC(0,1,1,0), + MASKTOVEC(0,1,1,1), + MASKTOVEC(1,0,0,0), + MASKTOVEC(1,0,0,1), + MASKTOVEC(1,0,1,0), + MASKTOVEC(1,0,1,1), + MASKTOVEC(1,1,0,0), + MASKTOVEC(1,1,0,1), + MASKTOVEC(1,1,1,0), + MASKTOVEC(1,1,1,1), +}; + +const __m256d gMaskToVecpd[] = +{ + MASKTOVEC(0, 0, 0, 0), + MASKTOVEC(0, 0, 0, 1), + MASKTOVEC(0, 0, 1, 0), + MASKTOVEC(0, 0, 1, 1), + MASKTOVEC(0, 1, 0, 0), + MASKTOVEC(0, 1, 0, 1), + MASKTOVEC(0, 1, 1, 0), + MASKTOVEC(0, 1, 1, 1), + MASKTOVEC(1, 0, 0, 0), + MASKTOVEC(1, 0, 0, 1), + MASKTOVEC(1, 0, 1, 0), + MASKTOVEC(1, 0, 1, 1), + MASKTOVEC(1, 1, 0, 0), + MASKTOVEC(1, 1, 0, 1), + MASKTOVEC(1, 1, 1, 0), + MASKTOVEC(1, 1, 1, 1), +}; + +struct POS +{ + int32_t x, y; +}; + +struct EDGE +{ + double a, b; // a, b edge coefficients in fix8 + double stepQuadX; // step to adjacent horizontal quad in fix16 + double stepQuadY; // step to adjacent vertical quad in fix16 + double stepRasterTileX; // step to adjacent horizontal raster tile in fix16 + double stepRasterTileY; // step to adjacent vertical raster tile in fix16 + + __m256d vQuadOffsets; // offsets for 4 samples of a quad + __m256d vRasterTileOffsets; // offsets for the 4 corners of a raster tile +}; + +////////////////////////////////////////////////////////////////////////// +/// @brief rasterize a raster tile partially covered by the triangle +/// @param vEdge0-2 - edge equations evaluated at sample pos at each of the 4 corners of a raster tile +/// @param vA, vB - A & B coefs for each edge of the triangle (Ax + Bx + C) +/// @param vStepQuad0-2 - edge equations evaluated at the UL corners of the 2x2 pixel quad. +/// Used to step between quads when sweeping over the raster tile. +template +INLINE uint64_t rasterizePartialTile(DRAW_CONTEXT *pDC, double startEdges[NumEdges], EDGE *pRastEdges) +{ + uint64_t coverageMask = 0; + + __m256d vEdges[NumEdges]; + __m256d vStepX[NumEdges]; + __m256d vStepY[NumEdges]; + + for (uint32_t e = 0; e < NumEdges; ++e) + { + // Step to the pixel sample locations of the 1st quad + vEdges[e] = _mm256_add_pd(_mm256_set1_pd(startEdges[e]), pRastEdges[e].vQuadOffsets); + + // compute step to next quad (mul by 2 in x and y direction) + vStepX[e] = _mm256_set1_pd(pRastEdges[e].stepQuadX); + vStepY[e] = _mm256_set1_pd(pRastEdges[e].stepQuadY); + } + + // fast unrolled version for 8x8 tile +#if KNOB_TILE_X_DIM == 8 && KNOB_TILE_Y_DIM == 8 + int edgeMask[NumEdges]; + uint64_t mask; + + auto eval_lambda = [&](int e){edgeMask[e] = _mm256_movemask_pd(vEdges[e]);}; + auto update_lambda = [&](int e){mask &= edgeMask[e];}; + auto incx_lambda = [&](int e){vEdges[e] = _mm256_add_pd(vEdges[e], vStepX[e]);}; + auto incy_lambda = [&](int e){vEdges[e] = _mm256_add_pd(vEdges[e], vStepY[e]);}; + auto decx_lambda = [&](int e){vEdges[e] = _mm256_sub_pd(vEdges[e], vStepX[e]);}; + +// evaluate which pixels in the quad are covered +#define EVAL \ + UnrollerL<0, NumEdges, 1>::step(eval_lambda); + + // update coverage mask +#define UPDATE_MASK(bit) \ + mask = edgeMask[0]; \ + UnrollerL<1, NumEdges, 1>::step(update_lambda); \ + coverageMask |= (mask << bit); + + // step in the +x direction to the next quad +#define INCX \ + UnrollerL<0, NumEdges, 1>::step(incx_lambda); + + // step in the +y direction to the next quad +#define INCY \ + UnrollerL<0, NumEdges, 1>::step(incy_lambda); + + // step in the -x direction to the next quad +#define DECX \ + UnrollerL<0, NumEdges, 1>::step(decx_lambda); + + // sweep 2x2 quad back and forth through the raster tile, + // computing coverage masks for the entire tile + + // raster tile + // 0 1 2 3 4 5 6 7 + // x x + // x x ------------------> + // x x | + // <-----------------x x V + // .. + + // row 0 + EVAL; + UPDATE_MASK(0); + INCX; + EVAL; + UPDATE_MASK(4); + INCX; + EVAL; + UPDATE_MASK(8); + INCX; + EVAL; + UPDATE_MASK(12); + INCY; + + //row 1 + EVAL; + UPDATE_MASK(28); + DECX; + EVAL; + UPDATE_MASK(24); + DECX; + EVAL; + UPDATE_MASK(20); + DECX; + EVAL; + UPDATE_MASK(16); + INCY; + + // row 2 + EVAL; + UPDATE_MASK(32); + INCX; + EVAL; + UPDATE_MASK(36); + INCX; + EVAL; + UPDATE_MASK(40); + INCX; + EVAL; + UPDATE_MASK(44); + INCY; + + // row 3 + EVAL; + UPDATE_MASK(60); + DECX; + EVAL; + UPDATE_MASK(56); + DECX; + EVAL; + UPDATE_MASK(52); + DECX; + EVAL; + UPDATE_MASK(48); +#else + uint32_t bit = 0; + for (uint32_t y = 0; y < KNOB_TILE_Y_DIM/2; ++y) + { + __m256d vStartOfRowEdge[NumEdges]; + for (uint32_t e = 0; e < NumEdges; ++e) + { + vStartOfRowEdge[e] = vEdges[e]; + } + + for (uint32_t x = 0; x < KNOB_TILE_X_DIM/2; ++x) + { + int edgeMask[NumEdges]; + for (uint32_t e = 0; e < NumEdges; ++e) + { + edgeMask[e] = _mm256_movemask_pd(vEdges[e]); + } + + uint64_t mask = edgeMask[0]; + for (uint32_t e = 1; e < NumEdges; ++e) + { + mask &= edgeMask[e]; + } + coverageMask |= (mask << bit); + + // step to the next pixel in the x + for (uint32_t e = 0; e < NumEdges; ++e) + { + vEdges[e] = _mm256_add_pd(vEdges[e], vStepX[e]); + } + bit+=4; + } + + // step to the next row + for (uint32_t e = 0; e < NumEdges; ++e) + { + vEdges[e] = _mm256_add_pd(vStartOfRowEdge[e], vStepY[e]); + } + } +#endif + return coverageMask; + +} +// Top left rule: +// Top: if an edge is horizontal, and it is above other edges in tri pixel space, it is a 'top' edge +// Left: if an edge is not horizontal, and it is on the left side of the triangle in pixel space, it is a 'left' edge +// Top left: a sample is in if it is a top or left edge. +// Out: !(horizontal && above) = !horizontal && below +// Out: !horizontal && left = !(!horizontal && left) = horizontal and right +INLINE __m256d adjustTopLeftRuleIntFix16(const __m128i vA, const __m128i vB, const __m256d vEdge) +{ + // if vA < 0, vC-- + // if vA == 0 && vB < 0, vC-- + + __m256d vEdgeOut = vEdge; + __m256d vEdgeAdjust = _mm256_sub_pd(vEdge, _mm256_set1_pd(1.0)); + + // if vA < 0 (line is not horizontal and below) + int msk = _mm_movemask_ps(_mm_castsi128_ps(vA)); + + // if vA == 0 && vB < 0 (line is horizontal and we're on the left edge of a tri) + __m128i vCmp = _mm_cmpeq_epi32(vA, _mm_setzero_si128()); + int msk2 = _mm_movemask_ps(_mm_castsi128_ps(vCmp)); + msk2 &= _mm_movemask_ps(_mm_castsi128_ps(vB)); + + // if either of these are true and we're on the line (edge == 0), bump it outside the line + vEdgeOut = _mm256_blendv_pd(vEdgeOut, vEdgeAdjust, gMaskToVecpd[msk | msk2]); + return vEdgeOut; +} + +// max(abs(dz/dx), abs(dz,dy) +INLINE float ComputeMaxDepthSlope(const SWR_TRIANGLE_DESC* pDesc) +{ + /* + // evaluate i,j at (0,0) + float i00 = pDesc->I[0] * 0.0f + pDesc->I[1] * 0.0f + pDesc->I[2]; + float j00 = pDesc->J[0] * 0.0f + pDesc->J[1] * 0.0f + pDesc->J[2]; + + // evaluate i,j at (1,0) + float i10 = pDesc->I[0] * 1.0f + pDesc->I[1] * 0.0f + pDesc->I[2]; + float j10 = pDesc->J[0] * 1.0f + pDesc->J[1] * 0.0f + pDesc->J[2]; + + // compute dz/dx + float d00 = pDesc->Z[0] * i00 + pDesc->Z[1] * j00 + pDesc->Z[2]; + float d10 = pDesc->Z[0] * i10 + pDesc->Z[1] * j10 + pDesc->Z[2]; + float dzdx = abs(d10 - d00); + + // evaluate i,j at (0,1) + float i01 = pDesc->I[0] * 0.0f + pDesc->I[1] * 1.0f + pDesc->I[2]; + float j01 = pDesc->J[0] * 0.0f + pDesc->J[1] * 1.0f + pDesc->J[2]; + + float d01 = pDesc->Z[0] * i01 + pDesc->Z[1] * j01 + pDesc->Z[2]; + float dzdy = abs(d01 - d00); + */ + + // optimized version of above + float dzdx = fabsf(pDesc->recipDet * (pDesc->Z[0] * pDesc->I[0] + pDesc->Z[1] * pDesc->J[0])); + float dzdy = fabsf(pDesc->recipDet * (pDesc->Z[0] * pDesc->I[1] + pDesc->Z[1] * pDesc->J[1])); + + return std::max(dzdx, dzdy); +} + +INLINE float ComputeBiasFactor(const SWR_RASTSTATE* pState, const SWR_TRIANGLE_DESC* pDesc, const float* z) +{ + if (pState->depthFormat == R24_UNORM_X8_TYPELESS) + { + return (1.0f / (1 << 24)); + } + else if (pState->depthFormat == R16_UNORM) + { + return (1.0f / (1 << 16)); + } + else + { + SWR_ASSERT(pState->depthFormat == R32_FLOAT); + + // for f32 depth, factor = 2^(exponent(max(abs(z) - 23) + float zMax = std::max(fabsf(z[0]), std::max(fabsf(z[1]), fabsf(z[2]))); + uint32_t zMaxInt = *(uint32_t*)&zMax; + zMaxInt &= 0x7f800000; + zMax = *(float*)&zMaxInt; + + return zMax * (1.0f / (1 << 23)); + } +} + +INLINE float ComputeDepthBias(const SWR_RASTSTATE* pState, const SWR_TRIANGLE_DESC* pTri, const float* z) +{ + if (pState->depthBias == 0 && pState->slopeScaledDepthBias == 0) + { + return 0.0f; + } + + float scale = pState->slopeScaledDepthBias; + if (scale != 0.0f) + { + scale *= ComputeMaxDepthSlope(pTri); + } + + float bias = pState->depthBias * ComputeBiasFactor(pState, pTri, z) + scale; + if (pState->depthBiasClamp > 0.0f) + { + bias = std::min(bias, pState->depthBiasClamp); + } + else if (pState->depthBiasClamp < 0.0f) + { + bias = std::max(bias, pState->depthBiasClamp); + } + + return bias; +} + +// Prevent DCE by writing coverage mask from rasterizer to volatile +#if KNOB_ENABLE_TOSS_POINTS +__declspec(thread) volatile uint64_t gToss; +#endif + +static const uint32_t vertsPerTri = 3, componentsPerAttrib = 4; +// try to avoid _chkstk insertions; make this thread local +static THREAD OSALIGN(float, 16) perspAttribsTLS[vertsPerTri * KNOB_NUM_ATTRIBUTES * componentsPerAttrib]; + +INLINE +void ComputeEdgeData(int32_t a, int32_t b, EDGE& edge) +{ + edge.a = a; + edge.b = b; + + // compute constant steps to adjacent quads + edge.stepQuadX = (double)((int64_t)a * (int64_t)(2 * FIXED_POINT_SCALE)); + edge.stepQuadY = (double)((int64_t)b * (int64_t)(2 * FIXED_POINT_SCALE)); + + // compute constant steps to adjacent raster tiles + edge.stepRasterTileX = (double)((int64_t)a * (int64_t)(KNOB_TILE_X_DIM * FIXED_POINT_SCALE)); + edge.stepRasterTileY = (double)((int64_t)b * (int64_t)(KNOB_TILE_Y_DIM * FIXED_POINT_SCALE)); + + // compute quad offsets + const __m256d vQuadOffsetsXIntFix8 = _mm256_set_pd(FIXED_POINT_SCALE, 0, FIXED_POINT_SCALE, 0); + const __m256d vQuadOffsetsYIntFix8 = _mm256_set_pd(FIXED_POINT_SCALE, FIXED_POINT_SCALE, 0, 0); + + __m256d vQuadStepXFix16 = _mm256_mul_pd(_mm256_set1_pd(edge.a), vQuadOffsetsXIntFix8); + __m256d vQuadStepYFix16 = _mm256_mul_pd(_mm256_set1_pd(edge.b), vQuadOffsetsYIntFix8); + edge.vQuadOffsets = _mm256_add_pd(vQuadStepXFix16, vQuadStepYFix16); + + // compute raster tile offsets + const __m256d vTileOffsetsXIntFix8 = _mm256_set_pd((KNOB_TILE_X_DIM - 1)*FIXED_POINT_SCALE, 0, (KNOB_TILE_X_DIM - 1)*FIXED_POINT_SCALE, 0); + const __m256d vTileOffsetsYIntFix8 = _mm256_set_pd((KNOB_TILE_Y_DIM - 1)*FIXED_POINT_SCALE, (KNOB_TILE_Y_DIM - 1)*FIXED_POINT_SCALE, 0, 0); + + __m256d vTileStepXFix16 = _mm256_mul_pd(_mm256_set1_pd(edge.a), vTileOffsetsXIntFix8); + __m256d vTileStepYFix16 = _mm256_mul_pd(_mm256_set1_pd(edge.b), vTileOffsetsYIntFix8); + edge.vRasterTileOffsets = _mm256_add_pd(vTileStepXFix16, vTileStepYFix16); +} + +INLINE +void ComputeEdgeData(const POS& p0, const POS& p1, EDGE& edge) +{ + ComputeEdgeData(p0.y - p1.y, p1.x - p0.x, edge); +} + +template +void RasterizeTriangle(DRAW_CONTEXT* pDC, uint32_t workerId, uint32_t macroTile, void* pDesc) +{ + const TRIANGLE_WORK_DESC &workDesc = *((TRIANGLE_WORK_DESC*)pDesc); +#if KNOB_ENABLE_TOSS_POINTS + if (KNOB_TOSS_BIN_TRIS) + { + return; + } +#endif + RDTSC_START(BERasterizeTriangle); + + RDTSC_START(BETriangleSetup); + const API_STATE &state = GetApiState(pDC); + const SWR_RASTSTATE &rastState = state.rastState; + const BACKEND_FUNCS& backendFuncs = pDC->pState->backendFuncs; + + OSALIGN(SWR_TRIANGLE_DESC, 16) triDesc; + triDesc.pUserClipBuffer = workDesc.pUserClipBuffer; + + __m128 vX, vY, vZ, vRecipW; + + // pTriBuffer data layout: grouped components of the 3 triangle points and 1 don't care + // eg: vX = [x0 x1 x2 dc] + vX = _mm_load_ps(workDesc.pTriBuffer); + vY = _mm_load_ps(workDesc.pTriBuffer + 4); + vZ = _mm_load_ps(workDesc.pTriBuffer + 8); + vRecipW = _mm_load_ps(workDesc.pTriBuffer + 12); + + // convert to fixed point + __m128i vXi = fpToFixedPoint(vX); + __m128i vYi = fpToFixedPoint(vY); + + // quantize floating point position to fixed point precision + // to prevent attribute creep around the triangle vertices + vX = _mm_mul_ps(_mm_cvtepi32_ps(vXi), _mm_set1_ps(1.0f / FIXED_POINT_SCALE)); + vY = _mm_mul_ps(_mm_cvtepi32_ps(vYi), _mm_set1_ps(1.0f / FIXED_POINT_SCALE)); + + // triangle setup - A and B edge equation coefs + __m128 vA, vB; + triangleSetupAB(vX, vY, vA, vB); + + __m128i vAi, vBi; + triangleSetupABInt(vXi, vYi, vAi, vBi); + + // determinant + float det = calcDeterminantInt(vAi, vBi); + + /// @todo: This test is flipped...we have a stray '-' sign somewhere + // Convert CW triangles to CCW + if (det > 0.0) + { + vA = _mm_mul_ps(vA, _mm_set1_ps(-1)); + vB = _mm_mul_ps(vB, _mm_set1_ps(-1)); + vAi = _mm_mullo_epi32(vAi, _mm_set1_epi32(-1)); + vBi = _mm_mullo_epi32(vBi, _mm_set1_epi32(-1)); + det = -det; + } + + __m128 vC; + // Finish triangle setup - C edge coef + triangleSetupC(vX, vY, vA, vB, vC); + + // compute barycentric i and j + // i = (A1x + B1y + C1)/det + // j = (A2x + B2y + C2)/det + __m128 vDet = _mm_set1_ps(det); + __m128 vRecipDet = _mm_div_ps(_mm_set1_ps(1.0f), vDet);//_mm_rcp_ps(vDet); + _mm_store_ss(&triDesc.recipDet, vRecipDet); + + // only extract coefs for 2 of the barycentrics; the 3rd can be + // determined from the barycentric equation: + // i + j + k = 1 <=> k = 1 - j - i + _MM_EXTRACT_FLOAT(triDesc.I[0], vA, 1); + _MM_EXTRACT_FLOAT(triDesc.I[1], vB, 1); + _MM_EXTRACT_FLOAT(triDesc.I[2], vC, 1); + _MM_EXTRACT_FLOAT(triDesc.J[0], vA, 2); + _MM_EXTRACT_FLOAT(triDesc.J[1], vB, 2); + _MM_EXTRACT_FLOAT(triDesc.J[2], vC, 2); + + OSALIGN(float, 16) oneOverW[4]; + _mm_store_ps(oneOverW, vRecipW); + triDesc.OneOverW[0] = oneOverW[0] - oneOverW[2]; + triDesc.OneOverW[1] = oneOverW[1] - oneOverW[2]; + triDesc.OneOverW[2] = oneOverW[2]; + + // calculate perspective correct coefs per vertex attrib + float* pPerspAttribs = perspAttribsTLS; + float* pAttribs = workDesc.pAttribs; + triDesc.pPerspAttribs = pPerspAttribs; + triDesc.pAttribs = pAttribs; + float *pRecipW = workDesc.pTriBuffer + 12; + triDesc.pRecipW = pRecipW; + __m128 vOneOverWV0 = _mm_broadcast_ss(pRecipW); + __m128 vOneOverWV1 = _mm_broadcast_ss(pRecipW+=1); + __m128 vOneOverWV2 = _mm_broadcast_ss(pRecipW+=1); + for(uint32_t i = 0; i < workDesc.numAttribs; i++) + { + __m128 attribA = _mm_load_ps(pAttribs); + __m128 attribB = _mm_load_ps(pAttribs+=4); + __m128 attribC = _mm_load_ps(pAttribs+=4); + pAttribs+=4; + + attribA = _mm_mul_ps(attribA, vOneOverWV0); + attribB = _mm_mul_ps(attribB, vOneOverWV1); + attribC = _mm_mul_ps(attribC, vOneOverWV2); + + _mm_store_ps(pPerspAttribs, attribA); + _mm_store_ps(pPerspAttribs+=4, attribB); + _mm_store_ps(pPerspAttribs+=4, attribC); + pPerspAttribs+=4; + } + + // compute bary Z + // zInterp = zVert0 + i(zVert1-zVert0) + j (zVert2 - zVert0) + OSALIGN(float, 16) a[4]; + _mm_store_ps(a, vZ); + triDesc.Z[0] = a[0] - a[2]; + triDesc.Z[1] = a[1] - a[2]; + triDesc.Z[2] = a[2]; + + // add depth bias + triDesc.Z[2] += ComputeDepthBias(&rastState, &triDesc, workDesc.pTriBuffer + 8); + + // Compute edge data + OSALIGNSIMD(int32_t) aAi[4], aBi[4]; + _mm_store_si128((__m128i*)aAi, vAi); + _mm_store_si128((__m128i*)aBi, vBi); + + const uint32_t numEdges = 3 + (RasterizeScissorEdges ? 4 : 0); + EDGE rastEdges[7]; + + // compute triangle edges + ComputeEdgeData(aAi[0], aBi[0], rastEdges[0]); + ComputeEdgeData(aAi[1], aBi[1], rastEdges[1]); + ComputeEdgeData(aAi[2], aBi[2], rastEdges[2]); + + // compute scissor edges if enabled + if (RasterizeScissorEdges) + { + POS topLeft{state.scissorInFixedPoint.left, state.scissorInFixedPoint.top}; + POS bottomLeft{state.scissorInFixedPoint.left, state.scissorInFixedPoint.bottom}; + POS topRight{state.scissorInFixedPoint.right, state.scissorInFixedPoint.top}; + POS bottomRight{state.scissorInFixedPoint.right, state.scissorInFixedPoint.bottom}; + + // construct 4 scissor edges in ccw direction + ComputeEdgeData(topLeft, bottomLeft, rastEdges[3]); + ComputeEdgeData(bottomLeft, bottomRight, rastEdges[4]); + ComputeEdgeData(bottomRight, topRight, rastEdges[5]); + ComputeEdgeData(topRight, topLeft, rastEdges[6]); + } + + // Calc bounding box of triangle + OSALIGN(BBOX, 16) bbox; + calcBoundingBoxInt(vXi, vYi, bbox); + + // Intersect with scissor/viewport + bbox.left = std::max(bbox.left, state.scissorInFixedPoint.left); + bbox.right = std::min(bbox.right - 1, state.scissorInFixedPoint.right); + bbox.top = std::max(bbox.top, state.scissorInFixedPoint.top); + bbox.bottom = std::min(bbox.bottom - 1, state.scissorInFixedPoint.bottom); + + triDesc.triFlags = workDesc.triFlags; + + // further constrain backend to intersecting bounding box of macro tile and scissored triangle bbox + uint32_t macroX, macroY; + MacroTileMgr::getTileIndices(macroTile, macroX, macroY); + int32_t macroBoxLeft = macroX * KNOB_MACROTILE_X_DIM_FIXED; + int32_t macroBoxRight = macroBoxLeft + KNOB_MACROTILE_X_DIM_FIXED - 1; + int32_t macroBoxTop = macroY * KNOB_MACROTILE_Y_DIM_FIXED; + int32_t macroBoxBottom = macroBoxTop + KNOB_MACROTILE_Y_DIM_FIXED - 1; + + OSALIGN(BBOX, 16) intersect; + intersect.left = std::max(bbox.left, macroBoxLeft); + intersect.top = std::max(bbox.top, macroBoxTop); + intersect.right = std::min(bbox.right, macroBoxRight); + intersect.bottom = std::min(bbox.bottom, macroBoxBottom); + + SWR_ASSERT(intersect.left <= intersect.right && intersect.top <= intersect.bottom && intersect.left >= 0 && intersect.right >= 0 && intersect.top >= 0 && intersect.bottom >= 0); + + RDTSC_STOP(BETriangleSetup, 0, pDC->drawId); + + // update triangle desc + uint32_t tileX = intersect.left >> (KNOB_TILE_X_DIM_SHIFT + FIXED_POINT_SHIFT); + uint32_t tileY = intersect.top >> (KNOB_TILE_Y_DIM_SHIFT + FIXED_POINT_SHIFT); + uint32_t maxTileX = intersect.right >> (KNOB_TILE_X_DIM_SHIFT + FIXED_POINT_SHIFT); + uint32_t maxTileY = intersect.bottom >> (KNOB_TILE_Y_DIM_SHIFT + FIXED_POINT_SHIFT); + uint32_t numTilesX = maxTileX - tileX + 1; + uint32_t numTilesY = maxTileY - tileY + 1; + + if (numTilesX == 0 || numTilesY == 0) + { + RDTSC_EVENT(BEEmptyTriangle, 1, 0); + RDTSC_STOP(BERasterizeTriangle, 1, 0); + return; + } + + RDTSC_START(BEStepSetup); + + // Step to pixel center of top-left pixel of the triangle bbox + // Align intersect bbox (top/left) to raster tile's (top/left). + int32_t x = AlignDown(intersect.left, (FIXED_POINT_SCALE * KNOB_TILE_X_DIM)); + int32_t y = AlignDown(intersect.top, (FIXED_POINT_SCALE * KNOB_TILE_Y_DIM)); + + if(sampleCount == SWR_MULTISAMPLE_1X) + { + // Add 0.5, in fixed point, to offset to pixel center + x += (FIXED_POINT_SCALE / 2); + y += (FIXED_POINT_SCALE / 2); + } + + __m128i vTopLeftX = _mm_set1_epi32(x); + __m128i vTopLeftY = _mm_set1_epi32(y); + + // evaluate edge equations at top-left pixel using 64bit math + // all other evaluations will be 32bit steps from it + // small triangles could skip this and do all 32bit math + // edge 0 + // + // line = Ax + By + C + // solving for C: + // C = -Ax - By + // we know x0 and y0 are on the line; plug them in: + // C = -Ax0 - By0 + // plug C back into line equation: + // line = Ax - Bx - Ax0 - Bx1 + // line = A(x - x0) + B(y - y0) + // line = A(x0+dX) + B(y0+dY) + C = Ax0 + AdX + By0 + BdY + c = AdX + BdY + + // edge 0 and 1 + // edge0 = A0(x - x0) + B0(y - y0) + // edge1 = A1(x - x1) + B1(y - y1) + __m128i vDeltaX = _mm_sub_epi32(vTopLeftX, vXi); + __m128i vDeltaY = _mm_sub_epi32(vTopLeftY, vYi); + + __m256d vEdgeFix16[7]; + + // evaluate A(dx) and B(dY) for all points + __m256d vAipd = _mm256_cvtepi32_pd(vAi); + __m256d vBipd = _mm256_cvtepi32_pd(vBi); + __m256d vDeltaXpd = _mm256_cvtepi32_pd(vDeltaX); + __m256d vDeltaYpd = _mm256_cvtepi32_pd(vDeltaY); + + __m256d vAiDeltaXFix16 = _mm256_mul_pd(vAipd, vDeltaXpd); + __m256d vBiDeltaYFix16 = _mm256_mul_pd(vBipd, vDeltaYpd); + __m256d vEdge = _mm256_add_pd(vAiDeltaXFix16, vBiDeltaYFix16); + + // adjust for top-left rule + vEdge = adjustTopLeftRuleIntFix16(vAi, vBi, vEdge); + + // broadcast respective edge results to all lanes + double* pEdge = (double*)&vEdge; + vEdgeFix16[0] = _mm256_set1_pd(pEdge[0]); + vEdgeFix16[1] = _mm256_set1_pd(pEdge[1]); + vEdgeFix16[2] = _mm256_set1_pd(pEdge[2]); + + // evaluate edge equations for scissor edges + if (RasterizeScissorEdges) + { + const BBOX &scissor = state.scissorInFixedPoint; + vEdgeFix16[3] = _mm256_set1_pd((rastEdges[3].a * (x - scissor.left)) + (rastEdges[3].b * (y - scissor.top))); + vEdgeFix16[4] = _mm256_set1_pd((rastEdges[4].a * (x - scissor.left)) + (rastEdges[4].b * (y - scissor.bottom))); + vEdgeFix16[5] = _mm256_set1_pd((rastEdges[5].a * (x - scissor.right)) + (rastEdges[5].b * (y - scissor.bottom))); + vEdgeFix16[6] = _mm256_set1_pd((rastEdges[6].a * (x - scissor.right)) + (rastEdges[6].b * (y - scissor.top))); + } + + // Evaluate edge equations at sample positions of each of the 4 corners of a raster tile + // used to for testing if entire raster tile is inside a triangle + vEdgeFix16[0] = _mm256_add_pd(vEdgeFix16[0], rastEdges[0].vRasterTileOffsets); + vEdgeFix16[1] = _mm256_add_pd(vEdgeFix16[1], rastEdges[1].vRasterTileOffsets); + vEdgeFix16[2] = _mm256_add_pd(vEdgeFix16[2], rastEdges[2].vRasterTileOffsets); + + // at this point vEdge has been evaluated at the UL pixel corners of raster tile bbox + // step sample positions to the raster tile bbox of multisample points + // min(xSamples),min(ySamples) ------ max(xSamples),min(ySamples) + // | | + // | | + // min(xSamples),max(ySamples) ------ max(xSamples),max(ySamples) + __m256d vEdge0TileBbox, vEdge1TileBbox, vEdge2TileBbox; + if (sampleCount > SWR_MULTISAMPLE_1X) + { + __m128i vTileSampleBBoxXh = MultisampleTraits::TileSampleOffsetsX(); + __m128i vTileSampleBBoxYh = MultisampleTraits::TileSampleOffsetsY(); + + __m256d vTileSampleBBoxXFix8 = _mm256_cvtepi32_pd(vTileSampleBBoxXh); + __m256d vTileSampleBBoxYFix8 = _mm256_cvtepi32_pd(vTileSampleBBoxYh); + + // step edge equation tests from Tile + // used to for testing if entire raster tile is inside a triangle + __m256d vResultAxFix16 = _mm256_mul_pd(_mm256_set1_pd(rastEdges[0].a), vTileSampleBBoxXFix8); + __m256d vResultByFix16 = _mm256_mul_pd(_mm256_set1_pd(rastEdges[0].b), vTileSampleBBoxYFix8); + vEdge0TileBbox = _mm256_add_pd(vResultAxFix16, vResultByFix16); + + vResultAxFix16 = _mm256_mul_pd(_mm256_set1_pd(rastEdges[1].a), vTileSampleBBoxXFix8); + vResultByFix16 = _mm256_mul_pd(_mm256_set1_pd(rastEdges[1].b), vTileSampleBBoxYFix8); + vEdge1TileBbox = _mm256_add_pd(vResultAxFix16, vResultByFix16); + + vResultAxFix16 = _mm256_mul_pd(_mm256_set1_pd(rastEdges[2].a), vTileSampleBBoxXFix8); + vResultByFix16 = _mm256_mul_pd(_mm256_set1_pd(rastEdges[2].b), vTileSampleBBoxYFix8); + vEdge2TileBbox = _mm256_add_pd(vResultAxFix16, vResultByFix16); + } + + RDTSC_STOP(BEStepSetup, 0, pDC->drawId); + + uint32_t tY = tileY; + uint32_t tX = tileX; + uint32_t maxY = maxTileY; + uint32_t maxX = maxTileX; + + // compute steps between raster tiles for render output buffers + static const uint32_t colorRasterTileStep{(KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * (FormatTraits::bpp / 8)) * MultisampleTraits::numSamples}; + static const uint32_t colorRasterTileRowStep{(KNOB_MACROTILE_X_DIM / KNOB_TILE_X_DIM) * colorRasterTileStep}; + static const uint32_t depthRasterTileStep{(KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * (FormatTraits::bpp / 8)) * MultisampleTraits::numSamples}; + static const uint32_t depthRasterTileRowStep{(KNOB_MACROTILE_X_DIM / KNOB_TILE_X_DIM)* depthRasterTileStep}; + static const uint32_t stencilRasterTileStep{(KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * (FormatTraits::bpp / 8)) * MultisampleTraits::numSamples}; + static const uint32_t stencilRasterTileRowStep{(KNOB_MACROTILE_X_DIM / KNOB_TILE_X_DIM) * stencilRasterTileStep}; + RenderOutputBuffers renderBuffers, currentRenderBufferRow; + + GetRenderHotTiles(pDC, macroTile, tileX, tileY, renderBuffers, MultisampleTraits::numSamples, + triDesc.triFlags.renderTargetArrayIndex); + currentRenderBufferRow = renderBuffers; + + // rasterize and generate coverage masks per sample + uint32_t maxSamples = MultisampleTraits::numSamples; + for (uint32_t tileY = tY; tileY <= maxY; ++tileY) + { + __m256d vStartOfRowEdge[numEdges]; + for (uint32_t e = 0; e < numEdges; ++e) + { + vStartOfRowEdge[e] = vEdgeFix16[e]; + } + + for (uint32_t tileX = tX; tileX <= maxX; ++tileX) + { + uint64_t anyCoveredSamples = 0; + + // is the corner of the edge outside of the raster tile? (vEdge < 0) + int mask0, mask1, mask2; + if (sampleCount == SWR_MULTISAMPLE_1X) + { + mask0 = _mm256_movemask_pd(vEdgeFix16[0]); + mask1 = _mm256_movemask_pd(vEdgeFix16[1]); + mask2 = _mm256_movemask_pd(vEdgeFix16[2]); + } + else + { + __m256d vSampleBboxTest0, vSampleBboxTest1, vSampleBboxTest2; + // evaluate edge equations at the tile multisample bounding box + vSampleBboxTest0 = _mm256_add_pd(vEdge0TileBbox, vEdgeFix16[0]); + vSampleBboxTest1 = _mm256_add_pd(vEdge1TileBbox, vEdgeFix16[1]); + vSampleBboxTest2 = _mm256_add_pd(vEdge2TileBbox, vEdgeFix16[2]); + mask0 = _mm256_movemask_pd(vSampleBboxTest0); + mask1 = _mm256_movemask_pd(vSampleBboxTest1); + mask2 = _mm256_movemask_pd(vSampleBboxTest2); + } + + for (uint32_t sampleNum = 0; sampleNum < maxSamples; sampleNum++) + { + // trivial reject, at least one edge has all 4 corners of raster tile outside + bool trivialReject = (!(mask0 && mask1 && mask2)) ? true : false; + + if (!trivialReject) + { + // trivial accept mask + triDesc.coverageMask[sampleNum] = 0xffffffffffffffffULL; + if ((mask0 & mask1 & mask2) == 0xf) + { + anyCoveredSamples = triDesc.coverageMask[sampleNum]; + // trivial accept, all 4 corners of all 3 edges are negative + // i.e. raster tile completely inside triangle + RDTSC_EVENT(BETrivialAccept, 1, 0); + } + else + { + __m256d vEdge0AtSample, vEdge1AtSample, vEdge2AtSample; + if(sampleCount == SWR_MULTISAMPLE_1X) + { + // should get optimized out for single sample case (global value numbering or copy propagation) + vEdge0AtSample = vEdgeFix16[0]; + vEdge1AtSample = vEdgeFix16[1]; + vEdge2AtSample = vEdgeFix16[2]; + } + else + { + __m128i vSampleOffsetXh = MultisampleTraits::vXi(sampleNum); + __m128i vSampleOffsetYh = MultisampleTraits::vYi(sampleNum); + __m256d vSampleOffsetX = _mm256_cvtepi32_pd(vSampleOffsetXh); + __m256d vSampleOffsetY = _mm256_cvtepi32_pd(vSampleOffsetYh); + + // *note*: none of this needs to be vectorized as rasterizePartialTile just takes vEdge[0] + // for each edge and broadcasts it before offsetting to individual pixel quads + + // step edge equation tests from UL tile corner to pixel sample position + __m256d vResultAxFix16 = _mm256_mul_pd(_mm256_set1_pd(rastEdges[0].a), vSampleOffsetX); + __m256d vResultByFix16 = _mm256_mul_pd(_mm256_set1_pd(rastEdges[0].b), vSampleOffsetY); + vEdge0AtSample = _mm256_add_pd(vResultAxFix16, vResultByFix16); + vEdge0AtSample = _mm256_add_pd(vEdgeFix16[0], vEdge0AtSample); + + vResultAxFix16 = _mm256_mul_pd(_mm256_set1_pd(rastEdges[1].a), vSampleOffsetX); + vResultByFix16 = _mm256_mul_pd(_mm256_set1_pd(rastEdges[1].b), vSampleOffsetY); + vEdge1AtSample = _mm256_add_pd(vResultAxFix16, vResultByFix16); + vEdge1AtSample = _mm256_add_pd(vEdgeFix16[1], vEdge1AtSample); + + vResultAxFix16 = _mm256_mul_pd(_mm256_set1_pd(rastEdges[2].a), vSampleOffsetX); + vResultByFix16 = _mm256_mul_pd(_mm256_set1_pd(rastEdges[2].b), vSampleOffsetY); + vEdge2AtSample = _mm256_add_pd(vResultAxFix16, vResultByFix16); + vEdge2AtSample = _mm256_add_pd(vEdgeFix16[2], vEdge2AtSample); + } + + double startQuadEdges[numEdges]; + const __m256i vLane0Mask = _mm256_set_epi32(0, 0, 0, 0, 0, 0, -1, -1); + _mm256_maskstore_pd(&startQuadEdges[0], vLane0Mask, vEdge0AtSample); + _mm256_maskstore_pd(&startQuadEdges[1], vLane0Mask, vEdge1AtSample); + _mm256_maskstore_pd(&startQuadEdges[2], vLane0Mask, vEdge2AtSample); + + for (uint32_t e = 3; e < numEdges; ++e) + { + _mm256_maskstore_pd(&startQuadEdges[e], vLane0Mask, vEdgeFix16[e]); + } + + // not trivial accept or reject, must rasterize full tile + RDTSC_START(BERasterizePartial); + if (RasterizeScissorEdges) + { + triDesc.coverageMask[sampleNum] = rasterizePartialTile<7>(pDC, startQuadEdges, rastEdges); + } + else + { + triDesc.coverageMask[sampleNum] = rasterizePartialTile<3>(pDC, startQuadEdges, rastEdges); + } + RDTSC_STOP(BERasterizePartial, 0, 0); + + anyCoveredSamples |= triDesc.coverageMask[sampleNum]; + } + } + else + { + // if we're calculating coverage per sample, need to store it off. otherwise no covered samples, don't need to do anything + if(sampleCount > SWR_MULTISAMPLE_1X) + { + triDesc.coverageMask[sampleNum] = 0; + } + RDTSC_EVENT(BETrivialReject, 1, 0); + } + } + +#if KNOB_ENABLE_TOSS_POINTS + if(KNOB_TOSS_RS) + { + gToss = triDesc.coverageMask[0]; + } + else +#endif + if(anyCoveredSamples) + { + RDTSC_START(BEPixelBackend); + backendFuncs.pfnBackend(pDC, workerId, tileX << KNOB_TILE_X_DIM_SHIFT, tileY << KNOB_TILE_Y_DIM_SHIFT, triDesc, renderBuffers); + RDTSC_STOP(BEPixelBackend, 0, 0); + } + + // step to the next tile in X + for (uint32_t e = 0; e < numEdges; ++e) + { + vEdgeFix16[e] = _mm256_add_pd(vEdgeFix16[e], _mm256_set1_pd(rastEdges[e].stepRasterTileX)); + } + StepRasterTileX(state.psState.numRenderTargets, renderBuffers, colorRasterTileStep, depthRasterTileStep, stencilRasterTileStep); + } + + // step to the next tile in Y + for (uint32_t e = 0; e < numEdges; ++e) + { + vEdgeFix16[e] = _mm256_add_pd(vStartOfRowEdge[e], _mm256_set1_pd(rastEdges[e].stepRasterTileY)); + } + StepRasterTileY(state.psState.numRenderTargets, renderBuffers, currentRenderBufferRow, colorRasterTileRowStep, depthRasterTileRowStep, stencilRasterTileRowStep); + } + + RDTSC_STOP(BERasterizeTriangle, 1, 0); +} + +void RasterizeTriPoint(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void* pData) +{ + const TRIANGLE_WORK_DESC& workDesc = *(const TRIANGLE_WORK_DESC*)pData; + const SWR_RASTSTATE& rastState = pDC->pState->state.rastState; + const SWR_BACKEND_STATE& backendState = pDC->pState->state.backendState; + + bool isPointSpriteTexCoordEnabled = backendState.pointSpriteTexCoordMask != 0; + + // load point vertex + float x = *workDesc.pTriBuffer; + float y = *(workDesc.pTriBuffer + 1); + float z = *(workDesc.pTriBuffer + 2); + + // create a copy of the triangle buffer to write our adjusted vertices to + OSALIGNSIMD(float) newTriBuffer[4 * 4]; + TRIANGLE_WORK_DESC newWorkDesc = workDesc; + newWorkDesc.pTriBuffer = &newTriBuffer[0]; + + // create a copy of the attrib buffer to write our adjusted attribs to + OSALIGNSIMD(float) newAttribBuffer[4 * 3 * KNOB_NUM_ATTRIBUTES]; + newWorkDesc.pAttribs = &newAttribBuffer[0]; + + newWorkDesc.pUserClipBuffer = workDesc.pUserClipBuffer; + newWorkDesc.numAttribs = workDesc.numAttribs; + newWorkDesc.triFlags = workDesc.triFlags; + + // construct two tris by bloating point by point size + float halfPointSize = workDesc.triFlags.pointSize * 0.5f; + float lowerX = x - halfPointSize; + float upperX = x + halfPointSize; + float lowerY = y - halfPointSize; + float upperY = y + halfPointSize; + + // tri 0 + float *pBuf = &newTriBuffer[0]; + *pBuf++ = lowerX; + *pBuf++ = lowerX; + *pBuf++ = upperX; + pBuf++; + *pBuf++ = lowerY; + *pBuf++ = upperY; + *pBuf++ = upperY; + pBuf++; + _mm_store_ps(pBuf, _mm_set1_ps(z)); + _mm_store_ps(pBuf+=4, _mm_set1_ps(1.0f)); + + // setup triangle rasterizer function + PFN_WORK_FUNC pfnTriRast; + if (rastState.samplePattern == SWR_MSAA_STANDARD_PATTERN) + { + pfnTriRast = gRasterizerTable[rastState.scissorEnable][rastState.sampleCount]; + } + else + { + // for center sample pattern, all samples are at pixel center; calculate coverage + // once at center and broadcast the results in the backend + pfnTriRast = gRasterizerTable[rastState.scissorEnable][SWR_MULTISAMPLE_1X]; + } + + // overwrite texcoords for point sprites + if (isPointSpriteTexCoordEnabled) + { + // copy original attribs + memcpy(&newAttribBuffer[0], workDesc.pAttribs, 4 * 3 * workDesc.numAttribs * sizeof(float)); + newWorkDesc.pAttribs = &newAttribBuffer[0]; + + // overwrite texcoord for point sprites + uint32_t texCoordMask = backendState.pointSpriteTexCoordMask; + DWORD texCoordAttrib = 0; + + while (_BitScanForward(&texCoordAttrib, texCoordMask)) + { + texCoordMask &= ~(1 << texCoordAttrib); + __m128* pTexAttrib = (__m128*)&newAttribBuffer[0] + 3 * texCoordAttrib; + if (rastState.pointSpriteTopOrigin) + { + pTexAttrib[0] = _mm_set_ps(1, 0, 0, 0); + pTexAttrib[1] = _mm_set_ps(1, 0, 1, 0); + pTexAttrib[2] = _mm_set_ps(1, 0, 1, 1); + } + else + { + pTexAttrib[0] = _mm_set_ps(1, 0, 1, 0); + pTexAttrib[1] = _mm_set_ps(1, 0, 0, 0); + pTexAttrib[2] = _mm_set_ps(1, 0, 0, 1); + } + } + } + else + { + // no texcoord overwrite, can reuse the attrib buffer from frontend + newWorkDesc.pAttribs = workDesc.pAttribs; + } + + pfnTriRast(pDC, workerId, macroTile, (void*)&newWorkDesc); + + // tri 1 + pBuf = &newTriBuffer[0]; + *pBuf++ = lowerX; + *pBuf++ = upperX; + *pBuf++ = upperX; + pBuf++; + *pBuf++ = lowerY; + *pBuf++ = upperY; + *pBuf++ = lowerY; + // z, w unchanged + + if (isPointSpriteTexCoordEnabled) + { + uint32_t texCoordMask = backendState.pointSpriteTexCoordMask; + DWORD texCoordAttrib = 0; + + while (_BitScanForward(&texCoordAttrib, texCoordMask)) + { + texCoordMask &= ~(1 << texCoordAttrib); + __m128* pTexAttrib = (__m128*)&newAttribBuffer[0] + 3 * texCoordAttrib; + if (rastState.pointSpriteTopOrigin) + { + pTexAttrib[0] = _mm_set_ps(1, 0, 0, 0); + pTexAttrib[1] = _mm_set_ps(1, 0, 1, 1); + pTexAttrib[2] = _mm_set_ps(1, 0, 0, 1); + + } + else + { + pTexAttrib[0] = _mm_set_ps(1, 0, 1, 0); + pTexAttrib[1] = _mm_set_ps(1, 0, 0, 1); + pTexAttrib[2] = _mm_set_ps(1, 0, 1, 1); + } + } + } + + pfnTriRast(pDC, workerId, macroTile, (void*)&newWorkDesc); +} + +void RasterizeSimplePoint(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void* pData) +{ +#if KNOB_ENABLE_TOSS_POINTS + if (KNOB_TOSS_BIN_TRIS) + { + return; + } +#endif + + const TRIANGLE_WORK_DESC& workDesc = *(const TRIANGLE_WORK_DESC*)pData; + const BACKEND_FUNCS& backendFuncs = pDC->pState->backendFuncs; + + // map x,y relative offsets from start of raster tile to bit position in + // coverage mask for the point + static const uint32_t coverageMap[8][8] = { + { 0, 1, 4, 5, 8, 9, 12, 13 }, + { 2, 3, 6, 7, 10, 11, 14, 15 }, + { 16, 17, 20, 21, 24, 25, 28, 29 }, + { 18, 19, 22, 23, 26, 27, 30, 31 }, + { 32, 33, 36, 37, 40, 41, 44, 45 }, + { 34, 35, 38, 39, 42, 43, 46, 47 }, + { 48, 49, 52, 53, 56, 57, 60, 61 }, + { 50, 51, 54, 55, 58, 59, 62, 63 } + }; + + OSALIGN(SWR_TRIANGLE_DESC, 16) triDesc; + + // pull point information from triangle buffer + // @todo use structs for readability + uint32_t tileAlignedX = *(uint32_t*)workDesc.pTriBuffer; + uint32_t tileAlignedY = *(uint32_t*)(workDesc.pTriBuffer + 1); + float z = *(workDesc.pTriBuffer + 2); + + // construct triangle descriptor for point + // no interpolation, set up i,j for constant interpolation of z and attribs + // @todo implement an optimized backend that doesn't require triangle information + + // compute coverage mask from x,y packed into the coverageMask flag + // mask indices by the maximum valid index for x/y of coveragemap. + uint32_t tX = workDesc.triFlags.coverageMask & 0x7; + uint32_t tY = (workDesc.triFlags.coverageMask >> 4) & 0x7; + // todo: multisample points? + triDesc.coverageMask[0] = 1ULL << coverageMap[tY][tX]; + + // no persp divide needed for points + triDesc.pAttribs = triDesc.pPerspAttribs = workDesc.pAttribs; + triDesc.triFlags = workDesc.triFlags; + triDesc.recipDet = 1.0f; + triDesc.OneOverW[0] = triDesc.OneOverW[1] = triDesc.OneOverW[2] = 1.0f; + triDesc.I[0] = triDesc.I[1] = triDesc.I[2] = 0.0f; + triDesc.J[0] = triDesc.J[1] = triDesc.J[2] = 0.0f; + triDesc.Z[0] = triDesc.Z[1] = triDesc.Z[2] = z; + + RenderOutputBuffers renderBuffers; + GetRenderHotTiles(pDC, macroTile, tileAlignedX >> KNOB_TILE_X_DIM_SHIFT , tileAlignedY >> KNOB_TILE_Y_DIM_SHIFT, + renderBuffers, 1, triDesc.triFlags.renderTargetArrayIndex); + + RDTSC_START(BEPixelBackend); + backendFuncs.pfnBackend(pDC, workerId, tileAlignedX, tileAlignedY, triDesc, renderBuffers); + RDTSC_STOP(BEPixelBackend, 0, 0); +} + +// Get pointers to hot tile memory for color RT, depth, stencil +void GetRenderHotTiles(DRAW_CONTEXT *pDC, uint32_t macroID, uint32_t tileX, uint32_t tileY, RenderOutputBuffers &renderBuffers, + uint32_t numSamples, uint32_t renderTargetArrayIndex) +{ + const API_STATE& state = GetApiState(pDC); + SWR_CONTEXT *pContext = pDC->pContext; + + uint32_t mx, my; + MacroTileMgr::getTileIndices(macroID, mx, my); + tileX -= KNOB_MACROTILE_X_DIM_IN_TILES * mx; + tileY -= KNOB_MACROTILE_Y_DIM_IN_TILES * my; + + // compute tile offset for active hottile buffers + const uint32_t pitch = KNOB_MACROTILE_X_DIM * FormatTraits::bpp / 8; + uint32_t offset = ComputeTileOffset2D::bpp> >(pitch, tileX, tileY); + offset*=numSamples; + + unsigned long rtSlot = 0; + uint32_t colorHottileEnableMask = state.colorHottileEnable; + while(_BitScanForward(&rtSlot, colorHottileEnableMask)) + { + HOTTILE *pColor = pContext->pHotTileMgr->GetHotTile(pContext, pDC, macroID, (SWR_RENDERTARGET_ATTACHMENT)(SWR_ATTACHMENT_COLOR0 + rtSlot), true, + numSamples, renderTargetArrayIndex); + pColor->state = HOTTILE_DIRTY; + renderBuffers.pColor[rtSlot] = pColor->pBuffer + offset; + + colorHottileEnableMask &= ~(1 << rtSlot); + } + if(state.depthHottileEnable) + { + const uint32_t pitch = KNOB_MACROTILE_X_DIM * FormatTraits::bpp / 8; + uint32_t offset = ComputeTileOffset2D::bpp> >(pitch, tileX, tileY); + offset*=numSamples; + HOTTILE *pDepth = pContext->pHotTileMgr->GetHotTile(pContext, pDC, macroID, SWR_ATTACHMENT_DEPTH, true, + numSamples, renderTargetArrayIndex); + pDepth->state = HOTTILE_DIRTY; + SWR_ASSERT(pDepth->pBuffer != nullptr); + renderBuffers.pDepth = pDepth->pBuffer + offset; + } + if(state.stencilHottileEnable) + { + const uint32_t pitch = KNOB_MACROTILE_X_DIM * FormatTraits::bpp / 8; + uint32_t offset = ComputeTileOffset2D::bpp> >(pitch, tileX, tileY); + offset*=numSamples; + HOTTILE* pStencil = pContext->pHotTileMgr->GetHotTile(pContext, pDC, macroID, SWR_ATTACHMENT_STENCIL, true, + numSamples, renderTargetArrayIndex); + pStencil->state = HOTTILE_DIRTY; + SWR_ASSERT(pStencil->pBuffer != nullptr); + renderBuffers.pStencil = pStencil->pBuffer + offset; + } +} + +INLINE +void StepRasterTileX(uint32_t NumRT, RenderOutputBuffers &buffers, uint32_t colorTileStep, uint32_t depthTileStep, uint32_t stencilTileStep) +{ + for(uint32_t rt = 0; rt < NumRT; ++rt) + { + buffers.pColor[rt] += colorTileStep; + } + + buffers.pDepth += depthTileStep; + buffers.pStencil += stencilTileStep; +} + +INLINE +void StepRasterTileY(uint32_t NumRT, RenderOutputBuffers &buffers, RenderOutputBuffers &startBufferRow, uint32_t colorRowStep, uint32_t depthRowStep, uint32_t stencilRowStep) +{ + for(uint32_t rt = 0; rt < NumRT; ++rt) + { + startBufferRow.pColor[rt] += colorRowStep; + buffers.pColor[rt] = startBufferRow.pColor[rt]; + } + startBufferRow.pDepth += depthRowStep; + buffers.pDepth = startBufferRow.pDepth; + + startBufferRow.pStencil += stencilRowStep; + buffers.pStencil = startBufferRow.pStencil; +} + +// initialize rasterizer function table +PFN_WORK_FUNC gRasterizerTable[2][SWR_MULTISAMPLE_TYPE_MAX] = +{ + RasterizeTriangle, + RasterizeTriangle, + RasterizeTriangle, + RasterizeTriangle, + RasterizeTriangle, + RasterizeTriangle, + RasterizeTriangle, + RasterizeTriangle, + RasterizeTriangle, + RasterizeTriangle +}; + +void RasterizeLine(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pData) +{ + const TRIANGLE_WORK_DESC &workDesc = *((TRIANGLE_WORK_DESC*)pData); +#if KNOB_ENABLE_TOSS_POINTS + if (KNOB_TOSS_BIN_TRIS) + { + return; + } +#endif + + // bloat line to two tris and call the triangle rasterizer twice + RDTSC_START(BERasterizeLine); + + const API_STATE &state = GetApiState(pDC); + const SWR_RASTSTATE &rastState = state.rastState; + + // macrotile dimensioning + uint32_t macroX, macroY; + MacroTileMgr::getTileIndices(macroTile, macroX, macroY); + int32_t macroBoxLeft = macroX * KNOB_MACROTILE_X_DIM_FIXED; + int32_t macroBoxRight = macroBoxLeft + KNOB_MACROTILE_X_DIM_FIXED - 1; + int32_t macroBoxTop = macroY * KNOB_MACROTILE_Y_DIM_FIXED; + int32_t macroBoxBottom = macroBoxTop + KNOB_MACROTILE_Y_DIM_FIXED - 1; + + // create a copy of the triangle buffer to write our adjusted vertices to + OSALIGNSIMD(float) newTriBuffer[4 * 4]; + TRIANGLE_WORK_DESC newWorkDesc = workDesc; + newWorkDesc.pTriBuffer = &newTriBuffer[0]; + + // create a copy of the attrib buffer to write our adjusted attribs to + OSALIGNSIMD(float) newAttribBuffer[4 * 3 * KNOB_NUM_ATTRIBUTES]; + newWorkDesc.pAttribs = &newAttribBuffer[0]; + + const __m128 vBloat0 = _mm_set_ps(0.5f, -0.5f, -0.5f, 0.5f); + const __m128 vBloat1 = _mm_set_ps(0.5f, 0.5f, 0.5f, -0.5f); + + __m128 vX, vY, vZ, vRecipW; + + vX = _mm_load_ps(workDesc.pTriBuffer); + vY = _mm_load_ps(workDesc.pTriBuffer + 4); + vZ = _mm_load_ps(workDesc.pTriBuffer + 8); + vRecipW = _mm_load_ps(workDesc.pTriBuffer + 12); + + // triangle 0 + // v0,v1 -> v0,v0,v1 + __m128 vXa = _mm_shuffle_ps(vX, vX, _MM_SHUFFLE(1, 1, 0, 0)); + __m128 vYa = _mm_shuffle_ps(vY, vY, _MM_SHUFFLE(1, 1, 0, 0)); + __m128 vZa = _mm_shuffle_ps(vZ, vZ, _MM_SHUFFLE(1, 1, 0, 0)); + __m128 vRecipWa = _mm_shuffle_ps(vRecipW, vRecipW, _MM_SHUFFLE(1, 1, 0, 0)); + + __m128 vLineWidth = _mm_set1_ps(pDC->pState->state.rastState.lineWidth); + __m128 vAdjust = _mm_mul_ps(vLineWidth, vBloat0); + if (workDesc.triFlags.yMajor) + { + vXa = _mm_add_ps(vAdjust, vXa); + } + else + { + vYa = _mm_add_ps(vAdjust, vYa); + } + + // Store triangle description for rasterizer + _mm_store_ps((float*)&newTriBuffer[0], vXa); + _mm_store_ps((float*)&newTriBuffer[4], vYa); + _mm_store_ps((float*)&newTriBuffer[8], vZa); + _mm_store_ps((float*)&newTriBuffer[12], vRecipWa); + + // binner bins 3 edges for lines as v0, v1, v1 + // tri0 needs v0, v0, v1 + for (uint32_t a = 0; a < workDesc.numAttribs; ++a) + { + __m128 vAttrib0 = _mm_load_ps(&workDesc.pAttribs[a*12 + 0]); + __m128 vAttrib1 = _mm_load_ps(&workDesc.pAttribs[a*12 + 4]); + + _mm_store_ps((float*)&newAttribBuffer[a*12 + 0], vAttrib0); + _mm_store_ps((float*)&newAttribBuffer[a*12 + 4], vAttrib0); + _mm_store_ps((float*)&newAttribBuffer[a*12 + 8], vAttrib1); + } + + // Store user clip distances for triangle 0 + float newClipBuffer[3 * 8]; + uint32_t numClipDist = _mm_popcnt_u32(state.rastState.clipDistanceMask); + if (numClipDist) + { + newWorkDesc.pUserClipBuffer = newClipBuffer; + + float* pOldBuffer = workDesc.pUserClipBuffer; + float* pNewBuffer = newClipBuffer; + for (uint32_t i = 0; i < numClipDist; ++i) + { + // read barycentric coeffs from binner + float a = *(pOldBuffer++); + float b = *(pOldBuffer++); + + // reconstruct original clip distance at vertices + float c0 = a + b; + float c1 = b; + + // construct triangle barycentrics + *(pNewBuffer++) = c0 - c1; + *(pNewBuffer++) = c0 - c1; + *(pNewBuffer++) = c1; + } + } + + // make sure this macrotile intersects the triangle + __m128i vXai = fpToFixedPoint(vXa); + __m128i vYai = fpToFixedPoint(vYa); + OSALIGN(BBOX, 16) bboxA; + calcBoundingBoxInt(vXai, vYai, bboxA); + + if (!(bboxA.left > macroBoxRight || + bboxA.left > state.scissorInFixedPoint.right || + bboxA.right - 1 < macroBoxLeft || + bboxA.right - 1 < state.scissorInFixedPoint.left || + bboxA.top > macroBoxBottom || + bboxA.top > state.scissorInFixedPoint.bottom || + bboxA.bottom - 1 < macroBoxTop || + bboxA.bottom - 1 < state.scissorInFixedPoint.top)) { + // rasterize triangle + gRasterizerTable[rastState.scissorEnable][rastState.sampleCount](pDC, workerId, macroTile, (void*)&newWorkDesc); + } + + // triangle 1 + // v0,v1 -> v1,v1,v0 + vXa = _mm_shuffle_ps(vX, vX, _MM_SHUFFLE(1, 0, 1, 1)); + vYa = _mm_shuffle_ps(vY, vY, _MM_SHUFFLE(1, 0, 1, 1)); + vZa = _mm_shuffle_ps(vZ, vZ, _MM_SHUFFLE(1, 0, 1, 1)); + vRecipWa = _mm_shuffle_ps(vRecipW, vRecipW, _MM_SHUFFLE(1, 0, 1, 1)); + + vAdjust = _mm_mul_ps(vLineWidth, vBloat1); + if (workDesc.triFlags.yMajor) + { + vXa = _mm_add_ps(vAdjust, vXa); + } + else + { + vYa = _mm_add_ps(vAdjust, vYa); + } + + // Store triangle description for rasterizer + _mm_store_ps((float*)&newTriBuffer[0], vXa); + _mm_store_ps((float*)&newTriBuffer[4], vYa); + _mm_store_ps((float*)&newTriBuffer[8], vZa); + _mm_store_ps((float*)&newTriBuffer[12], vRecipWa); + + // binner bins 3 edges for lines as v0, v1, v1 + // tri1 needs v1, v1, v0 + for (uint32_t a = 0; a < workDesc.numAttribs; ++a) + { + __m128 vAttrib0 = _mm_load_ps(&workDesc.pAttribs[a * 12 + 0]); + __m128 vAttrib1 = _mm_load_ps(&workDesc.pAttribs[a * 12 + 4]); + + _mm_store_ps((float*)&newAttribBuffer[a * 12 + 0], vAttrib1); + _mm_store_ps((float*)&newAttribBuffer[a * 12 + 4], vAttrib1); + _mm_store_ps((float*)&newAttribBuffer[a * 12 + 8], vAttrib0); + } + + // store user clip distance for triangle 1 + if (numClipDist) + { + float* pOldBuffer = workDesc.pUserClipBuffer; + float* pNewBuffer = newClipBuffer; + for (uint32_t i = 0; i < numClipDist; ++i) + { + // read barycentric coeffs from binner + float a = *(pOldBuffer++); + float b = *(pOldBuffer++); + + // reconstruct original clip distance at vertices + float c0 = a + b; + float c1 = b; + + // construct triangle barycentrics + *(pNewBuffer++) = c1 - c0; + *(pNewBuffer++) = c1 - c0; + *(pNewBuffer++) = c0; + } + } + + vXai = fpToFixedPoint(vXa); + vYai = fpToFixedPoint(vYa); + calcBoundingBoxInt(vXai, vYai, bboxA); + + if (!(bboxA.left > macroBoxRight || + bboxA.left > state.scissorInFixedPoint.right || + bboxA.right - 1 < macroBoxLeft || + bboxA.right - 1 < state.scissorInFixedPoint.left || + bboxA.top > macroBoxBottom || + bboxA.top > state.scissorInFixedPoint.bottom || + bboxA.bottom - 1 < macroBoxTop || + bboxA.bottom - 1 < state.scissorInFixedPoint.top)) { + // rasterize triangle + gRasterizerTable[rastState.scissorEnable][rastState.sampleCount](pDC, workerId, macroTile, (void*)&newWorkDesc); + } + + RDTSC_STOP(BERasterizeLine, 1, 0); +} + diff --git a/src/gallium/drivers/swr/rasterizer/core/rasterizer.h b/src/gallium/drivers/swr/rasterizer/core/rasterizer.h new file mode 100644 index 00000000000..bcfeef48410 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/rasterizer.h @@ -0,0 +1,35 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file rasterizer.h +* +* @brief Definitions for the rasterizer. +* +******************************************************************************/ +#pragma once + +#include "context.h" + +extern PFN_WORK_FUNC gRasterizerTable[2][SWR_MULTISAMPLE_TYPE_MAX]; +void RasterizeLine(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pData); +void RasterizeSimplePoint(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pData); +void RasterizeTriPoint(DRAW_CONTEXT *pDC, uint32_t workerId, uint32_t macroTile, void *pData); diff --git a/src/gallium/drivers/swr/rasterizer/core/rdtsc_core.cpp b/src/gallium/drivers/swr/rasterizer/core/rdtsc_core.cpp new file mode 100644 index 00000000000..4b6b536075b --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/rdtsc_core.cpp @@ -0,0 +1,91 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +****************************************************************************/ + +#include "rdtsc_core.h" +#include "common/rdtsc_buckets.h" + +// must match CORE_BUCKETS enum order +BUCKET_DESC gCoreBuckets[] = { + { "APIClearRenderTarget", "", true, 0xff0b8bea }, + { "APIDraw", "", true, 0xff000066 }, + { "APIDrawWakeAllThreads", "", false, 0xffffffff }, + { "APIDrawIndexed", "", true, 0xff000066 }, + { "APIDispatch", "", true, 0xff660000 }, + { "APIStoreTiles", "", true, 0xff00ffff }, + { "APIGetDrawContext", "", false, 0xffffffff }, + { "APISync", "", true, 0xff6666ff }, + { "APIWaitForIdle", "", true, 0xff0000ff }, + { "FEProcessDraw", "", true, 0xff009900 }, + { "FEProcessDrawIndexed", "", true, 0xff009900 }, + { "FEFetchShader", "", false, 0xffffffff }, + { "FEVertexShader", "", false, 0xffffffff }, + { "FEHullShader", "", false, 0xffffffff }, + { "FETessellation", "", false, 0xffffffff }, + { "FEDomainShader", "", false, 0xffffffff }, + { "FEGeometryShader", "", false, 0xffffffff }, + { "FEStreamout", "", false, 0xffffffff }, + { "FEPAAssemble", "", false, 0xffffffff }, + { "FEBinPoints", "", false, 0xff29b854 }, + { "FEBinLines", "", false, 0xff29b854 }, + { "FEBinTriangles", "", false, 0xff29b854 }, + { "FETriangleSetup", "", false, 0xffffffff }, + { "FEViewportCull", "", false, 0xffffffff }, + { "FEGuardbandClip", "", false, 0xffffffff }, + { "FEClipPoints", "", false, 0xffffffff }, + { "FEClipLines", "", false, 0xffffffff }, + { "FEClipTriangles", "", false, 0xffffffff }, + { "FECullZeroAreaAndBackface", "", false, 0xffffffff }, + { "FECullBetweenCenters", "", false, 0xffffffff }, + { "FEProcessStoreTiles", "", true, 0xff39c864 }, + { "FEProcessInvalidateTiles", "", true, 0xffffffff }, + { "WorkerWorkOnFifoBE", "", false, 0xff40261c }, + { "WorkerFoundWork", "", false, 0xff573326 }, + { "BELoadTiles", "", true, 0xffb0e2ff }, + { "BEDispatch", "", true, 0xff00a2ff }, + { "BEClear", "", true, 0xff00ccbb }, + { "BERasterizeLine", "", true, 0xffb26a4e }, + { "BERasterizeTriangle", "", true, 0xffb26a4e }, + { "BETriangleSetup", "", false, 0xffffffff }, + { "BEStepSetup", "", false, 0xffffffff }, + { "BECullZeroArea", "", false, 0xffffffff }, + { "BEEmptyTriangle", "", false, 0xffffffff }, + { "BETrivialAccept", "", false, 0xffffffff }, + { "BETrivialReject", "", false, 0xffffffff }, + { "BERasterizePartial", "", false, 0xffffffff }, + { "BEPixelBackend", "", false, 0xffffffff }, + { "BESetup", "", false, 0xffffffff }, + { "BEBarycentric", "", false, 0xffffffff }, + { "BEEarlyDepthTest", "", false, 0xffffffff }, + { "BEPixelShader", "", false, 0xffffffff }, + { "BELateDepthTest", "", false, 0xffffffff }, + { "BEOutputMerger", "", false, 0xffffffff }, + { "BEStoreTiles", "", true, 0xff00cccc }, + { "BEEndTile", "", false, 0xffffffff }, + { "WorkerWaitForThreadEvent", "", false, 0xffffffff }, +}; + +/// @todo bucketmanager and mapping should probably be a part of the SWR context +std::vector gBucketMap; +BucketManager gBucketMgr(KNOB_BUCKETS_ENABLE_THREADVIZ); + +uint32_t gCurrentFrame = 0; diff --git a/src/gallium/drivers/swr/rasterizer/core/rdtsc_core.h b/src/gallium/drivers/swr/rasterizer/core/rdtsc_core.h new file mode 100644 index 00000000000..5fcc40bf8ee --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/rdtsc_core.h @@ -0,0 +1,177 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +****************************************************************************/ + +#pragma once +#include "knobs.h" + +#include "common/os.h" +#include "common/rdtsc_buckets.h" + +#include + +enum CORE_BUCKETS +{ + APIClearRenderTarget, + APIDraw, + APIDrawWakeAllThreads, + APIDrawIndexed, + APIDispatch, + APIStoreTiles, + APIGetDrawContext, + APISync, + APIWaitForIdle, + FEProcessDraw, + FEProcessDrawIndexed, + FEFetchShader, + FEVertexShader, + FEHullShader, + FETessellation, + FEDomainShader, + FEGeometryShader, + FEStreamout, + FEPAAssemble, + FEBinPoints, + FEBinLines, + FEBinTriangles, + FETriangleSetup, + FEViewportCull, + FEGuardbandClip, + FEClipPoints, + FEClipLines, + FEClipTriangles, + FECullZeroAreaAndBackface, + FECullBetweenCenters, + FEProcessStoreTiles, + FEProcessInvalidateTiles, + WorkerWorkOnFifoBE, + WorkerFoundWork, + BELoadTiles, + BEDispatch, + BEClear, + BERasterizeLine, + BERasterizeTriangle, + BETriangleSetup, + BEStepSetup, + BECullZeroArea, + BEEmptyTriangle, + BETrivialAccept, + BETrivialReject, + BERasterizePartial, + BEPixelBackend, + BESetup, + BEBarycentric, + BEEarlyDepthTest, + BEPixelShader, + BELateDepthTest, + BEOutputMerger, + BEStoreTiles, + BEEndTile, + WorkerWaitForThreadEvent, + + NumBuckets +}; + +void rdtscReset(); +void rdtscInit(int threadId); +void rdtscStart(uint32_t bucketId); +void rdtscStop(uint32_t bucketId, uint32_t count, uint64_t drawId); +void rdtscEvent(uint32_t bucketId, uint32_t count1, uint32_t count2); +void rdtscEndFrame(); + +#ifdef KNOB_ENABLE_RDTSC +#define RDTSC_RESET() rdtscReset() +#define RDTSC_INIT(threadId) rdtscInit(threadId) +#define RDTSC_START(bucket) rdtscStart(bucket) +#define RDTSC_STOP(bucket, count, draw) rdtscStop(bucket, count, draw) +#define RDTSC_EVENT(bucket, count1, count2) rdtscEvent(bucket, count1, count2) +#define RDTSC_ENDFRAME() rdtscEndFrame() +#else +#define RDTSC_RESET() +#define RDTSC_INIT(threadId) +#define RDTSC_START(bucket) +#define RDTSC_STOP(bucket, count, draw) +#define RDTSC_EVENT(bucket, count1, count2) +#define RDTSC_ENDFRAME() +#endif + +extern std::vector gBucketMap; +extern BucketManager gBucketMgr; +extern BUCKET_DESC gCoreBuckets[]; +extern uint32_t gCurrentFrame; + +INLINE void rdtscReset() +{ + gCurrentFrame = 0; + gBucketMgr.ClearThreads(); + gBucketMgr.ClearBuckets(); +} + +INLINE void rdtscInit(int threadId) +{ + // register all the buckets once + if (threadId == 0) + { + gBucketMap.resize(NumBuckets); + for (uint32_t i = 0; i < NumBuckets; ++i) + { + gBucketMap[i] = gBucketMgr.RegisterBucket(gCoreBuckets[i]); + } + } + + std::string name = threadId == 0 ? "API" : "WORKER"; + gBucketMgr.RegisterThread(name); +} + +INLINE void rdtscStart(uint32_t bucketId) +{ + uint32_t id = gBucketMap[bucketId]; + gBucketMgr.StartBucket(id); +} + +INLINE void rdtscStop(uint32_t bucketId, uint32_t count, uint64_t drawId) +{ + uint32_t id = gBucketMap[bucketId]; + gBucketMgr.StopBucket(id); +} + +INLINE void rdtscEvent(uint32_t bucketId, uint32_t count1, uint32_t count2) +{ + uint32_t id = gBucketMap[bucketId]; + gBucketMgr.AddEvent(id, count1); +} + +INLINE void rdtscEndFrame() +{ + gCurrentFrame++; + + if (gCurrentFrame == KNOB_BUCKETS_START_FRAME) + { + gBucketMgr.StartCapture(); + } + + if (gCurrentFrame == KNOB_BUCKETS_END_FRAME) + { + gBucketMgr.StopCapture(); + gBucketMgr.PrintReport("rdtsc.txt"); + } +} diff --git a/src/gallium/drivers/swr/rasterizer/core/state.h b/src/gallium/drivers/swr/rasterizer/core/state.h new file mode 100644 index 00000000000..2758555fd4b --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/state.h @@ -0,0 +1,1027 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file state.h +* +* @brief Definitions for API state. +* +******************************************************************************/ +#pragma once + +#include "common/formats.h" +#include "common/simdintrin.h" + +// clear flags +#define SWR_CLEAR_NONE 0 +#define SWR_CLEAR_COLOR (1 << 0) +#define SWR_CLEAR_DEPTH (1 << 1) +#define SWR_CLEAR_STENCIL (1 << 2) + +enum DRIVER_TYPE +{ + DX, + GL +}; + +////////////////////////////////////////////////////////////////////////// +/// PRIMITIVE_TOPOLOGY. +////////////////////////////////////////////////////////////////////////// +enum PRIMITIVE_TOPOLOGY +{ + TOP_UNKNOWN = 0x0, + TOP_POINT_LIST = 0x1, + TOP_LINE_LIST = 0x2, + TOP_LINE_STRIP = 0x3, + TOP_TRIANGLE_LIST = 0x4, + TOP_TRIANGLE_STRIP = 0x5, + TOP_TRIANGLE_FAN = 0x6, + TOP_QUAD_LIST = 0x7, + TOP_QUAD_STRIP = 0x8, + TOP_LINE_LIST_ADJ = 0x9, + TOP_LISTSTRIP_ADJ = 0xA, + TOP_TRI_LIST_ADJ = 0xB, + TOP_TRI_STRIP_ADJ = 0xC, + TOP_TRI_STRIP_REVERSE = 0xD, + TOP_POLYGON = 0xE, + TOP_RECT_LIST = 0xF, + TOP_LINE_LOOP = 0x10, + TOP_POINT_LIST_BF = 0x11, + TOP_LINE_STRIP_CONT = 0x12, + TOP_LINE_STRIP_BF = 0x13, + TOP_LINE_STRIP_CONT_BF = 0x14, + TOP_TRIANGLE_FAN_NOSTIPPLE = 0x16, + TOP_TRIANGLE_DISC = 0x17, /// @todo What is this?? + + TOP_PATCHLIST_BASE = 0x1F, // Invalid topology, used to calculate num verts for a patchlist. + TOP_PATCHLIST_1 = 0x20, // List of 1-vertex patches + TOP_PATCHLIST_2 = 0x21, + TOP_PATCHLIST_3 = 0x22, + TOP_PATCHLIST_4 = 0x23, + TOP_PATCHLIST_5 = 0x24, + TOP_PATCHLIST_6 = 0x25, + TOP_PATCHLIST_7 = 0x26, + TOP_PATCHLIST_8 = 0x27, + TOP_PATCHLIST_9 = 0x28, + TOP_PATCHLIST_10 = 0x29, + TOP_PATCHLIST_11 = 0x2A, + TOP_PATCHLIST_12 = 0x2B, + TOP_PATCHLIST_13 = 0x2C, + TOP_PATCHLIST_14 = 0x2D, + TOP_PATCHLIST_15 = 0x2E, + TOP_PATCHLIST_16 = 0x2F, + TOP_PATCHLIST_17 = 0x30, + TOP_PATCHLIST_18 = 0x31, + TOP_PATCHLIST_19 = 0x32, + TOP_PATCHLIST_20 = 0x33, + TOP_PATCHLIST_21 = 0x34, + TOP_PATCHLIST_22 = 0x35, + TOP_PATCHLIST_23 = 0x36, + TOP_PATCHLIST_24 = 0x37, + TOP_PATCHLIST_25 = 0x38, + TOP_PATCHLIST_26 = 0x39, + TOP_PATCHLIST_27 = 0x3A, + TOP_PATCHLIST_28 = 0x3B, + TOP_PATCHLIST_29 = 0x3C, + TOP_PATCHLIST_30 = 0x3D, + TOP_PATCHLIST_31 = 0x3E, + TOP_PATCHLIST_32 = 0x3F, // List of 32-vertex patches +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_SHADER_TYPE +////////////////////////////////////////////////////////////////////////// +enum SWR_SHADER_TYPE +{ + SHADER_VERTEX, + SHADER_GEOMETRY, + SHADER_DOMAIN, + SHADER_HULL, + SHADER_PIXEL, + SHADER_COMPUTE, + + NUM_SHADER_TYPES, +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_RENDERTARGET_ATTACHMENT +/// @todo Its not clear what an "attachment" means. Its not common term. +////////////////////////////////////////////////////////////////////////// +enum SWR_RENDERTARGET_ATTACHMENT +{ + SWR_ATTACHMENT_COLOR0, + SWR_ATTACHMENT_COLOR1, + SWR_ATTACHMENT_COLOR2, + SWR_ATTACHMENT_COLOR3, + SWR_ATTACHMENT_COLOR4, + SWR_ATTACHMENT_COLOR5, + SWR_ATTACHMENT_COLOR6, + SWR_ATTACHMENT_COLOR7, + SWR_ATTACHMENT_DEPTH, + SWR_ATTACHMENT_STENCIL, + + SWR_NUM_ATTACHMENTS +}; + +#define SWR_NUM_RENDERTARGETS 8 + +#define SWR_ATTACHMENT_COLOR0_BIT 0x001 +#define SWR_ATTACHMENT_COLOR1_BIT 0x002 +#define SWR_ATTACHMENT_COLOR2_BIT 0x004 +#define SWR_ATTACHMENT_COLOR3_BIT 0x008 +#define SWR_ATTACHMENT_COLOR4_BIT 0x010 +#define SWR_ATTACHMENT_COLOR5_BIT 0x020 +#define SWR_ATTACHMENT_COLOR6_BIT 0x040 +#define SWR_ATTACHMENT_COLOR7_BIT 0x080 +#define SWR_ATTACHMENT_DEPTH_BIT 0x100 +#define SWR_ATTACHMENT_STENCIL_BIT 0x200 +#define SWR_ATTACHMENT_MASK_ALL 0x3ff +#define SWR_ATTACHMENT_MASK_COLOR 0x0ff + + +////////////////////////////////////////////////////////////////////////// +/// @brief SWR Inner Tessellation factor ID +/// See above GetTessFactorOutputPosition code for documentation +enum SWR_INNER_TESSFACTOR_ID +{ + SWR_QUAD_U_TRI_INSIDE, + SWR_QUAD_V_INSIDE, + + SWR_NUM_INNER_TESS_FACTORS, +}; + +////////////////////////////////////////////////////////////////////////// +/// @brief SWR Outer Tessellation factor ID +/// See above GetTessFactorOutputPosition code for documentation +enum SWR_OUTER_TESSFACTOR_ID +{ + SWR_QUAD_U_EQ0_TRI_U_LINE_DETAIL, + SWR_QUAD_V_EQ0_TRI_V_LINE_DENSITY, + SWR_QUAD_U_EQ1_TRI_W, + SWR_QUAD_V_EQ1, + + SWR_NUM_OUTER_TESS_FACTORS, +}; + + +///////////////////////////////////////////////////////////////////////// +/// simdvertex +/// @brief Defines a vertex element that holds all the data for SIMD vertices. +/// Contains position in clip space, hardcoded to attribute 0, +/// space for up to 32 attributes, as well as any SGV values generated +/// by the pipeline +///////////////////////////////////////////////////////////////////////// +#define VERTEX_POSITION_SLOT 0 +#define VERTEX_ATTRIB_START_SLOT 1 +#define VERTEX_ATTRIB_END_SLOT 32 +#define VERTEX_RTAI_SLOT 33 // GS writes RenderTargetArrayIndex here +#define VERTEX_PRIMID_SLOT 34 // GS writes PrimId here +#define VERTEX_CLIPCULL_DIST_LO_SLOT 35 // VS writes lower 4 clip/cull dist +#define VERTEX_CLIPCULL_DIST_HI_SLOT 36 // VS writes upper 4 clip/cull dist +#define VERTEX_POINT_SIZE_SLOT 37 // VS writes point size here +static_assert(VERTEX_POINT_SIZE_SLOT < KNOB_NUM_ATTRIBUTES, "Mismatched attribute slot size"); + +// SoAoSoA +struct simdvertex +{ + simdvector attrib[KNOB_NUM_ATTRIBUTES]; +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_VS_CONTEXT +/// @brief Input to vertex shader +///////////////////////////////////////////////////////////////////////// +struct SWR_VS_CONTEXT +{ + simdvertex* pVin; // IN: SIMD input vertex data store + simdvertex* pVout; // OUT: SIMD output vertex data store + + uint32_t InstanceID; // IN: Instance ID, constant across all verts of the SIMD + simdscalari VertexID; // IN: Vertex ID + simdscalari mask; // IN: Active mask for shader +}; + +///////////////////////////////////////////////////////////////////////// +/// ScalarCPoint +/// @brief defines a control point element as passed from the output +/// of the hull shader to the input of the domain shader +///////////////////////////////////////////////////////////////////////// +struct ScalarAttrib +{ + float x; + float y; + float z; + float w; +}; + +struct ScalarCPoint +{ + ScalarAttrib attrib[KNOB_NUM_ATTRIBUTES]; +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_TESSELLATION_FACTORS +/// @brief Tessellation factors structure (non-vector) +///////////////////////////////////////////////////////////////////////// +struct SWR_TESSELLATION_FACTORS +{ + float OuterTessFactors[SWR_NUM_OUTER_TESS_FACTORS]; + float InnerTessFactors[SWR_NUM_INNER_TESS_FACTORS]; +}; + +#define MAX_NUM_VERTS_PER_PRIM 32 // support up to 32 control point patches +struct ScalarPatch +{ + SWR_TESSELLATION_FACTORS tessFactors; + ScalarCPoint cp[MAX_NUM_VERTS_PER_PRIM]; + ScalarCPoint patchData; +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_HS_CONTEXT +/// @brief Input to hull shader +///////////////////////////////////////////////////////////////////////// +struct SWR_HS_CONTEXT +{ + simdvertex vert[MAX_NUM_VERTS_PER_PRIM]; // IN: (SIMD) input primitive data + simdscalari PrimitiveID; // IN: (SIMD) primitive ID generated from the draw call + simdscalari mask; // IN: Active mask for shader + ScalarPatch* pCPout; // OUT: Output control point patch + // SIMD-sized-array of SCALAR patches +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_DS_CONTEXT +/// @brief Input to domain shader +///////////////////////////////////////////////////////////////////////// +struct SWR_DS_CONTEXT +{ + uint32_t PrimitiveID; // IN: (SCALAR) PrimitiveID for the patch associated with the DS invocation + uint32_t vectorOffset; // IN: (SCALAR) vector index offset into SIMD data. + uint32_t vectorStride; // IN: (SCALAR) stride (in vectors) of output data per attribute-component + ScalarPatch* pCpIn; // IN: (SCALAR) Control patch + simdscalar* pDomainU; // IN: (SIMD) Domain Point U coords + simdscalar* pDomainV; // IN: (SIMD) Domain Point V coords + simdscalari mask; // IN: Active mask for shader + simdscalar* pOutputData; // OUT: (SIMD) Vertex Attributes (2D array of vectors, one row per attribute-component) +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_GS_CONTEXT +/// @brief Input to geometry shader. +///////////////////////////////////////////////////////////////////////// +struct SWR_GS_CONTEXT +{ + simdvertex vert[MAX_NUM_VERTS_PER_PRIM]; // IN: input primitive data for SIMD prims + simdscalari PrimitiveID; // IN: input primitive ID generated from the draw call + uint32_t InstanceID; // IN: input instance ID + simdscalari mask; // IN: Active mask for shader + uint8_t* pStream; // OUT: output stream (contains vertices for all output streams) + uint8_t* pCutOrStreamIdBuffer; // OUT: cut or stream id buffer + simdscalari vertexCount; // OUT: num vertices emitted per SIMD lane +}; + +struct PixelPositions +{ + simdscalar UL; + simdscalar center; + simdscalar sample; + simdscalar centroid; +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_PS_CONTEXT +/// @brief Input to pixel shader. +///////////////////////////////////////////////////////////////////////// +struct SWR_PS_CONTEXT +{ + PixelPositions vX; // IN: x location(s) of pixels + PixelPositions vY; // IN: x location(s) of pixels + simdscalar vZ; // INOUT: z location of pixels + simdscalari activeMask; // OUT: mask for kill + simdscalar inputMask; // IN: input coverage mask for all samples + simdscalari oMask; // OUT: mask for output coverage + + PixelPositions vI; // barycentric coords evaluated at pixel center, sample position, centroid + PixelPositions vJ; + PixelPositions vOneOverW; // IN: 1/w + + const float* pAttribs; // IN: pointer to attribute barycentric coefficients + const float* pPerspAttribs; // IN: pointer to attribute/w barycentric coefficients + const float* pRecipW; // IN: pointer to 1/w coord for each vertex + const float *I; // IN: Barycentric A, B, and C coefs used to compute I + const float *J; // IN: Barycentric A, B, and C coefs used to compute J + float recipDet; // IN: 1/Det, used when barycentric interpolating attributes + const float* pSamplePosX; // IN: array of sample positions + const float* pSamplePosY; // IN: array of sample positions + simdvector shaded[SWR_NUM_RENDERTARGETS]; + // OUT: result color per rendertarget + + uint32_t frontFace; // IN: front- 1, back- 0 + uint32_t primID; // IN: primitive ID + uint32_t sampleIndex; // IN: sampleIndex +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_CS_CONTEXT +/// @brief Input to compute shader. +///////////////////////////////////////////////////////////////////////// +struct SWR_CS_CONTEXT +{ + // The ThreadGroupId is the current thread group index relative + // to all thread groups in the Dispatch call. The ThreadId, ThreadIdInGroup, + // and ThreadIdInGroupFlattened can be derived from ThreadGroupId in the shader. + + // Compute shader accepts the following system values. + // o ThreadId - Current thread id relative to all other threads in dispatch. + // o ThreadGroupId - Current thread group id relative to all other groups in dispatch. + // o ThreadIdInGroup - Current thread relative to all threads in the current thread group. + // o ThreadIdInGroupFlattened - Flattened linear id derived from ThreadIdInGroup. + // + // All of these system values can be computed in the shader. They will be + // derived from the current tile counter. The tile counter is an atomic counter that + // resides in the draw context and is initialized to the product of the dispatch dims. + // + // tileCounter = dispatchDims.x * dispatchDims.y * dispatchDims.z + // + // Each CPU worker thread will atomically decrement this counter and passes the current + // count into the shader. When the count reaches 0 then all thread groups in the + // dispatch call have been completed. + + uint32_t tileCounter; // The tile counter value for this thread group. + + // Dispatch dimensions used by shader to compute system values from the tile counter. + uint32_t dispatchDims[3]; + + uint8_t* pTGSM; // Thread Group Shared Memory pointer. + + uint8_t* pSpillFillBuffer; // Spill/fill buffer for barrier support +}; + +// enums +enum SWR_TILE_MODE +{ + SWR_TILE_NONE = 0x0, // Linear mode (no tiling) + SWR_TILE_MODE_WMAJOR, // W major tiling + SWR_TILE_MODE_XMAJOR, // X major tiling + SWR_TILE_MODE_YMAJOR, // Y major tiling + SWR_TILE_SWRZ, // SWR-Z tiling + + SWR_TILE_MODE_COUNT +}; + +enum SWR_SURFACE_TYPE +{ + SURFACE_1D = 0, + SURFACE_2D = 1, + SURFACE_3D = 2, + SURFACE_CUBE = 3, + SURFACE_BUFFER = 4, + SURFACE_STRUCTURED_BUFFER = 5, + SURFACE_NULL = 7 +}; + +enum SWR_ZFUNCTION +{ + ZFUNC_ALWAYS, + ZFUNC_NEVER, + ZFUNC_LT, + ZFUNC_EQ, + ZFUNC_LE, + ZFUNC_GT, + ZFUNC_NE, + ZFUNC_GE, + NUM_ZFUNC +}; + +enum SWR_STENCILOP +{ + STENCILOP_KEEP, + STENCILOP_ZERO, + STENCILOP_REPLACE, + STENCILOP_INCRSAT, + STENCILOP_DECRSAT, + STENCILOP_INCR, + STENCILOP_DECR, + STENCILOP_INVERT +}; + +enum SWR_BLEND_FACTOR +{ + BLENDFACTOR_ONE, + BLENDFACTOR_SRC_COLOR, + BLENDFACTOR_SRC_ALPHA, + BLENDFACTOR_DST_ALPHA, + BLENDFACTOR_DST_COLOR, + BLENDFACTOR_SRC_ALPHA_SATURATE, + BLENDFACTOR_CONST_COLOR, + BLENDFACTOR_CONST_ALPHA, + BLENDFACTOR_SRC1_COLOR, + BLENDFACTOR_SRC1_ALPHA, + BLENDFACTOR_ZERO, + BLENDFACTOR_INV_SRC_COLOR, + BLENDFACTOR_INV_SRC_ALPHA, + BLENDFACTOR_INV_DST_ALPHA, + BLENDFACTOR_INV_DST_COLOR, + BLENDFACTOR_INV_CONST_COLOR, + BLENDFACTOR_INV_CONST_ALPHA, + BLENDFACTOR_INV_SRC1_COLOR, + BLENDFACTOR_INV_SRC1_ALPHA +}; + +enum SWR_BLEND_OP +{ + BLENDOP_ADD, + BLENDOP_SUBTRACT, + BLENDOP_REVSUBTRACT, + BLENDOP_MIN, + BLENDOP_MAX, +}; + +enum SWR_LOGIC_OP +{ + LOGICOP_CLEAR, + LOGICOP_NOR, + LOGICOP_AND_INVERTED, + LOGICOP_COPY_INVERTED, + LOGICOP_AND_REVERSE, + LOGICOP_INVERT, + LOGICOP_XOR, + LOGICOP_NAND, + LOGICOP_AND, + LOGICOP_EQUIV, + LOGICOP_NOOP, + LOGICOP_OR_INVERTED, + LOGICOP_COPY, + LOGICOP_OR_REVERSE, + LOGICOP_OR, + LOGICOP_SET, +}; + +struct SWR_SURFACE_STATE +{ + uint8_t *pBaseAddress; + SWR_SURFACE_TYPE type; // @llvm_enum + SWR_FORMAT format; // @llvm_enum + uint32_t width; + uint32_t height; + uint32_t depth; + uint32_t numSamples; + uint32_t samplePattern; + uint32_t pitch; + uint32_t qpitch; + uint32_t minLod; // for sampled surfaces, the most detailed LOD that can be accessed by sampler + uint32_t maxLod; // for sampled surfaces, the max LOD that can be accessed + float resourceMinLod; // for sampled surfaces, the most detailed fractional mip that can be accessed by sampler + uint32_t lod; // for render targets, the lod being rendered to + uint32_t arrayIndex; // for render targets, the array index being rendered to for arrayed surfaces + SWR_TILE_MODE tileMode; // @llvm_enum + bool bInterleavedSamples; // are MSAA samples stored interleaved or planar + uint32_t halign; + uint32_t valign; + uint32_t xOffset; + uint32_t yOffset; + + uint32_t lodOffsets[2][15]; // lod offsets for sampled surfaces + + uint8_t *pAuxBaseAddress; // Used for compression, append/consume counter, etc. +}; + +// vertex fetch state +// WARNING- any changes to this struct need to be reflected +// in the fetch shader jit +struct SWR_VERTEX_BUFFER_STATE +{ + uint32_t index; + uint32_t pitch; + const uint8_t *pData; + uint32_t size; + uint32_t numaNode; + uint32_t maxVertex; // size / pitch. precalculated value used by fetch shader for OOB checks + uint32_t partialInboundsSize; // size % pitch. precalculated value used by fetch shader for partially OOB vertices +}; + +struct SWR_INDEX_BUFFER_STATE +{ + // Format type for indices (e.g. UINT16, UINT32, etc.) + SWR_FORMAT format; // @llvm_enum + const void *pIndices; + uint32_t size; +}; + + +////////////////////////////////////////////////////////////////////////// +/// SWR_FETCH_CONTEXT +/// @brief Input to fetch shader. +/// @note WARNING - Changes to this struct need to be reflected in the +/// fetch shader jit. +///////////////////////////////////////////////////////////////////////// +struct SWR_FETCH_CONTEXT +{ + const SWR_VERTEX_BUFFER_STATE* pStreams; // IN: array of bound vertex buffers + const int32_t* pIndices; // IN: pointer to index buffer for indexed draws + const int32_t* pLastIndex; // IN: pointer to end of index buffer, used for bounds checking + uint32_t CurInstance; // IN: current instance + uint32_t BaseVertex; // IN: base vertex + uint32_t StartVertex; // IN: start vertex + uint32_t StartInstance; // IN: start instance + simdscalari VertexID; // OUT: vector of vertex IDs + simdscalari CutMask; // OUT: vector mask of indices which have the cut index value +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_STATS +/// +/// @brief All statistics generated by SWR go here. These are public +/// to driver. +///////////////////////////////////////////////////////////////////////// +struct SWR_STATS +{ + // Occlusion Query + uint64_t DepthPassCount; // Number of passing depth tests. Not exact. + + // Pipeline Stats + uint64_t IaVertices; // Number of Fetch Shader vertices + uint64_t IaPrimitives; // Number of PA primitives. + uint64_t VsInvocations; // Number of Vertex Shader invocations + uint64_t HsInvocations; // Number of Hull Shader invocations + uint64_t DsInvocations; // Number of Domain Shader invocations + uint64_t GsInvocations; // Number of Geometry Shader invocations + uint64_t PsInvocations; // Number of Pixel Shader invocations + uint64_t CsInvocations; // Number of Compute Shader invocations + uint64_t CInvocations; // Number of clipper invocations + uint64_t CPrimitives; // Number of clipper primitives. + uint64_t GsPrimitives; // Number of prims GS outputs. + + // Streamout Stats + uint32_t SoWriteOffset[4]; + uint64_t SoPrimStorageNeeded[4]; + uint64_t SoNumPrimsWritten[4]; +}; + +////////////////////////////////////////////////////////////////////////// +/// STREAMOUT_BUFFERS +///////////////////////////////////////////////////////////////////////// + +#define MAX_SO_STREAMS 4 +#define MAX_ATTRIBUTES 32 + +struct SWR_STREAMOUT_BUFFER +{ + bool enable; + + // Pointers to streamout buffers. + uint32_t* pBuffer; + + // Size of buffer in dwords. + uint32_t bufferSize; + + // Vertex pitch of buffer in dwords. + uint32_t pitch; + + // Offset into buffer in dwords. SOS will increment this offset. + uint32_t streamOffset; + + // Offset to the SO write offset. If not null then we update offset here. + uint32_t* pWriteOffset; + +}; + +////////////////////////////////////////////////////////////////////////// +/// STREAMOUT_STATE +///////////////////////////////////////////////////////////////////////// +struct SWR_STREAMOUT_STATE +{ + // This disables stream output. + bool soEnable; + + // which streams are enabled for streamout + bool streamEnable[MAX_SO_STREAMS]; + + // If set then do not send any streams to the rasterizer. + bool rasterizerDisable; + + // Specifies which stream to send to the rasterizer. + uint32_t streamToRasterizer; + + // The stream masks specify which attributes are sent to which streams. + // These masks help the FE to setup the pPrimData buffer that is passed + // the the Stream Output Shader (SOS) function. + uint32_t streamMasks[MAX_SO_STREAMS]; + + // Number of attributes, including position, per vertex that are streamed out. + // This should match number of bits in stream mask. + uint32_t streamNumEntries[MAX_SO_STREAMS]; +}; + +////////////////////////////////////////////////////////////////////////// +/// STREAMOUT_CONTEXT - Passed to SOS +///////////////////////////////////////////////////////////////////////// +struct SWR_STREAMOUT_CONTEXT +{ + uint32_t* pPrimData; + SWR_STREAMOUT_BUFFER* pBuffer[MAX_SO_STREAMS]; + + // Num prims written for this stream + uint32_t numPrimsWritten; + + // Num prims that should have been written if there were no overflow. + uint32_t numPrimStorageNeeded; +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_GS_STATE - Geometry shader state +///////////////////////////////////////////////////////////////////////// +struct SWR_GS_STATE +{ + bool gsEnable; + + // number of input attributes per vertex. used by the frontend to + // optimize assembling primitives for GS + uint32_t numInputAttribs; + + // output topology - can be point, tristrip, or linestrip + PRIMITIVE_TOPOLOGY outputTopology; // @llvm_enum + + // maximum number of verts that can be emitted by a single instance of the GS + uint32_t maxNumVerts; + + // instance count + uint32_t instanceCount; + + // geometry shader emits renderTargetArrayIndex + bool emitsRenderTargetArrayIndex; + + // geometry shader emits PrimitiveID + bool emitsPrimitiveID; + + // if true, geometry shader emits a single stream, with separate cut buffer. + // if false, geometry shader emits vertices for multiple streams to the stream buffer, with a separate StreamID buffer + // to map vertices to streams + bool isSingleStream; + + // when single stream is enabled, singleStreamID dictates which stream is being output. + // field ignored if isSingleStream is false + uint32_t singleStreamID; +}; + + +////////////////////////////////////////////////////////////////////////// +/// SWR_TS_OUTPUT_TOPOLOGY - Defines data output by the tessellator / DS +///////////////////////////////////////////////////////////////////////// +enum SWR_TS_OUTPUT_TOPOLOGY +{ + SWR_TS_OUTPUT_POINT, + SWR_TS_OUTPUT_LINE, + SWR_TS_OUTPUT_TRI_CW, + SWR_TS_OUTPUT_TRI_CCW, + + SWR_TS_OUTPUT_TOPOLOGY_COUNT +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_TS_PARTITIONING - Defines tessellation algorithm +///////////////////////////////////////////////////////////////////////// +enum SWR_TS_PARTITIONING +{ + SWR_TS_INTEGER, + SWR_TS_ODD_FRACTIONAL, + SWR_TS_EVEN_FRACTIONAL, + + SWR_TS_PARTITIONING_COUNT +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_TS_DOMAIN - Defines Tessellation Domain +///////////////////////////////////////////////////////////////////////// +enum SWR_TS_DOMAIN +{ + SWR_TS_QUAD, + SWR_TS_TRI, + SWR_TS_ISOLINE, + + SWR_TS_DOMAIN_COUNT +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_TS_STATE - Tessellation state +///////////////////////////////////////////////////////////////////////// +struct SWR_TS_STATE +{ + bool tsEnable; + SWR_TS_OUTPUT_TOPOLOGY tsOutputTopology; // @llvm_enum + SWR_TS_PARTITIONING partitioning; // @llvm_enum + SWR_TS_DOMAIN domain; // @llvm_enum + + PRIMITIVE_TOPOLOGY postDSTopology; // @llvm_enum + + uint32_t numHsInputAttribs; + uint32_t numHsOutputAttribs; + uint32_t numDsOutputAttribs; +}; + +// output merger state +struct SWR_RENDER_TARGET_BLEND_STATE +{ + uint8_t writeDisableRed : 1; + uint8_t writeDisableGreen : 1; + uint8_t writeDisableBlue : 1; + uint8_t writeDisableAlpha : 1; +}; +static_assert(sizeof(SWR_RENDER_TARGET_BLEND_STATE) == 1, "Invalid SWR_RENDER_TARGET_BLEND_STATE size"); + +#define SWR_MAX_NUM_MULTISAMPLES 16 +enum SWR_MULTISAMPLE_COUNT +{ + SWR_MULTISAMPLE_1X = 0, + SWR_MULTISAMPLE_2X, + SWR_MULTISAMPLE_4X, + SWR_MULTISAMPLE_8X, + SWR_MULTISAMPLE_16X, + SWR_MULTISAMPLE_TYPE_MAX +}; + +struct SWR_BLEND_STATE +{ + // constant blend factor color in RGBA float + float constantColor[4]; + + // alpha test reference value in unorm8 or float32 + uint32_t alphaTestReference; + uint32_t sampleMask; + // all RT's have the same sample count + ///@todo move this to Output Merger state when we refactor + SWR_MULTISAMPLE_COUNT sampleCount; // @llvm_enum + + SWR_RENDER_TARGET_BLEND_STATE renderTarget[SWR_NUM_RENDERTARGETS]; +}; +static_assert(sizeof(SWR_BLEND_STATE) == 36, "Invalid SWR_BLEND_STATE size"); + +////////////////////////////////////////////////////////////////////////// +/// FUNCTION POINTERS FOR SHADERS + +typedef void(__cdecl *PFN_FETCH_FUNC)(SWR_FETCH_CONTEXT& fetchInfo, simdvertex& out); +typedef void(__cdecl *PFN_VERTEX_FUNC)(HANDLE hPrivateData, SWR_VS_CONTEXT* pVsContext); +typedef void(__cdecl *PFN_HS_FUNC)(HANDLE hPrivateData, SWR_HS_CONTEXT* pHsContext); +typedef void(__cdecl *PFN_DS_FUNC)(HANDLE hPrivateData, SWR_DS_CONTEXT* pDsContext); +typedef void(__cdecl *PFN_GS_FUNC)(HANDLE hPrivateData, SWR_GS_CONTEXT* pGsContext); +typedef void(__cdecl *PFN_CS_FUNC)(HANDLE hPrivateData, SWR_CS_CONTEXT* pCsContext); +typedef void(__cdecl *PFN_SO_FUNC)(SWR_STREAMOUT_CONTEXT& soContext); +typedef void(__cdecl *PFN_PIXEL_KERNEL)(HANDLE hPrivateData, SWR_PS_CONTEXT *pContext); +typedef void(__cdecl *PFN_BLEND_JIT_FUNC)(const SWR_BLEND_STATE*, simdvector&, simdvector&, uint32_t, BYTE*, simdvector&, simdscalari*, simdscalari*); + +////////////////////////////////////////////////////////////////////////// +/// FRONTEND_STATE +///////////////////////////////////////////////////////////////////////// +struct SWR_FRONTEND_STATE +{ + // skip clip test, perspective divide, and viewport transform + // intended for verts in screen space + bool vpTransformDisable; + union + { + struct + { + uint32_t triFan : 2; + uint32_t lineStripList : 1; + uint32_t triStripList : 2; + }; + uint32_t bits; + }provokingVertex; + uint32_t topologyProvokingVertex; // provoking vertex for the draw topology +}; + +////////////////////////////////////////////////////////////////////////// +/// VIEWPORT_MATRIX +///////////////////////////////////////////////////////////////////////// +struct SWR_VIEWPORT_MATRIX +{ + float m00; + float m11; + float m22; + float m30; + float m31; + float m32; +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_VIEWPORT +///////////////////////////////////////////////////////////////////////// +struct SWR_VIEWPORT +{ + float x; + float y; + float width; + float height; + float minZ; + float maxZ; +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_CULLMODE +////////////////////////////////////////////////////////////////////////// +enum SWR_CULLMODE +{ + SWR_CULLMODE_BOTH, + SWR_CULLMODE_NONE, + SWR_CULLMODE_FRONT, + SWR_CULLMODE_BACK +}; + +enum SWR_FILLMODE +{ + SWR_FILLMODE_POINT, + SWR_FILLMODE_WIREFRAME, + SWR_FILLMODE_SOLID +}; + +enum SWR_FRONTWINDING +{ + SWR_FRONTWINDING_CW, + SWR_FRONTWINDING_CCW +}; + + +enum SWR_MSAA_SAMPLE_PATTERN +{ + SWR_MSAA_CENTER_PATTERN, + SWR_MSAA_STANDARD_PATTERN, + SWR_MSAA_SAMPLE_PATTERN_MAX +}; + +enum SWR_PIXEL_LOCATION +{ + SWR_PIXEL_LOCATION_CENTER, + SWR_PIXEL_LOCATION_UL, +}; + +// fixed point screen space sample locations within a pixel +struct SWR_MULTISAMPLE_POS +{ + uint32_t x; + uint32_t y; +}; + +enum SWR_MSAA_RASTMODE +{ + SWR_MSAA_RASTMODE_OFF_PIXEL, + SWR_MSAA_RASTMODE_OFF_PATTERN, + SWR_MSAA_RASTMODE_ON_PIXEL, + SWR_MSAA_RASTMODE_ON_PATTERN +}; + +////////////////////////////////////////////////////////////////////////// +/// SWR_RASTSTATE +////////////////////////////////////////////////////////////////////////// +struct SWR_RASTSTATE +{ + uint32_t cullMode : 2; + uint32_t fillMode : 2; + uint32_t frontWinding : 1; + uint32_t scissorEnable : 1; + uint32_t depthClipEnable : 1; + float pointSize; + float lineWidth; + + // point size output from the VS + bool pointParam; + + // point sprite + bool pointSpriteEnable; + bool pointSpriteTopOrigin; + + // depth bias + float depthBias; + float slopeScaledDepthBias; + float depthBiasClamp; + SWR_FORMAT depthFormat; // @llvm_enum + + ///@todo: MSAA lines + // multisample state for MSAA lines + bool msaaRastEnable; + SWR_MSAA_RASTMODE rastMode; // @llvm_enum + + // sample count the rasterizer is running at + SWR_MULTISAMPLE_COUNT sampleCount; // @llvm_enum + bool bForcedSampleCount; + uint32_t pixelLocation; // UL or Center + bool pixelOffset; // offset pixel positions by .5 in both the horizontal and vertical direction + SWR_MULTISAMPLE_POS iSamplePos[SWR_MAX_NUM_MULTISAMPLES]; + SWR_MSAA_SAMPLE_PATTERN samplePattern; // @llvm_enum + + // user clip/cull distance enables + uint8_t cullDistanceMask; + uint8_t clipDistanceMask; +}; + +// backend state +struct SWR_BACKEND_STATE +{ + uint32_t constantInterpolationMask; + uint32_t pointSpriteTexCoordMask; + uint8_t numAttributes; + uint8_t numComponents[KNOB_NUM_ATTRIBUTES]; +}; + +union SWR_DEPTH_STENCIL_STATE +{ + struct + { + // dword 0 + uint32_t depthWriteEnable : 1; + uint32_t depthTestEnable : 1; + uint32_t stencilWriteEnable : 1; + uint32_t stencilTestEnable : 1; + uint32_t doubleSidedStencilTestEnable : 1; + + uint32_t depthTestFunc : 3; + uint32_t stencilTestFunc : 3; + + uint32_t backfaceStencilPassDepthPassOp : 3; + uint32_t backfaceStencilPassDepthFailOp : 3; + uint32_t backfaceStencilFailOp : 3; + uint32_t backfaceStencilTestFunc : 3; + uint32_t stencilPassDepthPassOp : 3; + uint32_t stencilPassDepthFailOp : 3; + uint32_t stencilFailOp : 3; + + // dword 1 + uint8_t backfaceStencilWriteMask; + uint8_t backfaceStencilTestMask; + uint8_t stencilWriteMask; + uint8_t stencilTestMask; + + // dword 2 + uint8_t backfaceStencilRefValue; + uint8_t stencilRefValue; + }; + uint32_t value[3]; +}; + +enum SWR_SHADING_RATE +{ + SWR_SHADING_RATE_PIXEL, + SWR_SHADING_RATE_SAMPLE, + SWR_SHADING_RATE_COARSE, + SWR_SHADING_RATE_MAX, +}; + +enum SWR_INPUT_COVERAGE +{ + SWR_INPUT_COVERAGE_NONE, + SWR_INPUT_COVERAGE_NORMAL, + SWR_INPUT_COVERAGE_MAX, +}; + +enum SWR_PS_POSITION_OFFSET +{ + SWR_PS_POSITION_SAMPLE_NONE, + SWR_PS_POSITION_SAMPLE_OFFSET, + SWR_PS_POSITION_CENTROID_OFFSET, + SWR_PS_POSITION_OFFSET_MAX, +}; + +enum SWR_BARYCENTRICS_MASK +{ + SWR_BARYCENTRIC_PER_PIXEL_MASK = 0x1, + SWR_BARYCENTRIC_CENTROID_MASK = 0x2, + SWR_BARYCENTRIC_PER_SAMPLE_MASK = 0x4, + SWR_BARYCENTRICS_MASK_MAX = 0x8 +}; + +// pixel shader state +struct SWR_PS_STATE +{ + // dword 0-1 + PFN_PIXEL_KERNEL pfnPixelShader; // @llvm_pfn + + // dword 2 + uint32_t killsPixel : 1; // pixel shader can kill pixels + uint32_t inputCoverage : 1; // type of input coverage PS uses + uint32_t writesODepth : 1; // pixel shader writes to depth + uint32_t usesSourceDepth : 1; // pixel shader reads depth + uint32_t shadingRate : 2; // shading per pixel / sample / coarse pixel + uint32_t numRenderTargets : 4; // number of render target outputs in use (0-8) + uint32_t posOffset : 2; // type of offset (none, sample, centroid) to add to pixel position + uint32_t barycentricsMask : 3; // which type(s) of barycentric coords does the PS interpolate attributes with + uint32_t usesUAV : 1; // pixel shader accesses UAV + uint32_t forceEarlyZ : 1; // force execution of early depth/stencil test +}; diff --git a/src/gallium/drivers/swr/rasterizer/core/tessellator.h b/src/gallium/drivers/swr/rasterizer/core/tessellator.h new file mode 100644 index 00000000000..915ac77897b --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/tessellator.h @@ -0,0 +1,88 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file tessellator.h +* +* @brief Tessellator fixed function unit interface definition +* +******************************************************************************/ +#pragma once + +/// Allocate and initialize a new tessellation context +HANDLE SWR_API TSInitCtx( + SWR_TS_DOMAIN tsDomain, ///< [IN] Tessellation domain (isoline, quad, triangle) + SWR_TS_PARTITIONING tsPartitioning, ///< [IN] Tessellation partitioning algorithm + SWR_TS_OUTPUT_TOPOLOGY tsOutputTopology, ///< [IN] Tessellation output topology + void* pContextMem, ///< [IN] Memory to use for the context + size_t& memSize); ///< [INOUT] In: Amount of memory in pContextMem. Out: Mem required + +/// Destroy & de-allocate tessellation context +void SWR_API TSDestroyCtx( + HANDLE tsCtx); ///< [IN] Tessellation context to be destroyed + +struct SWR_TS_TESSELLATED_DATA +{ + uint32_t NumPrimitives; + uint32_t NumDomainPoints; + + uint32_t* ppIndices[3]; + float* pDomainPointsU; + float* pDomainPointsV; + // For Tri: pDomainPointsW[i] = 1.0f - pDomainPointsU[i] - pDomainPointsV[i] +}; + +/// Perform Tessellation +void SWR_API TSTessellate( + HANDLE tsCtx, ///< [IN] Tessellation Context + const SWR_TESSELLATION_FACTORS& tsTessFactors, ///< [IN] Tessellation Factors + SWR_TS_TESSELLATED_DATA& tsTessellatedData); ///< [OUT] Tessellated Data + + + +/// @TODO - Implement OSS tessellator + +INLINE HANDLE SWR_API TSInitCtx( + SWR_TS_DOMAIN tsDomain, + SWR_TS_PARTITIONING tsPartitioning, + SWR_TS_OUTPUT_TOPOLOGY tsOutputTopology, + void* pContextMem, + size_t& memSize) +{ + SWR_ASSERT(0, "%s: Not Implemented", __FUNCTION__); + return NULL; +} + + +INLINE void SWR_API TSDestroyCtx(HANDLE tsCtx) +{ + SWR_ASSERT(0, "%s: Not Implemented", __FUNCTION__); +} + + +INLINE void SWR_API TSTessellate( + HANDLE tsCtx, + const SWR_TESSELLATION_FACTORS& tsTessFactors, + SWR_TS_TESSELLATED_DATA& tsTessellatedData) +{ + SWR_ASSERT(0, "%s: Not Implemented", __FUNCTION__); +} + diff --git a/src/gallium/drivers/swr/rasterizer/core/threads.cpp b/src/gallium/drivers/swr/rasterizer/core/threads.cpp new file mode 100644 index 00000000000..24c5588bfec --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/threads.cpp @@ -0,0 +1,962 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +****************************************************************************/ + +#include +#include +#include +#include +#include +#include +#include +#include +#include + +#if defined(__linux__) || defined(__gnu_linux__) +#include +#include +#include +#endif + +#include "common/os.h" +#include "context.h" +#include "frontend.h" +#include "backend.h" +#include "rasterizer.h" +#include "rdtsc_core.h" +#include "tilemgr.h" +#include "core/multisample.h" + + + + +// ThreadId +struct Core +{ + uint32_t procGroup = 0; + std::vector threadIds; +}; + +struct NumaNode +{ + std::vector cores; +}; + +typedef std::vector CPUNumaNodes; + +void CalculateProcessorTopology(CPUNumaNodes& out_nodes, uint32_t& out_numThreadsPerProcGroup) +{ + out_nodes.clear(); + out_numThreadsPerProcGroup = 0; + +#if defined(_WIN32) + + SYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX buffer[KNOB_MAX_NUM_THREADS]; + DWORD bufSize = sizeof(buffer); + + BOOL ret = GetLogicalProcessorInformationEx(RelationProcessorCore, buffer, &bufSize); + SWR_ASSERT(ret != FALSE, "Failed to get Processor Topology Information"); + + uint32_t count = bufSize / buffer->Size; + PSYSTEM_LOGICAL_PROCESSOR_INFORMATION_EX pBuffer = buffer; + + for (uint32_t i = 0; i < count; ++i) + { + SWR_ASSERT(pBuffer->Relationship == RelationProcessorCore); + for (uint32_t g = 0; g < pBuffer->Processor.GroupCount; ++g) + { + auto& gmask = pBuffer->Processor.GroupMask[g]; + uint32_t threadId = 0; + uint32_t procGroup = gmask.Group; + + Core* pCore = nullptr; + + uint32_t numThreads = (uint32_t)_mm_popcount_sizeT(gmask.Mask); + + while (BitScanForwardSizeT((unsigned long*)&threadId, gmask.Mask)) + { + // clear mask + gmask.Mask &= ~(KAFFINITY(1) << threadId); + + // Find Numa Node + PROCESSOR_NUMBER procNum = {}; + procNum.Group = WORD(procGroup); + procNum.Number = UCHAR(threadId); + + uint32_t numaId = 0; + ret = GetNumaProcessorNodeEx(&procNum, (PUSHORT)&numaId); + SWR_ASSERT(ret); + + // Store data + if (out_nodes.size() <= numaId) out_nodes.resize(numaId + 1); + auto& numaNode = out_nodes[numaId]; + + uint32_t coreId = 0; + + if (nullptr == pCore) + { + numaNode.cores.push_back(Core()); + pCore = &numaNode.cores.back(); + pCore->procGroup = procGroup; +#if !defined(_WIN64) + coreId = (uint32_t)numaNode.cores.size(); + if ((coreId * numThreads) >= 32) + { + // Windows doesn't return threadIds >= 32 for a processor group correctly + // when running a 32-bit application. + // Just save -1 as the threadId + threadId = uint32_t(-1); + } +#endif + } + pCore->threadIds.push_back(threadId); + if (procGroup == 0) + { + out_numThreadsPerProcGroup++; + } + } + } + pBuffer = PtrAdd(pBuffer, pBuffer->Size); + } + + +#elif defined(__linux__) || defined (__gnu_linux__) + + // Parse /proc/cpuinfo to get full topology + std::ifstream input("/proc/cpuinfo"); + std::string line; + char* c; + uint32_t threadId = uint32_t(-1); + uint32_t coreId = uint32_t(-1); + uint32_t numaId = uint32_t(-1); + + while (std::getline(input, line)) + { + if (line.find("processor") != std::string::npos) + { + if (threadId != uint32_t(-1)) + { + // Save information. + if (out_nodes.size() <= numaId) out_nodes.resize(numaId + 1); + auto& numaNode = out_nodes[numaId]; + if (numaNode.cores.size() <= coreId) numaNode.cores.resize(coreId + 1); + auto& core = numaNode.cores[coreId]; + + core.procGroup = coreId; + core.threadIds.push_back(threadId); + + out_numThreadsPerProcGroup++; + } + + auto data_start = line.find(": ") + 2; + threadId = std::strtoul(&line.c_str()[data_start], &c, 10); + continue; + } + if (line.find("core id") != std::string::npos) + { + auto data_start = line.find(": ") + 2; + coreId = std::strtoul(&line.c_str()[data_start], &c, 10); + continue; + } + if (line.find("physical id") != std::string::npos) + { + auto data_start = line.find(": ") + 2; + numaId = std::strtoul(&line.c_str()[data_start], &c, 10); + continue; + } + } + + if (threadId != uint32_t(-1)) + { + // Save information. + if (out_nodes.size() <= numaId) out_nodes.resize(numaId + 1); + auto& numaNode = out_nodes[numaId]; + if (numaNode.cores.size() <= coreId) numaNode.cores.resize(coreId + 1); + auto& core = numaNode.cores[coreId]; + + core.procGroup = coreId; + core.threadIds.push_back(threadId); + out_numThreadsPerProcGroup++; + } + + for (uint32_t node = 0; node < out_nodes.size(); node++) { + auto& numaNode = out_nodes[node]; + auto it = numaNode.cores.begin(); + for ( ; it != numaNode.cores.end(); ) { + if (it->threadIds.size() == 0) + numaNode.cores.erase(it); + else + ++it; + } + } + +#else + +#error Unsupported platform + +#endif +} + + +void bindThread(uint32_t threadId, uint32_t procGroupId = 0, bool bindProcGroup=false) +{ + // Only bind threads when MAX_WORKER_THREADS isn't set. + if (KNOB_MAX_WORKER_THREADS && bindProcGroup == false) + { + return; + } + +#if defined(_WIN32) + { + GROUP_AFFINITY affinity = {}; + affinity.Group = procGroupId; + +#if !defined(_WIN64) + if (threadId >= 32) + { + // In a 32-bit process on Windows it is impossible to bind + // to logical processors 32-63 within a processor group. + // In this case set the mask to 0 and let the system assign + // the processor. Hopefully it will make smart choices. + affinity.Mask = 0; + } + else +#endif + { + // If KNOB_MAX_WORKER_THREADS is set, only bind to the proc group, + // Not the individual HW thread. + if (!KNOB_MAX_WORKER_THREADS) + { + affinity.Mask = KAFFINITY(1) << threadId; + } + } + + SetThreadGroupAffinity(GetCurrentThread(), &affinity, nullptr); + } +#else + cpu_set_t cpuset; + pthread_t thread = pthread_self(); + CPU_ZERO(&cpuset); + CPU_SET(threadId, &cpuset); + + pthread_setaffinity_np(thread, sizeof(cpu_set_t), &cpuset); +#endif +} + +INLINE +uint64_t GetEnqueuedDraw(SWR_CONTEXT *pContext) +{ + //uint64_t result = _InterlockedCompareExchange64((volatile __int64*)&pContext->DrawEnqueued, 0, 0); + //return result; + return pContext->DrawEnqueued; +} + +INLINE +DRAW_CONTEXT *GetDC(SWR_CONTEXT *pContext, uint64_t drawId) +{ + return &pContext->dcRing[(drawId-1) % KNOB_MAX_DRAWS_IN_FLIGHT]; +} + +// returns true if dependency not met +INLINE +bool CheckDependency(SWR_CONTEXT *pContext, DRAW_CONTEXT *pDC, uint64_t lastRetiredDraw) +{ + return (pDC->dependency > lastRetiredDraw); +} + +void ClearColorHotTile(const HOTTILE* pHotTile) // clear a macro tile from float4 clear data. +{ + // Load clear color into SIMD register... + float *pClearData = (float*)(pHotTile->clearData); + simdscalar valR = _simd_broadcast_ss(&pClearData[0]); + simdscalar valG = _simd_broadcast_ss(&pClearData[1]); + simdscalar valB = _simd_broadcast_ss(&pClearData[2]); + simdscalar valA = _simd_broadcast_ss(&pClearData[3]); + + float *pfBuf = (float*)pHotTile->pBuffer; + uint32_t numSamples = pHotTile->numSamples; + + for (uint32_t row = 0; row < KNOB_MACROTILE_Y_DIM; row += KNOB_TILE_Y_DIM) + { + for (uint32_t col = 0; col < KNOB_MACROTILE_X_DIM; col += KNOB_TILE_X_DIM) + { + for (uint32_t si = 0; si < (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * numSamples); si += SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM) //SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM); si++) + { + _simd_store_ps(pfBuf, valR); + pfBuf += KNOB_SIMD_WIDTH; + _simd_store_ps(pfBuf, valG); + pfBuf += KNOB_SIMD_WIDTH; + _simd_store_ps(pfBuf, valB); + pfBuf += KNOB_SIMD_WIDTH; + _simd_store_ps(pfBuf, valA); + pfBuf += KNOB_SIMD_WIDTH; + } + } + } +} + +void ClearDepthHotTile(const HOTTILE* pHotTile) // clear a macro tile from float4 clear data. +{ + // Load clear color into SIMD register... + float *pClearData = (float*)(pHotTile->clearData); + simdscalar valZ = _simd_broadcast_ss(&pClearData[0]); + + float *pfBuf = (float*)pHotTile->pBuffer; + uint32_t numSamples = pHotTile->numSamples; + + for (uint32_t row = 0; row < KNOB_MACROTILE_Y_DIM; row += KNOB_TILE_Y_DIM) + { + for (uint32_t col = 0; col < KNOB_MACROTILE_X_DIM; col += KNOB_TILE_X_DIM) + { + for (uint32_t si = 0; si < (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * numSamples); si += SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM) + { + _simd_store_ps(pfBuf, valZ); + pfBuf += KNOB_SIMD_WIDTH; + } + } + } +} + +void ClearStencilHotTile(const HOTTILE* pHotTile) +{ + // convert from F32 to U8. + uint8_t clearVal = (uint8_t)(pHotTile->clearData[0]); + //broadcast 32x into __m256i... + simdscalari valS = _simd_set1_epi8(clearVal); + + simdscalari* pBuf = (simdscalari*)pHotTile->pBuffer; + uint32_t numSamples = pHotTile->numSamples; + + for (uint32_t row = 0; row < KNOB_MACROTILE_Y_DIM; row += KNOB_TILE_Y_DIM) + { + for (uint32_t col = 0; col < KNOB_MACROTILE_X_DIM; col += KNOB_TILE_X_DIM) + { + // We're putting 4 pixels in each of the 32-bit slots, so increment 4 times as quickly. + for (uint32_t si = 0; si < (KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * numSamples); si += SIMD_TILE_X_DIM * SIMD_TILE_Y_DIM * 4) + { + _simd_store_si(pBuf, valS); + pBuf += 1; + } + } + } +} + +// for draw calls, we initialize the active hot tiles and perform deferred +// load on them if tile is in invalid state. we do this in the outer thread loop instead of inside +// the draw routine itself mainly for performance, to avoid unnecessary setup +// every triangle +// @todo support deferred clear +INLINE +void InitializeHotTiles(SWR_CONTEXT* pContext, DRAW_CONTEXT* pDC, uint32_t macroID, const TRIANGLE_WORK_DESC* pWork) +{ + const API_STATE& state = GetApiState(pDC); + HotTileMgr *pHotTileMgr = pContext->pHotTileMgr; + + uint32_t x, y; + MacroTileMgr::getTileIndices(macroID, x, y); + x *= KNOB_MACROTILE_X_DIM; + y *= KNOB_MACROTILE_Y_DIM; + + uint32_t numSamples = GetNumSamples(state.rastState.sampleCount); + + // check RT if enabled + unsigned long rtSlot = 0; + uint32_t colorHottileEnableMask = state.colorHottileEnable; + while(_BitScanForward(&rtSlot, colorHottileEnableMask)) + { + HOTTILE* pHotTile = pHotTileMgr->GetHotTile(pContext, pDC, macroID, (SWR_RENDERTARGET_ATTACHMENT)(SWR_ATTACHMENT_COLOR0 + rtSlot), true, numSamples); + + if (pHotTile->state == HOTTILE_INVALID) + { + RDTSC_START(BELoadTiles); + // invalid hottile before draw requires a load from surface before we can draw to it + pContext->pfnLoadTile(GetPrivateState(pDC), KNOB_COLOR_HOT_TILE_FORMAT, (SWR_RENDERTARGET_ATTACHMENT)(SWR_ATTACHMENT_COLOR0 + rtSlot), x, y, pHotTile->renderTargetArrayIndex, pHotTile->pBuffer); + pHotTile->state = HOTTILE_DIRTY; + RDTSC_STOP(BELoadTiles, 0, 0); + } + else if (pHotTile->state == HOTTILE_CLEAR) + { + RDTSC_START(BELoadTiles); + // Clear the tile. + ClearColorHotTile(pHotTile); + pHotTile->state = HOTTILE_DIRTY; + RDTSC_STOP(BELoadTiles, 0, 0); + } + colorHottileEnableMask &= ~(1 << rtSlot); + } + + // check depth if enabled + if (state.depthHottileEnable) + { + HOTTILE* pHotTile = pHotTileMgr->GetHotTile(pContext, pDC, macroID, SWR_ATTACHMENT_DEPTH, true, numSamples); + if (pHotTile->state == HOTTILE_INVALID) + { + RDTSC_START(BELoadTiles); + // invalid hottile before draw requires a load from surface before we can draw to it + pContext->pfnLoadTile(GetPrivateState(pDC), KNOB_DEPTH_HOT_TILE_FORMAT, SWR_ATTACHMENT_DEPTH, x, y, pHotTile->renderTargetArrayIndex, pHotTile->pBuffer); + pHotTile->state = HOTTILE_DIRTY; + RDTSC_STOP(BELoadTiles, 0, 0); + } + else if (pHotTile->state == HOTTILE_CLEAR) + { + RDTSC_START(BELoadTiles); + // Clear the tile. + ClearDepthHotTile(pHotTile); + pHotTile->state = HOTTILE_DIRTY; + RDTSC_STOP(BELoadTiles, 0, 0); + } + } + + // check stencil if enabled + if (state.stencilHottileEnable) + { + HOTTILE* pHotTile = pHotTileMgr->GetHotTile(pContext, pDC, macroID, SWR_ATTACHMENT_STENCIL, true, numSamples); + if (pHotTile->state == HOTTILE_INVALID) + { + RDTSC_START(BELoadTiles); + // invalid hottile before draw requires a load from surface before we can draw to it + pContext->pfnLoadTile(GetPrivateState(pDC), KNOB_STENCIL_HOT_TILE_FORMAT, SWR_ATTACHMENT_STENCIL, x, y, pHotTile->renderTargetArrayIndex, pHotTile->pBuffer); + pHotTile->state = HOTTILE_DIRTY; + RDTSC_STOP(BELoadTiles, 0, 0); + } + else if (pHotTile->state == HOTTILE_CLEAR) + { + RDTSC_START(BELoadTiles); + // Clear the tile. + ClearStencilHotTile(pHotTile); + pHotTile->state = HOTTILE_DIRTY; + RDTSC_STOP(BELoadTiles, 0, 0); + } + } +} + +INLINE bool FindFirstIncompleteDraw(SWR_CONTEXT* pContext, uint64_t& curDrawBE) +{ + // increment our current draw id to the first incomplete draw + uint64_t drawEnqueued = GetEnqueuedDraw(pContext); + while (curDrawBE < drawEnqueued) + { + DRAW_CONTEXT *pDC = &pContext->dcRing[curDrawBE % KNOB_MAX_DRAWS_IN_FLIGHT]; + + // If its not compute and FE is not done then break out of loop. + if (!pDC->doneFE && !pDC->isCompute) break; + + bool isWorkComplete = (pDC->isCompute) ? + pDC->pDispatch->isWorkComplete() : pDC->pTileMgr->isWorkComplete(); + + if (isWorkComplete) + { + curDrawBE++; + InterlockedIncrement(&pDC->threadsDoneBE); + } + else + { + break; + } + } + + // If there are no more incomplete draws then return false. + return (curDrawBE >= drawEnqueued) ? false : true; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief If there is any BE work then go work on it. +/// @param pContext - pointer to SWR context. +/// @param workerId - The unique worker ID that is assigned to this thread. +/// @param curDrawBE - This tracks the draw contexts that this thread has processed. Each worker thread +/// has its own curDrawBE counter and this ensures that each worker processes all the +/// draws in order. +/// @param lockedTiles - This is the set of tiles locked by other threads. Each thread maintains its +/// own set and each time it fails to lock a macrotile, because its already locked, +/// then it will add that tile to the lockedTiles set. As a worker begins to work +/// on future draws the lockedTiles ensure that it doesn't work on tiles that may +/// still have work pending in a previous draw. Additionally, the lockedTiles is +/// hueristic that can steer a worker back to the same macrotile that it had been +/// working on in a previous draw. +void WorkOnFifoBE( + SWR_CONTEXT *pContext, + uint32_t workerId, + uint64_t &curDrawBE, + std::unordered_set& lockedTiles) +{ + // Find the first incomplete draw that has pending work. If no such draw is found then + // return. FindFirstIncompleteDraw is responsible for incrementing the curDrawBE. + if (FindFirstIncompleteDraw(pContext, curDrawBE) == false) + { + return; + } + + uint64_t lastRetiredDraw = pContext->dcRing[curDrawBE % KNOB_MAX_DRAWS_IN_FLIGHT].drawId - 1; + + // Reset our history for locked tiles. We'll have to re-learn which tiles are locked. + lockedTiles.clear(); + + // Try to work on each draw in order of the available draws in flight. + // 1. If we're on curDrawBE, we can work on any macrotile that is available. + // 2. If we're trying to work on draws after curDrawBE, we are restricted to + // working on those macrotiles that are known to be complete in the prior draw to + // maintain order. The locked tiles provides the history to ensures this. + for (uint64_t i = curDrawBE; i < GetEnqueuedDraw(pContext); ++i) + { + DRAW_CONTEXT *pDC = &pContext->dcRing[i % KNOB_MAX_DRAWS_IN_FLIGHT]; + + if (pDC->isCompute) return; // We don't look at compute work. + + // First wait for FE to be finished with this draw. This keeps threading model simple + // but if there are lots of bubbles between draws then serializing FE and BE may + // need to be revisited. + if (!pDC->doneFE) return; + + // If this draw is dependent on a previous draw then we need to bail. + if (CheckDependency(pContext, pDC, lastRetiredDraw)) + { + return; + } + + // Grab the list of all dirty macrotiles. A tile is dirty if it has work queued to it. + std::vector ¯oTiles = pDC->pTileMgr->getDirtyTiles(); + + for (uint32_t tileID : macroTiles) + { + MacroTileQueue &tile = pDC->pTileMgr->getMacroTileQueue(tileID); + + // can only work on this draw if it's not in use by other threads + if (lockedTiles.find(tileID) == lockedTiles.end()) + { + if (tile.getNumQueued()) + { + if (tile.tryLock()) + { + BE_WORK *pWork; + + RDTSC_START(WorkerFoundWork); + + uint32_t numWorkItems = tile.getNumQueued(); + + if (numWorkItems != 0) + { + pWork = tile.peek(); + SWR_ASSERT(pWork); + if (pWork->type == DRAW) + { + InitializeHotTiles(pContext, pDC, tileID, (const TRIANGLE_WORK_DESC*)&pWork->desc); + } + } + + while ((pWork = tile.peek()) != nullptr) + { + pWork->pfnWork(pDC, workerId, tileID, &pWork->desc); + tile.dequeue(); + } + RDTSC_STOP(WorkerFoundWork, numWorkItems, pDC->drawId); + + _ReadWriteBarrier(); + + pDC->pTileMgr->markTileComplete(tileID); + + // Optimization: If the draw is complete and we're the last one to have worked on it then + // we can reset the locked list as we know that all previous draws before the next are guaranteed to be complete. + if ((curDrawBE == i) && pDC->pTileMgr->isWorkComplete()) + { + // We can increment the current BE and safely move to next draw since we know this draw is complete. + curDrawBE++; + InterlockedIncrement(&pDC->threadsDoneBE); + + lastRetiredDraw++; + + lockedTiles.clear(); + break; + } + } + else + { + // This tile is already locked. So let's add it to our locked tiles set. This way we don't try locking this one again. + lockedTiles.insert(tileID); + } + } + } + } + } +} + +void WorkOnFifoFE(SWR_CONTEXT *pContext, uint32_t workerId, uint64_t &curDrawFE, UCHAR numaNode) +{ + // Try to grab the next DC from the ring + uint64_t drawEnqueued = GetEnqueuedDraw(pContext); + while (curDrawFE < drawEnqueued) + { + uint32_t dcSlot = curDrawFE % KNOB_MAX_DRAWS_IN_FLIGHT; + DRAW_CONTEXT *pDC = &pContext->dcRing[dcSlot]; + if (pDC->isCompute || pDC->doneFE || pDC->FeLock) + { + curDrawFE++; + InterlockedIncrement(&pDC->threadsDoneFE); + } + else + { + break; + } + } + + uint64_t curDraw = curDrawFE; + while (curDraw < drawEnqueued) + { + uint32_t dcSlot = curDraw % KNOB_MAX_DRAWS_IN_FLIGHT; + DRAW_CONTEXT *pDC = &pContext->dcRing[dcSlot]; + + if (!pDC->isCompute && !pDC->FeLock) + { + uint32_t initial = InterlockedCompareExchange((volatile uint32_t*)&pDC->FeLock, 1, 0); + if (initial == 0) + { + // successfully grabbed the DC, now run the FE + pDC->FeWork.pfnWork(pContext, pDC, workerId, &pDC->FeWork.desc); + + _ReadWriteBarrier(); + pDC->doneFE = true; + } + } + curDraw++; + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief If there is any compute work then go work on it. +/// @param pContext - pointer to SWR context. +/// @param workerId - The unique worker ID that is assigned to this thread. +/// @param curDrawBE - This tracks the draw contexts that this thread has processed. Each worker thread +/// has its own curDrawBE counter and this ensures that each worker processes all the +/// draws in order. +void WorkOnCompute( + SWR_CONTEXT *pContext, + uint32_t workerId, + uint64_t& curDrawBE) +{ + if (FindFirstIncompleteDraw(pContext, curDrawBE) == false) + { + return; + } + + uint64_t lastRetiredDraw = pContext->dcRing[curDrawBE % KNOB_MAX_DRAWS_IN_FLIGHT].drawId - 1; + + DRAW_CONTEXT *pDC = &pContext->dcRing[curDrawBE % KNOB_MAX_DRAWS_IN_FLIGHT]; + if (pDC->isCompute == false) return; + + // check dependencies + if (CheckDependency(pContext, pDC, lastRetiredDraw)) + { + return; + } + + SWR_ASSERT(pDC->pDispatch != nullptr); + DispatchQueue& queue = *pDC->pDispatch; + + // Is there any work remaining? + if (queue.getNumQueued() > 0) + { + bool lastToComplete = false; + + uint32_t threadGroupId = 0; + while (queue.getWork(threadGroupId)) + { + ProcessComputeBE(pDC, workerId, threadGroupId); + + lastToComplete = queue.finishedWork(); + } + + _ReadWriteBarrier(); + + if (lastToComplete) + { + SWR_ASSERT(queue.isWorkComplete() == true); + pDC->doneCompute = true; + } + } +} + +DWORD workerThreadMain(LPVOID pData) +{ + THREAD_DATA *pThreadData = (THREAD_DATA*)pData; + SWR_CONTEXT *pContext = pThreadData->pContext; + uint32_t threadId = pThreadData->threadId; + uint32_t workerId = pThreadData->workerId; + + bindThread(threadId, pThreadData->procGroupId, pThreadData->forceBindProcGroup); + + RDTSC_INIT(threadId); + + int numaNode = (int)pThreadData->numaId; + + // flush denormals to 0 + _mm_setcsr(_mm_getcsr() | _MM_FLUSH_ZERO_ON | _MM_DENORMALS_ZERO_ON); + + // Track tiles locked by other threads. If we try to lock a macrotile and find its already + // locked then we'll add it to this list so that we don't try and lock it again. + std::unordered_set lockedTiles; + + // each worker has the ability to work on any of the queued draws as long as certain + // conditions are met. the data associated + // with a draw is guaranteed to be active as long as a worker hasn't signaled that he + // has moved on to the next draw when he determines there is no more work to do. The api + // thread will not increment the head of the dc ring until all workers have moved past the + // current head. + // the logic to determine what to work on is: + // 1- try to work on the FE any draw that is queued. For now there are no dependencies + // on the FE work, so any worker can grab any FE and process in parallel. Eventually + // we'll need dependency tracking to force serialization on FEs. The worker will try + // to pick an FE by atomically incrementing a counter in the swr context. he'll keep + // trying until he reaches the tail. + // 2- BE work must be done in strict order. we accomplish this today by pulling work off + // the oldest draw (ie the head) of the dcRing. the worker can determine if there is + // any work left by comparing the total # of binned work items and the total # of completed + // work items. If they are equal, then there is no more work to do for this draw, and + // the worker can safely increment its oldestDraw counter and move on to the next draw. + std::unique_lock lock(pContext->WaitLock, std::defer_lock); + + auto threadHasWork = [&](uint64_t curDraw) { return curDraw != pContext->DrawEnqueued; }; + + uint64_t curDrawBE = 1; + uint64_t curDrawFE = 1; + + while (pContext->threadPool.inThreadShutdown == false) + { + uint32_t loop = 0; + while (loop++ < KNOB_WORKER_SPIN_LOOP_COUNT && !threadHasWork(curDrawBE)) + { + _mm_pause(); + } + + if (!threadHasWork(curDrawBE)) + { + lock.lock(); + + // check for thread idle condition again under lock + if (threadHasWork(curDrawBE)) + { + lock.unlock(); + continue; + } + + if (pContext->threadPool.inThreadShutdown) + { + lock.unlock(); + break; + } + + RDTSC_START(WorkerWaitForThreadEvent); + + pContext->FifosNotEmpty.wait(lock); + lock.unlock(); + + RDTSC_STOP(WorkerWaitForThreadEvent, 0, 0); + + if (pContext->threadPool.inThreadShutdown) + { + break; + } + } + + RDTSC_START(WorkerWorkOnFifoBE); + WorkOnFifoBE(pContext, workerId, curDrawBE, lockedTiles); + RDTSC_STOP(WorkerWorkOnFifoBE, 0, 0); + + WorkOnCompute(pContext, workerId, curDrawBE); + + WorkOnFifoFE(pContext, workerId, curDrawFE, numaNode); + } + + return 0; +} + +DWORD workerThreadInit(LPVOID pData) +{ +#if defined(_WIN32) + __try +#endif // _WIN32 + { + return workerThreadMain(pData); + } + +#if defined(_WIN32) + __except(EXCEPTION_CONTINUE_SEARCH) + { + } + +#endif // _WIN32 + + return 1; +} + +void CreateThreadPool(SWR_CONTEXT *pContext, THREAD_POOL *pPool) +{ + bindThread(0); + + CPUNumaNodes nodes; + uint32_t numThreadsPerProcGroup = 0; + CalculateProcessorTopology(nodes, numThreadsPerProcGroup); + + uint32_t numHWNodes = (uint32_t)nodes.size(); + uint32_t numHWCoresPerNode = (uint32_t)nodes[0].cores.size(); + uint32_t numHWHyperThreads = (uint32_t)nodes[0].cores[0].threadIds.size(); + + uint32_t numNodes = numHWNodes; + uint32_t numCoresPerNode = numHWCoresPerNode; + uint32_t numHyperThreads = numHWHyperThreads; + + if (KNOB_MAX_NUMA_NODES) + { + numNodes = std::min(numNodes, KNOB_MAX_NUMA_NODES); + } + + if (KNOB_MAX_CORES_PER_NUMA_NODE) + { + numCoresPerNode = std::min(numCoresPerNode, KNOB_MAX_CORES_PER_NUMA_NODE); + } + + if (KNOB_MAX_THREADS_PER_CORE) + { + numHyperThreads = std::min(numHyperThreads, KNOB_MAX_THREADS_PER_CORE); + } + + // Calculate numThreads + uint32_t numThreads = numNodes * numCoresPerNode * numHyperThreads; + + if (KNOB_MAX_WORKER_THREADS) + { + uint32_t maxHWThreads = numHWNodes * numHWCoresPerNode * numHWHyperThreads; + numThreads = std::min(KNOB_MAX_WORKER_THREADS, maxHWThreads); + } + + if (numThreads > KNOB_MAX_NUM_THREADS) + { + printf("WARNING: system thread count %u exceeds max %u, " + "performance will be degraded\n", + numThreads, KNOB_MAX_NUM_THREADS); + } + + if (numThreads == 1) + { + // If only 1 worker thread, try to move it to an available + // HW thread. If that fails, use the API thread. + if (numCoresPerNode < numHWCoresPerNode) + { + numCoresPerNode++; + } + else if (numHyperThreads < numHWHyperThreads) + { + numHyperThreads++; + } + else if (numNodes < numHWNodes) + { + numNodes++; + } + else + { + pPool->numThreads = 0; + SET_KNOB(SINGLE_THREADED, true); + return; + } + } + else + { + // Save a HW thread for the API thread. + numThreads--; + } + + pPool->numThreads = numThreads; + pContext->NumWorkerThreads = pPool->numThreads; + + pPool->inThreadShutdown = false; + pPool->pThreadData = (THREAD_DATA *)malloc(pPool->numThreads * sizeof(THREAD_DATA)); + + if (KNOB_MAX_WORKER_THREADS) + { + bool bForceBindProcGroup = (numThreads > numThreadsPerProcGroup); + uint32_t numProcGroups = (numThreads + numThreadsPerProcGroup - 1) / numThreadsPerProcGroup; + // When MAX_WORKER_THREADS is set we don't bother to bind to specific HW threads + // But Windows will still require binding to specific process groups + for (uint32_t workerId = 0; workerId < numThreads; ++workerId) + { + pPool->pThreadData[workerId].workerId = workerId; + pPool->pThreadData[workerId].procGroupId = workerId % numProcGroups; + pPool->pThreadData[workerId].threadId = 0; + pPool->pThreadData[workerId].numaId = 0; + pPool->pThreadData[workerId].pContext = pContext; + pPool->pThreadData[workerId].forceBindProcGroup = bForceBindProcGroup; + pPool->threads[workerId] = new std::thread(workerThreadInit, &pPool->pThreadData[workerId]); + } + } + else + { + uint32_t workerId = 0; + for (uint32_t n = 0; n < numNodes; ++n) + { + auto& node = nodes[n]; + + uint32_t numCores = numCoresPerNode; + for (uint32_t c = 0; c < numCores; ++c) + { + auto& core = node.cores[c]; + for (uint32_t t = 0; t < numHyperThreads; ++t) + { + if (c == 0 && n == 0 && t == 0) + { + // Skip core 0, thread0 on node 0 to reserve for API thread + continue; + } + + pPool->pThreadData[workerId].workerId = workerId; + pPool->pThreadData[workerId].procGroupId = core.procGroup; + pPool->pThreadData[workerId].threadId = core.threadIds[t]; + pPool->pThreadData[workerId].numaId = n; + pPool->pThreadData[workerId].pContext = pContext; + pPool->threads[workerId] = new std::thread(workerThreadInit, &pPool->pThreadData[workerId]); + + ++workerId; + } + } + } + } +} + +void DestroyThreadPool(SWR_CONTEXT *pContext, THREAD_POOL *pPool) +{ + if (!KNOB_SINGLE_THREADED) + { + // Inform threads to finish up + std::unique_lock lock(pContext->WaitLock); + pPool->inThreadShutdown = true; + _mm_mfence(); + pContext->FifosNotEmpty.notify_all(); + lock.unlock(); + + // Wait for threads to finish and destroy them + for (uint32_t t = 0; t < pPool->numThreads; ++t) + { + pPool->threads[t]->join(); + delete(pPool->threads[t]); + } + + // Clean up data used by threads + free(pPool->pThreadData); + } +} diff --git a/src/gallium/drivers/swr/rasterizer/core/threads.h b/src/gallium/drivers/swr/rasterizer/core/threads.h new file mode 100644 index 00000000000..0fa7196f5ac --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/threads.h @@ -0,0 +1,63 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file threads.h +* +* @brief Definitions for SWR threading model. +* +******************************************************************************/ +#pragma once + +#include "knobs.h" + +#include +#include +typedef std::thread* THREAD_PTR; + +struct SWR_CONTEXT; + +struct THREAD_DATA +{ + uint32_t procGroupId; // Will always be 0 for non-Windows OS + uint32_t threadId; // within the procGroup for Windows + uint32_t numaId; // NUMA node id + uint32_t workerId; + SWR_CONTEXT *pContext; + bool forceBindProcGroup; // Only useful when KNOB_MAX_WORKER_THREADS is set. +}; + + +struct THREAD_POOL +{ + THREAD_PTR threads[KNOB_MAX_NUM_THREADS]; + uint32_t numThreads; + volatile bool inThreadShutdown; + THREAD_DATA *pThreadData; +}; + +void CreateThreadPool(SWR_CONTEXT *pContext, THREAD_POOL *pPool); +void DestroyThreadPool(SWR_CONTEXT *pContext, THREAD_POOL *pPool); + +// Expose FE and BE worker functions to the API thread if single threaded +void WorkOnFifoFE(SWR_CONTEXT *pContext, uint32_t workerId, uint64_t &curDrawFE, UCHAR numaNode); +void WorkOnFifoBE(SWR_CONTEXT *pContext, uint32_t workerId, uint64_t &curDrawBE, std::unordered_set &usedTiles); +void WorkOnCompute(SWR_CONTEXT *pContext, uint32_t workerId, uint64_t &curDrawBE); diff --git a/src/gallium/drivers/swr/rasterizer/core/tilemgr.cpp b/src/gallium/drivers/swr/rasterizer/core/tilemgr.cpp new file mode 100644 index 00000000000..860393661e2 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/tilemgr.cpp @@ -0,0 +1,105 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file tilemgr.cpp +* +* @brief Implementation for Macro Tile Manager which provides the facilities +* for threads to work on an macro tile. +* +******************************************************************************/ +#include + +#include "fifo.hpp" +#include "tilemgr.h" + +#define TILE_ID(x,y) ((x << 16 | y)) + +// override new/delete for alignment +void *MacroTileMgr::operator new(size_t size) +{ + return _aligned_malloc(size, 64); +} + +void MacroTileMgr::operator delete(void *p) +{ + _aligned_free(p); +} + +void* DispatchQueue::operator new(size_t size) +{ + return _aligned_malloc(size, 64); +} + +void DispatchQueue::operator delete(void *p) +{ + _aligned_free(p); +} + +MacroTileMgr::MacroTileMgr(Arena& arena) : mArena(arena) +{ +} + +void MacroTileMgr::initialize() +{ + mWorkItemsProduced = 0; + mWorkItemsConsumed = 0; + + mDirtyTiles.clear(); +} + +void MacroTileMgr::enqueue(uint32_t x, uint32_t y, BE_WORK *pWork) +{ + // Should not enqueue more then what we have backing for in the hot tile manager. + SWR_ASSERT(x < KNOB_NUM_HOT_TILES_X); + SWR_ASSERT(y < KNOB_NUM_HOT_TILES_Y); + + uint32_t id = TILE_ID(x, y); + + MacroTileQueue &tile = mTiles[id]; + tile.mWorkItemsFE++; + + if (tile.mWorkItemsFE == 1) + { + tile.clear(mArena); + mDirtyTiles.push_back(id); + } + + mWorkItemsProduced++; + tile.enqueue_try_nosync(mArena, pWork); +} + +void MacroTileMgr::markTileComplete(uint32_t id) +{ + SWR_ASSERT(mTiles.find(id) != mTiles.end()); + MacroTileQueue &tile = mTiles[id]; + uint32_t numTiles = tile.mWorkItemsFE; + InterlockedExchangeAdd(&mWorkItemsConsumed, numTiles); + + _ReadWriteBarrier(); + tile.mWorkItemsBE += numTiles; + SWR_ASSERT(tile.mWorkItemsFE == tile.mWorkItemsBE); + + // clear out tile, but defer fifo clear until the next DC first queues to it. + // this prevents worker threads from constantly locking a completed macro tile + tile.mWorkItemsFE = 0; + tile.mWorkItemsBE = 0; +} diff --git a/src/gallium/drivers/swr/rasterizer/core/tilemgr.h b/src/gallium/drivers/swr/rasterizer/core/tilemgr.h new file mode 100644 index 00000000000..9137941bad4 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/tilemgr.h @@ -0,0 +1,390 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file tilemgr.h +* +* @brief Definitions for Macro Tile Manager which provides the facilities +* for threads to work on an macro tile. +* +******************************************************************************/ +#pragma once + +#include +#include +#include "common/formats.h" +#include "fifo.hpp" +#include "context.h" +#include "format_traits.h" + +////////////////////////////////////////////////////////////////////////// +/// MacroTile - work queue for a tile. +////////////////////////////////////////////////////////////////////////// +struct MacroTileQueue +{ + MacroTileQueue() { } + ~MacroTileQueue() { } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Returns number of work items queued for this tile. + uint32_t getNumQueued() + { + return mFifo.getNumQueued(); + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Attempt to lock the work fifo. If already locked then return false. + bool tryLock() + { + return mFifo.tryLock(); + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Clear fifo and unlock it. + void clear(Arena& arena) + { + mFifo.clear(arena); + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Peek at work sitting at the front of the fifo. + BE_WORK* peek() + { + return mFifo.peek(); + } + + bool enqueue_try_nosync(Arena& arena, const BE_WORK* entry) + { + return mFifo.enqueue_try_nosync(arena, entry); + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Move to next work item + void dequeue() + { + mFifo.dequeue_noinc(); + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Destroy fifo + void destroy() + { + mFifo.destroy(); + } + + ///@todo This will all be private. + uint32_t mWorkItemsFE = 0; + uint32_t mWorkItemsBE = 0; + +private: + QUEUE mFifo; +}; + +////////////////////////////////////////////////////////////////////////// +/// MacroTileMgr - Manages macrotiles for a draw. +////////////////////////////////////////////////////////////////////////// +class MacroTileMgr +{ +public: + MacroTileMgr(Arena& arena); + ~MacroTileMgr() + { + for (auto &tile : mTiles) + { + tile.second.destroy(); + } + } + + void initialize(); + INLINE std::vector& getDirtyTiles() { return mDirtyTiles; } + INLINE MacroTileQueue& getMacroTileQueue(uint32_t id) { return mTiles[id]; } + void markTileComplete(uint32_t id); + + INLINE bool isWorkComplete() + { + return mWorkItemsProduced == mWorkItemsConsumed; + } + + void enqueue(uint32_t x, uint32_t y, BE_WORK *pWork); + + static INLINE void getTileIndices(uint32_t tileID, uint32_t &x, uint32_t &y) + { + y = tileID & 0xffff; + x = (tileID >> 16) & 0xffff; + } + + void *operator new(size_t size); + void operator delete (void *p); + +private: + Arena& mArena; + SWR_FORMAT mFormat; + std::unordered_map mTiles; + + // Any tile that has work queued to it is a dirty tile. + std::vector mDirtyTiles; + + OSALIGNLINE(LONG) mWorkItemsProduced; + OSALIGNLINE(volatile LONG) mWorkItemsConsumed; +}; + +////////////////////////////////////////////////////////////////////////// +/// DispatchQueue - work queue for dispatch +////////////////////////////////////////////////////////////////////////// +class DispatchQueue +{ +public: + DispatchQueue() {} + + ////////////////////////////////////////////////////////////////////////// + /// @brief Setup the producer consumer counts. + void initialize(uint32_t totalTasks, void* pTaskData) + { + // The available and outstanding counts start with total tasks. + // At the start there are N tasks available and outstanding. + // When both the available and outstanding counts have reached 0 then all work has completed. + // When a worker starts on a threadgroup then it decrements the available count. + // When a worker completes a threadgroup then it decrements the outstanding count. + + mTasksAvailable = totalTasks; + mTasksOutstanding = totalTasks; + + mpTaskData = pTaskData; + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Returns number of tasks available for this dispatch. + uint32_t getNumQueued() + { + return (mTasksAvailable > 0) ? mTasksAvailable : 0; + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Atomically decrement the work available count. If the result + // is greater than 0 then we can on the associated thread group. + // Otherwise, there is no more work to do. + bool getWork(uint32_t& groupId) + { + LONG result = InterlockedDecrement(&mTasksAvailable); + + if (result >= 0) + { + groupId = result; + return true; + } + + return false; + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Atomically decrement the outstanding count. A worker is notifying + /// us that he just finished some work. Also, return true if we're + /// the last worker to complete this dispatch. + bool finishedWork() + { + LONG result = InterlockedDecrement(&mTasksOutstanding); + SWR_ASSERT(result >= 0, "Should never oversubscribe work"); + + return (result == 0) ? true : false; + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Work is complete once both the available/outstanding counts have reached 0. + bool isWorkComplete() + { + return ((mTasksAvailable <= 0) && + (mTasksOutstanding <= 0)); + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Return pointer to task data. + const void* GetTasksData() + { + return mpTaskData; + } + + void *operator new(size_t size); + void operator delete (void *p); + + void* mpTaskData; // The API thread will set this up and the callback task function will interpet this. + + OSALIGNLINE(volatile LONG) mTasksAvailable{ 0 }; + OSALIGNLINE(volatile LONG) mTasksOutstanding{ 0 }; +}; + + +enum HOTTILE_STATE +{ + HOTTILE_INVALID, // tile is in unitialized state and should be loaded with surface contents before rendering + HOTTILE_CLEAR, // tile should be cleared + HOTTILE_DIRTY, // tile has been rendered to + HOTTILE_RESOLVED, // tile has been stored to memory +}; + +struct HOTTILE +{ + BYTE *pBuffer; + HOTTILE_STATE state; + DWORD clearData[4]; // May need to change based on pfnClearTile implementation. Reorder for alignment? + uint32_t numSamples; + uint32_t renderTargetArrayIndex; // current render target array index loaded +}; + +union HotTileSet +{ + struct + { + HOTTILE Color[SWR_NUM_RENDERTARGETS]; + HOTTILE Depth; + HOTTILE Stencil; + }; + HOTTILE Attachment[SWR_NUM_ATTACHMENTS]; +}; + +class HotTileMgr +{ +public: + HotTileMgr() + { + memset(&mHotTiles[0][0], 0, sizeof(mHotTiles)); + + // cache hottile size + for (uint32_t i = SWR_ATTACHMENT_COLOR0; i <= SWR_ATTACHMENT_COLOR7; ++i) + { + mHotTileSize[i] = KNOB_MACROTILE_X_DIM * KNOB_MACROTILE_Y_DIM * FormatTraits::bpp / 8; + } + mHotTileSize[SWR_ATTACHMENT_DEPTH] = KNOB_MACROTILE_X_DIM * KNOB_MACROTILE_Y_DIM * FormatTraits::bpp / 8; + mHotTileSize[SWR_ATTACHMENT_STENCIL] = KNOB_MACROTILE_X_DIM * KNOB_MACROTILE_Y_DIM * FormatTraits::bpp / 8; + } + + ~HotTileMgr() + { + for (int x = 0; x < KNOB_NUM_HOT_TILES_X; ++x) + { + for (int y = 0; y < KNOB_NUM_HOT_TILES_Y; ++y) + { + for (int a = 0; a < SWR_NUM_ATTACHMENTS; ++a) + { + if (mHotTiles[x][y].Attachment[a].pBuffer != NULL) + { + _aligned_free(mHotTiles[x][y].Attachment[a].pBuffer); + mHotTiles[x][y].Attachment[a].pBuffer = NULL; + } + } + } + } + } + + HOTTILE *GetHotTile(SWR_CONTEXT* pContext, DRAW_CONTEXT* pDC, uint32_t macroID, SWR_RENDERTARGET_ATTACHMENT attachment, bool create, uint32_t numSamples = 1, + uint32_t renderTargetArrayIndex = 0) + { + uint32_t x, y; + MacroTileMgr::getTileIndices(macroID, x, y); + + assert(x < KNOB_NUM_HOT_TILES_X); + assert(y < KNOB_NUM_HOT_TILES_Y); + + HotTileSet &tile = mHotTiles[x][y]; + HOTTILE& hotTile = tile.Attachment[attachment]; + if (hotTile.pBuffer == NULL) + { + if (create) + { + uint32_t size = numSamples * mHotTileSize[attachment]; + hotTile.pBuffer = (BYTE*)_aligned_malloc(size, KNOB_SIMD_WIDTH * 4); + hotTile.state = HOTTILE_INVALID; + hotTile.numSamples = numSamples; + hotTile.renderTargetArrayIndex = renderTargetArrayIndex; + } + else + { + return NULL; + } + } + else + { + // free the old tile and create a new one with enough space to hold all samples + if (numSamples > hotTile.numSamples) + { + // tile should be either uninitialized or resolved if we're deleting and switching to a + // new sample count + assert((hotTile.state == HOTTILE_INVALID) || + (hotTile.state == HOTTILE_RESOLVED) || + (hotTile.state == HOTTILE_CLEAR)); + _aligned_free(hotTile.pBuffer); + + uint32_t size = numSamples * mHotTileSize[attachment]; + hotTile.pBuffer = (BYTE*)_aligned_malloc(size, KNOB_SIMD_WIDTH * 4); + hotTile.state = HOTTILE_INVALID; + hotTile.numSamples = numSamples; + } + + // if requested render target array index isn't currently loaded, need to store out the current hottile + // and load the requested array slice + if (renderTargetArrayIndex != hotTile.renderTargetArrayIndex) + { + SWR_FORMAT format; + switch (attachment) + { + case SWR_ATTACHMENT_COLOR0: + case SWR_ATTACHMENT_COLOR1: + case SWR_ATTACHMENT_COLOR2: + case SWR_ATTACHMENT_COLOR3: + case SWR_ATTACHMENT_COLOR4: + case SWR_ATTACHMENT_COLOR5: + case SWR_ATTACHMENT_COLOR6: + case SWR_ATTACHMENT_COLOR7: format = KNOB_COLOR_HOT_TILE_FORMAT; break; + case SWR_ATTACHMENT_DEPTH: format = KNOB_DEPTH_HOT_TILE_FORMAT; break; + case SWR_ATTACHMENT_STENCIL: format = KNOB_STENCIL_HOT_TILE_FORMAT; break; + default: SWR_ASSERT(false, "Unknown attachment: %d", attachment); format = KNOB_COLOR_HOT_TILE_FORMAT; break; + } + + if (hotTile.state == HOTTILE_DIRTY) + { + pContext->pfnStoreTile(GetPrivateState(pDC), format, attachment, + x * KNOB_MACROTILE_X_DIM, y * KNOB_MACROTILE_Y_DIM, hotTile.renderTargetArrayIndex, hotTile.pBuffer); + } + + pContext->pfnLoadTile(GetPrivateState(pDC), format, attachment, + x * KNOB_MACROTILE_X_DIM, y * KNOB_MACROTILE_Y_DIM, renderTargetArrayIndex, hotTile.pBuffer); + + hotTile.renderTargetArrayIndex = renderTargetArrayIndex; + hotTile.state = HOTTILE_DIRTY; + } + } + return &tile.Attachment[attachment]; + } + + HotTileSet &GetHotTile(uint32_t macroID) + { + uint32_t x, y; + MacroTileMgr::getTileIndices(macroID, x, y); + assert(x < KNOB_NUM_HOT_TILES_X); + assert(y < KNOB_NUM_HOT_TILES_Y); + + return mHotTiles[x][y]; + } + +private: + HotTileSet mHotTiles[KNOB_NUM_HOT_TILES_X][KNOB_NUM_HOT_TILES_Y]; + uint32_t mHotTileSize[SWR_NUM_ATTACHMENTS]; +}; + diff --git a/src/gallium/drivers/swr/rasterizer/core/utils.cpp b/src/gallium/drivers/swr/rasterizer/core/utils.cpp new file mode 100644 index 00000000000..f36452f2cec --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/utils.cpp @@ -0,0 +1,148 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file utils.cpp +* +* @brief Utilities used by SWR core. +* +******************************************************************************/ +#if defined(_WIN32) + +#include +#include +#include +#include + +using namespace Gdiplus; + +int GetEncoderClsid(const WCHAR* format, CLSID* pClsid) +{ + uint32_t num = 0; // number of image encoders + uint32_t size = 0; // size of the image encoder array in bytes + + ImageCodecInfo* pImageCodecInfo = nullptr; + + GetImageEncodersSize(&num, &size); + if(size == 0) + return -1; // Failure + + pImageCodecInfo = (ImageCodecInfo*)(malloc(size)); + if(pImageCodecInfo == nullptr) + return -1; // Failure + + GetImageEncoders(num, size, pImageCodecInfo); + + for(uint32_t j = 0; j < num; ++j) + { + if( wcscmp(pImageCodecInfo[j].MimeType, format) == 0 ) + { + *pClsid = pImageCodecInfo[j].Clsid; + free(pImageCodecInfo); + return j; // Success + } + } + + free(pImageCodecInfo); + return -1; // Failure +} + +void SaveImageToPNGFile( + const WCHAR *pFilename, + void *pBuffer, + uint32_t width, + uint32_t height) +{ + // dump pixels to a png + // Initialize GDI+. + GdiplusStartupInput gdiplusStartupInput; + ULONG_PTR gdiplusToken; + GdiplusStartup(&gdiplusToken, &gdiplusStartupInput, nullptr); + + Bitmap *bitmap = new Bitmap(width, height); + BYTE *pBytes = (BYTE*)pBuffer; + static const uint32_t bytesPerPixel = 4; + for (uint32_t y = 0; y < height; ++y) + for (uint32_t x = 0; x < width; ++x) + { + uint32_t pixel = *(uint32_t*)pBytes; + if (pixel == 0xcdcdcdcd) + { + pixel = 0xFFFF00FF; + } + else if (pixel == 0xdddddddd) + { + pixel = 0x80FF0000; + } + else + { + pixel |= 0xFF000000; + } + Color color(pixel); + bitmap->SetPixel(x, y, color); + pBytes += bytesPerPixel; + } + + // Save image. + CLSID pngClsid; + GetEncoderClsid(L"image/png", &pngClsid); + bitmap->Save(pFilename, &pngClsid, nullptr); + + delete bitmap; + + GdiplusShutdown(gdiplusToken); +} + +void OpenBitmapFromFile( + const WCHAR *pFilename, + void **pBuffer, + uint32_t *width, + uint32_t *height) +{ + GdiplusStartupInput gdiplusStartupInput; + ULONG_PTR gdiplusToken; + GdiplusStartup(&gdiplusToken, &gdiplusStartupInput, nullptr); + + Bitmap *bitmap = new Bitmap(pFilename); + + *width = bitmap->GetWidth(); + *height = bitmap->GetHeight(); + *pBuffer = new BYTE[*width * *height * 4]; // width * height * |RGBA| + + // The folder 'stb_image' contains a PNG open/close module which + // is far less painful than this is, yo. + Gdiplus::Color clr; + for (uint32_t y = 0, idx = 0; y < *height; ++y) + { + for (uint32_t x = 0; x < *width; ++x, idx += 4) + { + bitmap->GetPixel(x, *height - y - 1, &clr); + ((BYTE*)*pBuffer)[idx + 0] = clr.GetBlue(); + ((BYTE*)*pBuffer)[idx + 1] = clr.GetGreen(); + ((BYTE*)*pBuffer)[idx + 2] = clr.GetRed(); + ((BYTE*)*pBuffer)[idx + 3] = clr.GetAlpha(); + } + } + + delete bitmap; + bitmap = 0; +} +#endif diff --git a/src/gallium/drivers/swr/rasterizer/core/utils.h b/src/gallium/drivers/swr/rasterizer/core/utils.h new file mode 100644 index 00000000000..8a59ef24fee --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/core/utils.h @@ -0,0 +1,831 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file utils.h +* +* @brief Utilities used by SWR core. +* +******************************************************************************/ +#pragma once + +#include +#include "common/os.h" +#include "common/simdintrin.h" +#include "common/swr_assert.h" + +#if defined(_WIN32) +void SaveImageToPNGFile( + const WCHAR *pFilename, + void *pBuffer, + uint32_t width, + uint32_t height); + +void OpenBitmapFromFile( + const WCHAR *pFilename, + void **pBuffer, + uint32_t *width, + uint32_t *height); +#endif + +/// @todo assume linux is always 64 bit +#if defined(_WIN64) || defined(__linux__) || defined(__gnu_linux__) +#define _MM_INSERT_EPI64 _mm_insert_epi64 +#define _MM_EXTRACT_EPI64 _mm_extract_epi64 +#else +INLINE INT64 _MM_EXTRACT_EPI64(__m128i a, const int32_t ndx) +{ + OSALIGNLINE(uint32_t) elems[4]; + _mm_store_si128((__m128i*)elems, a); + if (ndx == 0) + { + uint64_t foo = elems[0]; + foo |= (uint64_t)elems[1] << 32; + return foo; + } + else + { + uint64_t foo = elems[2]; + foo |= (uint64_t)elems[3] << 32; + return foo; + } +} + +INLINE __m128i _MM_INSERT_EPI64(__m128i a, INT64 b, const int32_t ndx) +{ + OSALIGNLINE(int64_t) elems[2]; + _mm_store_si128((__m128i*)elems, a); + if (ndx == 0) + { + elems[0] = b; + } + else + { + elems[1] = b; + } + __m128i out; + out = _mm_load_si128((const __m128i*)elems); + return out; +} +#endif + +OSALIGNLINE(struct) BBOX +{ + int top, bottom, left, right; + + BBOX() {} + BBOX(int t, int b, int l, int r) : top(t), bottom(b), left(l), right(r) {} + + bool operator==(const BBOX& rhs) + { + return (this->top == rhs.top && + this->bottom == rhs.bottom && + this->left == rhs.left && + this->right == rhs.right); + } + + bool operator!=(const BBOX& rhs) + { + return !(*this == rhs); + } +}; + +struct simdBBox +{ + simdscalari top, bottom, left, right; +}; + +INLINE +void vTranspose(__m128 &row0, __m128 &row1, __m128 &row2, __m128 &row3) +{ + __m128i row0i = _mm_castps_si128(row0); + __m128i row1i = _mm_castps_si128(row1); + __m128i row2i = _mm_castps_si128(row2); + __m128i row3i = _mm_castps_si128(row3); + + __m128i vTemp = row2i; + row2i = _mm_unpacklo_epi32(row2i, row3i); + vTemp = _mm_unpackhi_epi32(vTemp, row3i); + + row3i = row0i; + row0i = _mm_unpacklo_epi32(row0i, row1i); + row3i = _mm_unpackhi_epi32(row3i, row1i); + + row1i = row0i; + row0i = _mm_unpacklo_epi64(row0i, row2i); + row1i = _mm_unpackhi_epi64(row1i, row2i); + + row2i = row3i; + row2i = _mm_unpacklo_epi64(row2i, vTemp); + row3i = _mm_unpackhi_epi64(row3i, vTemp); + + row0 = _mm_castsi128_ps(row0i); + row1 = _mm_castsi128_ps(row1i); + row2 = _mm_castsi128_ps(row2i); + row3 = _mm_castsi128_ps(row3i); +} + +INLINE +void vTranspose(__m128i &row0, __m128i &row1, __m128i &row2, __m128i &row3) +{ + __m128i vTemp = row2; + row2 = _mm_unpacklo_epi32(row2, row3); + vTemp = _mm_unpackhi_epi32(vTemp, row3); + + row3 = row0; + row0 = _mm_unpacklo_epi32(row0, row1); + row3 = _mm_unpackhi_epi32(row3, row1); + + row1 = row0; + row0 = _mm_unpacklo_epi64(row0, row2); + row1 = _mm_unpackhi_epi64(row1, row2); + + row2 = row3; + row2 = _mm_unpacklo_epi64(row2, vTemp); + row3 = _mm_unpackhi_epi64(row3, vTemp); +} + +#define GCC_VERSION (__GNUC__ * 10000 \ + + __GNUC_MINOR__ * 100 \ + + __GNUC_PATCHLEVEL__) + +#if defined(__GNUC__) && (GCC_VERSION < 40900) +#define _mm_undefined_ps _mm_setzero_ps +#define _mm_undefined_si128 _mm_setzero_si128 +#if KNOB_SIMD_WIDTH == 8 +#define _mm256_undefined_ps _mm256_setzero_ps +#endif +#endif + +#if KNOB_SIMD_WIDTH == 8 +INLINE +void vTranspose3x8(__m128 (&vDst)[8], __m256 &vSrc0, __m256 &vSrc1, __m256 &vSrc2) +{ + __m256 r0r2 = _mm256_unpacklo_ps(vSrc0, vSrc2); //x0z0x1z1 x4z4x5z5 + __m256 r1rx = _mm256_unpacklo_ps(vSrc1, _mm256_undefined_ps()); //y0w0y1w1 y4w4y5w5 + __m256 r02r1xlolo = _mm256_unpacklo_ps(r0r2, r1rx); //x0y0z0w0 x4y4z4w4 + __m256 r02r1xlohi = _mm256_unpackhi_ps(r0r2, r1rx); //x1y1z1w1 x5y5z5w5 + + r0r2 = _mm256_unpackhi_ps(vSrc0, vSrc2); //x2z2x3z3 x6z6x7z7 + r1rx = _mm256_unpackhi_ps(vSrc1, _mm256_undefined_ps()); //y2w2y3w3 y6w6yw77 + __m256 r02r1xhilo = _mm256_unpacklo_ps(r0r2, r1rx); //x2y2z2w2 x6y6z6w6 + __m256 r02r1xhihi = _mm256_unpackhi_ps(r0r2, r1rx); //x3y3z3w3 x7y7z7w7 + + vDst[0] = _mm256_castps256_ps128(r02r1xlolo); + vDst[1] = _mm256_castps256_ps128(r02r1xlohi); + vDst[2] = _mm256_castps256_ps128(r02r1xhilo); + vDst[3] = _mm256_castps256_ps128(r02r1xhihi); + + vDst[4] = _mm256_extractf128_ps(r02r1xlolo, 1); + vDst[5] = _mm256_extractf128_ps(r02r1xlohi, 1); + vDst[6] = _mm256_extractf128_ps(r02r1xhilo, 1); + vDst[7] = _mm256_extractf128_ps(r02r1xhihi, 1); +} + +INLINE +void vTranspose4x8(__m128 (&vDst)[8], __m256 &vSrc0, __m256 &vSrc1, __m256 &vSrc2, __m256 &vSrc3) +{ + __m256 r0r2 = _mm256_unpacklo_ps(vSrc0, vSrc2); //x0z0x1z1 x4z4x5z5 + __m256 r1rx = _mm256_unpacklo_ps(vSrc1, vSrc3); //y0w0y1w1 y4w4y5w5 + __m256 r02r1xlolo = _mm256_unpacklo_ps(r0r2, r1rx); //x0y0z0w0 x4y4z4w4 + __m256 r02r1xlohi = _mm256_unpackhi_ps(r0r2, r1rx); //x1y1z1w1 x5y5z5w5 + + r0r2 = _mm256_unpackhi_ps(vSrc0, vSrc2); //x2z2x3z3 x6z6x7z7 + r1rx = _mm256_unpackhi_ps(vSrc1, vSrc3) ; //y2w2y3w3 y6w6yw77 + __m256 r02r1xhilo = _mm256_unpacklo_ps(r0r2, r1rx); //x2y2z2w2 x6y6z6w6 + __m256 r02r1xhihi = _mm256_unpackhi_ps(r0r2, r1rx); //x3y3z3w3 x7y7z7w7 + + vDst[0] = _mm256_castps256_ps128(r02r1xlolo); + vDst[1] = _mm256_castps256_ps128(r02r1xlohi); + vDst[2] = _mm256_castps256_ps128(r02r1xhilo); + vDst[3] = _mm256_castps256_ps128(r02r1xhihi); + + vDst[4] = _mm256_extractf128_ps(r02r1xlolo, 1); + vDst[5] = _mm256_extractf128_ps(r02r1xlohi, 1); + vDst[6] = _mm256_extractf128_ps(r02r1xhilo, 1); + vDst[7] = _mm256_extractf128_ps(r02r1xhihi, 1); +} + +INLINE +void vTranspose8x8(__m256 (&vDst)[8], const __m256 &vMask0, const __m256 &vMask1, const __m256 &vMask2, const __m256 &vMask3, const __m256 &vMask4, const __m256 &vMask5, const __m256 &vMask6, const __m256 &vMask7) +{ + __m256 __t0 = _mm256_unpacklo_ps(vMask0, vMask1); + __m256 __t1 = _mm256_unpackhi_ps(vMask0, vMask1); + __m256 __t2 = _mm256_unpacklo_ps(vMask2, vMask3); + __m256 __t3 = _mm256_unpackhi_ps(vMask2, vMask3); + __m256 __t4 = _mm256_unpacklo_ps(vMask4, vMask5); + __m256 __t5 = _mm256_unpackhi_ps(vMask4, vMask5); + __m256 __t6 = _mm256_unpacklo_ps(vMask6, vMask7); + __m256 __t7 = _mm256_unpackhi_ps(vMask6, vMask7); + __m256 __tt0 = _mm256_shuffle_ps(__t0,__t2,_MM_SHUFFLE(1,0,1,0)); + __m256 __tt1 = _mm256_shuffle_ps(__t0,__t2,_MM_SHUFFLE(3,2,3,2)); + __m256 __tt2 = _mm256_shuffle_ps(__t1,__t3,_MM_SHUFFLE(1,0,1,0)); + __m256 __tt3 = _mm256_shuffle_ps(__t1,__t3,_MM_SHUFFLE(3,2,3,2)); + __m256 __tt4 = _mm256_shuffle_ps(__t4,__t6,_MM_SHUFFLE(1,0,1,0)); + __m256 __tt5 = _mm256_shuffle_ps(__t4,__t6,_MM_SHUFFLE(3,2,3,2)); + __m256 __tt6 = _mm256_shuffle_ps(__t5,__t7,_MM_SHUFFLE(1,0,1,0)); + __m256 __tt7 = _mm256_shuffle_ps(__t5,__t7,_MM_SHUFFLE(3,2,3,2)); + vDst[0] = _mm256_permute2f128_ps(__tt0, __tt4, 0x20); + vDst[1] = _mm256_permute2f128_ps(__tt1, __tt5, 0x20); + vDst[2] = _mm256_permute2f128_ps(__tt2, __tt6, 0x20); + vDst[3] = _mm256_permute2f128_ps(__tt3, __tt7, 0x20); + vDst[4] = _mm256_permute2f128_ps(__tt0, __tt4, 0x31); + vDst[5] = _mm256_permute2f128_ps(__tt1, __tt5, 0x31); + vDst[6] = _mm256_permute2f128_ps(__tt2, __tt6, 0x31); + vDst[7] = _mm256_permute2f128_ps(__tt3, __tt7, 0x31); +} + +INLINE +void vTranspose8x8(__m256 (&vDst)[8], const __m256i &vMask0, const __m256i &vMask1, const __m256i &vMask2, const __m256i &vMask3, const __m256i &vMask4, const __m256i &vMask5, const __m256i &vMask6, const __m256i &vMask7) +{ + vTranspose8x8(vDst, _mm256_castsi256_ps(vMask0), _mm256_castsi256_ps(vMask1), _mm256_castsi256_ps(vMask2), _mm256_castsi256_ps(vMask3), + _mm256_castsi256_ps(vMask4), _mm256_castsi256_ps(vMask5), _mm256_castsi256_ps(vMask6), _mm256_castsi256_ps(vMask7)); +} +#endif + +////////////////////////////////////////////////////////////////////////// +/// TranposeSingleComponent +////////////////////////////////////////////////////////////////////////// +template +struct TransposeSingleComponent +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Pass-thru for single component. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + INLINE static void Transpose(const BYTE* pSrc, BYTE* pDst) + { + memcpy(pDst, pSrc, (bpp * KNOB_SIMD_WIDTH) / 8); + } +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose8_8_8_8 +////////////////////////////////////////////////////////////////////////// +struct Transpose8_8_8_8 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 8_8_8_8 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + INLINE static void Transpose(const BYTE* pSrc, BYTE* pDst) + { + simdscalari src = _simd_load_si((const simdscalari*)pSrc); +#if KNOB_SIMD_WIDTH == 8 +#if KNOB_ARCH == KNOB_ARCH_AVX + __m128i c0c1 = _mm256_castsi256_si128(src); // rrrrrrrrgggggggg + __m128i c2c3 = _mm_castps_si128(_mm256_extractf128_ps(_mm256_castsi256_ps(src), 1)); // bbbbbbbbaaaaaaaa + __m128i c0c2 = _mm_unpacklo_epi64(c0c1, c2c3); // rrrrrrrrbbbbbbbb + __m128i c1c3 = _mm_unpackhi_epi64(c0c1, c2c3); // ggggggggaaaaaaaa + __m128i c01 = _mm_unpacklo_epi8(c0c2, c1c3); // rgrgrgrgrgrgrgrg + __m128i c23 = _mm_unpackhi_epi8(c0c2, c1c3); // babababababababa + __m128i c0123lo = _mm_unpacklo_epi16(c01, c23); // rgbargbargbargba + __m128i c0123hi = _mm_unpackhi_epi16(c01, c23); // rgbargbargbargba + _mm_store_si128((__m128i*)pDst, c0123lo); + _mm_store_si128((__m128i*)(pDst + 16), c0123hi); +#elif KNOB_ARCH == KNOB_ARCH_AVX2 + simdscalari dst01 = _mm256_shuffle_epi8(src, + _mm256_set_epi32(0x0f078080, 0x0e068080, 0x0d058080, 0x0c048080, 0x80800b03, 0x80800a02, 0x80800901, 0x80800800)); + simdscalari dst23 = _mm256_permute2x128_si256(src, src, 0x01); + dst23 = _mm256_shuffle_epi8(dst23, + _mm256_set_epi32(0x80800f07, 0x80800e06, 0x80800d05, 0x80800c04, 0x0b038080, 0x0a028080, 0x09018080, 0x08008080)); + simdscalari dst = _mm256_or_si256(dst01, dst23); + _simd_store_si((simdscalari*)pDst, dst); +#endif +#else +#error Unsupported vector width +#endif + } +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose8_8_8 +////////////////////////////////////////////////////////////////////////// +struct Transpose8_8_8 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 8_8_8 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + INLINE static void Transpose(const BYTE* pSrc, BYTE* pDst) = delete; +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose8_8 +////////////////////////////////////////////////////////////////////////// +struct Transpose8_8 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 8_8 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + INLINE static void Transpose(const BYTE* pSrc, BYTE* pDst) + { + simdscalari src = _simd_load_si((const simdscalari*)pSrc); + +#if KNOB_SIMD_WIDTH == 8 + __m128i rg = _mm256_castsi256_si128(src); // rrrrrrrr gggggggg + __m128i g = _mm_unpackhi_epi64(rg, rg); // gggggggg gggggggg + rg = _mm_unpacklo_epi8(rg, g); + _mm_store_si128((__m128i*)pDst, rg); +#else +#error Unsupported vector width +#endif + } +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose32_32_32_32 +////////////////////////////////////////////////////////////////////////// +struct Transpose32_32_32_32 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 32_32_32_32 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + INLINE static void Transpose(const BYTE* pSrc, BYTE* pDst) + { +#if KNOB_SIMD_WIDTH == 8 + simdscalar src0 = _simd_load_ps((const float*)pSrc); + simdscalar src1 = _simd_load_ps((const float*)pSrc + 8); + simdscalar src2 = _simd_load_ps((const float*)pSrc + 16); + simdscalar src3 = _simd_load_ps((const float*)pSrc + 24); + + __m128 vDst[8]; + vTranspose4x8(vDst, src0, src1, src2, src3); + _mm_store_ps((float*)pDst, vDst[0]); + _mm_store_ps((float*)pDst+4, vDst[1]); + _mm_store_ps((float*)pDst+8, vDst[2]); + _mm_store_ps((float*)pDst+12, vDst[3]); + _mm_store_ps((float*)pDst+16, vDst[4]); + _mm_store_ps((float*)pDst+20, vDst[5]); + _mm_store_ps((float*)pDst+24, vDst[6]); + _mm_store_ps((float*)pDst+28, vDst[7]); +#else +#error Unsupported vector width +#endif + } +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose32_32_32 +////////////////////////////////////////////////////////////////////////// +struct Transpose32_32_32 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 32_32_32 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + INLINE static void Transpose(const BYTE* pSrc, BYTE* pDst) + { +#if KNOB_SIMD_WIDTH == 8 + simdscalar src0 = _simd_load_ps((const float*)pSrc); + simdscalar src1 = _simd_load_ps((const float*)pSrc + 8); + simdscalar src2 = _simd_load_ps((const float*)pSrc + 16); + + __m128 vDst[8]; + vTranspose3x8(vDst, src0, src1, src2); + _mm_store_ps((float*)pDst, vDst[0]); + _mm_store_ps((float*)pDst + 4, vDst[1]); + _mm_store_ps((float*)pDst + 8, vDst[2]); + _mm_store_ps((float*)pDst + 12, vDst[3]); + _mm_store_ps((float*)pDst + 16, vDst[4]); + _mm_store_ps((float*)pDst + 20, vDst[5]); + _mm_store_ps((float*)pDst + 24, vDst[6]); + _mm_store_ps((float*)pDst + 28, vDst[7]); +#else +#error Unsupported vector width +#endif + } +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose32_32 +////////////////////////////////////////////////////////////////////////// +struct Transpose32_32 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 32_32 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + INLINE static void Transpose(const BYTE* pSrc, BYTE* pDst) + { + const float* pfSrc = (const float*)pSrc; + __m128 src_r0 = _mm_load_ps(pfSrc + 0); + __m128 src_r1 = _mm_load_ps(pfSrc + 4); + __m128 src_g0 = _mm_load_ps(pfSrc + 8); + __m128 src_g1 = _mm_load_ps(pfSrc + 12); + + __m128 dst0 = _mm_unpacklo_ps(src_r0, src_g0); + __m128 dst1 = _mm_unpackhi_ps(src_r0, src_g0); + __m128 dst2 = _mm_unpacklo_ps(src_r1, src_g1); + __m128 dst3 = _mm_unpackhi_ps(src_r1, src_g1); + + float* pfDst = (float*)pDst; + _mm_store_ps(pfDst + 0, dst0); + _mm_store_ps(pfDst + 4, dst1); + _mm_store_ps(pfDst + 8, dst2); + _mm_store_ps(pfDst + 12, dst3); + } +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose16_16_16_16 +////////////////////////////////////////////////////////////////////////// +struct Transpose16_16_16_16 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 16_16_16_16 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + INLINE static void Transpose(const BYTE* pSrc, BYTE* pDst) + { +#if KNOB_SIMD_WIDTH == 8 + simdscalari src_rg = _simd_load_si((const simdscalari*)pSrc); + simdscalari src_ba = _simd_load_si((const simdscalari*)(pSrc + sizeof(simdscalari))); + + __m128i src_r = _mm256_extractf128_si256(src_rg, 0); + __m128i src_g = _mm256_extractf128_si256(src_rg, 1); + __m128i src_b = _mm256_extractf128_si256(src_ba, 0); + __m128i src_a = _mm256_extractf128_si256(src_ba, 1); + + __m128i rg0 = _mm_unpacklo_epi16(src_r, src_g); + __m128i rg1 = _mm_unpackhi_epi16(src_r, src_g); + __m128i ba0 = _mm_unpacklo_epi16(src_b, src_a); + __m128i ba1 = _mm_unpackhi_epi16(src_b, src_a); + + __m128i dst0 = _mm_unpacklo_epi32(rg0, ba0); + __m128i dst1 = _mm_unpackhi_epi32(rg0, ba0); + __m128i dst2 = _mm_unpacklo_epi32(rg1, ba1); + __m128i dst3 = _mm_unpackhi_epi32(rg1, ba1); + + _mm_store_si128(((__m128i*)pDst) + 0, dst0); + _mm_store_si128(((__m128i*)pDst) + 1, dst1); + _mm_store_si128(((__m128i*)pDst) + 2, dst2); + _mm_store_si128(((__m128i*)pDst) + 3, dst3); +#else +#error Unsupported vector width +#endif + } +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose16_16_16 +////////////////////////////////////////////////////////////////////////// +struct Transpose16_16_16 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 16_16_16 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + INLINE static void Transpose(const BYTE* pSrc, BYTE* pDst) + { +#if KNOB_SIMD_WIDTH == 8 + simdscalari src_rg = _simd_load_si((const simdscalari*)pSrc); + + __m128i src_r = _mm256_extractf128_si256(src_rg, 0); + __m128i src_g = _mm256_extractf128_si256(src_rg, 1); + __m128i src_b = _mm_load_si128((const __m128i*)(pSrc + sizeof(simdscalari))); + __m128i src_a = _mm_undefined_si128(); + + __m128i rg0 = _mm_unpacklo_epi16(src_r, src_g); + __m128i rg1 = _mm_unpackhi_epi16(src_r, src_g); + __m128i ba0 = _mm_unpacklo_epi16(src_b, src_a); + __m128i ba1 = _mm_unpackhi_epi16(src_b, src_a); + + __m128i dst0 = _mm_unpacklo_epi32(rg0, ba0); + __m128i dst1 = _mm_unpackhi_epi32(rg0, ba0); + __m128i dst2 = _mm_unpacklo_epi32(rg1, ba1); + __m128i dst3 = _mm_unpackhi_epi32(rg1, ba1); + + _mm_store_si128(((__m128i*)pDst) + 0, dst0); + _mm_store_si128(((__m128i*)pDst) + 1, dst1); + _mm_store_si128(((__m128i*)pDst) + 2, dst2); + _mm_store_si128(((__m128i*)pDst) + 3, dst3); +#else +#error Unsupported vector width +#endif + } +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose16_16 +////////////////////////////////////////////////////////////////////////// +struct Transpose16_16 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 16_16 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + INLINE static void Transpose(const BYTE* pSrc, BYTE* pDst) + { + simdscalar src = _simd_load_ps((const float*)pSrc); + +#if KNOB_SIMD_WIDTH == 8 + __m128 comp0 = _mm256_castps256_ps128(src); + __m128 comp1 = _mm256_extractf128_ps(src, 1); + + __m128i comp0i = _mm_castps_si128(comp0); + __m128i comp1i = _mm_castps_si128(comp1); + + __m128i resLo = _mm_unpacklo_epi16(comp0i, comp1i); + __m128i resHi = _mm_unpackhi_epi16(comp0i, comp1i); + + _mm_store_si128((__m128i*)pDst, resLo); + _mm_store_si128((__m128i*)pDst + 1, resHi); +#else +#error Unsupported vector width +#endif + } +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose24_8 +////////////////////////////////////////////////////////////////////////// +struct Transpose24_8 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 24_8 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + static void Transpose(const BYTE* pSrc, BYTE* pDst) = delete; +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose32_8_24 +////////////////////////////////////////////////////////////////////////// +struct Transpose32_8_24 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 32_8_24 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + static void Transpose(const BYTE* pSrc, BYTE* pDst) = delete; +}; + + + +////////////////////////////////////////////////////////////////////////// +/// Transpose4_4_4_4 +////////////////////////////////////////////////////////////////////////// +struct Transpose4_4_4_4 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 4_4_4_4 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + static void Transpose(const BYTE* pSrc, BYTE* pDst) = delete; +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose5_6_5 +////////////////////////////////////////////////////////////////////////// +struct Transpose5_6_5 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 5_6_5 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + static void Transpose(const BYTE* pSrc, BYTE* pDst) = delete; +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose9_9_9_5 +////////////////////////////////////////////////////////////////////////// +struct Transpose9_9_9_5 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 9_9_9_5 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + static void Transpose(const BYTE* pSrc, BYTE* pDst) = delete; +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose5_5_5_1 +////////////////////////////////////////////////////////////////////////// +struct Transpose5_5_5_1 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 5_5_5_1 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + static void Transpose(const BYTE* pSrc, BYTE* pDst) = delete; +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose10_10_10_2 +////////////////////////////////////////////////////////////////////////// +struct Transpose10_10_10_2 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 10_10_10_2 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + static void Transpose(const BYTE* pSrc, BYTE* pDst) = delete; +}; + +////////////////////////////////////////////////////////////////////////// +/// Transpose11_11_10 +////////////////////////////////////////////////////////////////////////// +struct Transpose11_11_10 +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Performs an SOA to AOS conversion for packed 11_11_10 data. + /// @param pSrc - source data in SOA form + /// @param pDst - output data in AOS form + static void Transpose(const BYTE* pSrc, BYTE* pDst) = delete; +}; + +// helper function to unroll loops +template +struct UnrollerL { + template + INLINE static void step(Lambda& func) { + func(Begin); + UnrollerL::step(func); + } +}; + +template +struct UnrollerL { + template + static void step(Lambda& func) { + } +}; + +// general CRC compute +INLINE +uint32_t ComputeCRC(uint32_t crc, const void *pData, uint32_t size) +{ +#if defined(_WIN64) || defined(__linux__) || defined(__gnu_linux__) + uint32_t sizeInQwords = size / sizeof(uint64_t); + uint32_t sizeRemainderBytes = size % sizeof(uint64_t); + uint64_t* pDataWords = (uint64_t*)pData; + for (uint32_t i = 0; i < sizeInQwords; ++i) + { + crc = (uint32_t)_mm_crc32_u64(crc, *pDataWords++); + } +#else + uint32_t sizeInDwords = size / sizeof(uint32_t); + uint32_t sizeRemainderBytes = size % sizeof(uint32_t); + uint32_t* pDataWords = (uint32_t*)pData; + for (uint32_t i = 0; i < sizeInDwords; ++i) + { + crc = _mm_crc32_u32(crc, *pDataWords++); + } +#endif + + BYTE* pRemainderBytes = (BYTE*)pDataWords; + for (uint32_t i = 0; i < sizeRemainderBytes; ++i) + { + crc = _mm_crc32_u8(crc, *pRemainderBytes++); + } + + return crc; +} + +////////////////////////////////////////////////////////////////////////// +/// Add byte offset to any-type pointer +////////////////////////////////////////////////////////////////////////// +template +INLINE +static T* PtrAdd(T* p, intptr_t offset) +{ + intptr_t intp = reinterpret_cast(p); + return reinterpret_cast(intp + offset); +} + +////////////////////////////////////////////////////////////////////////// +/// Is a power-of-2? +////////////////////////////////////////////////////////////////////////// +template +INLINE +static bool IsPow2(T value) +{ + return value == (value & (0 - value)); +} + +////////////////////////////////////////////////////////////////////////// +/// Align down to specified alignment +/// Note: IsPow2(alignment) MUST be true +////////////////////////////////////////////////////////////////////////// +template +INLINE +static T1 AlignDownPow2(T1 value, T2 alignment) +{ + SWR_ASSERT(IsPow2(alignment)); + return value & ~T1(alignment - 1); +} + +////////////////////////////////////////////////////////////////////////// +/// Align up to specified alignment +/// Note: IsPow2(alignment) MUST be true +////////////////////////////////////////////////////////////////////////// +template +INLINE +static T1 AlignUpPow2(T1 value, T2 alignment) +{ + return AlignDownPow2(value + T1(alignment - 1), alignment); +} + +////////////////////////////////////////////////////////////////////////// +/// Align up ptr to specified alignment +/// Note: IsPow2(alignment) MUST be true +////////////////////////////////////////////////////////////////////////// +template +INLINE +static T1* AlignUpPow2(T1* value, T2 alignment) +{ + return reinterpret_cast( + AlignDownPow2(reinterpret_cast(value) + uintptr_t(alignment - 1), alignment)); +} + +////////////////////////////////////////////////////////////////////////// +/// Align down to specified alignment +////////////////////////////////////////////////////////////////////////// +template +INLINE +static T1 AlignDown(T1 value, T2 alignment) +{ + if (IsPow2(alignment)) { return AlignDownPow2(value, alignment); } + return value - T1(value % alignment); +} + +////////////////////////////////////////////////////////////////////////// +/// Align down to specified alignment +////////////////////////////////////////////////////////////////////////// +template +INLINE +static T1* AlignDown(T1* value, T2 alignment) +{ + return (T1*)AlignDown(uintptr_t(value), alignment); +} + +////////////////////////////////////////////////////////////////////////// +/// Align up to specified alignment +/// Note: IsPow2(alignment) MUST be true +////////////////////////////////////////////////////////////////////////// +template +INLINE +static T1 AlignUp(T1 value, T2 alignment) +{ + return AlignDown(value + T1(alignment - 1), alignment); +} + +////////////////////////////////////////////////////////////////////////// +/// Align up to specified alignment +/// Note: IsPow2(alignment) MUST be true +////////////////////////////////////////////////////////////////////////// +template +INLINE +static T1* AlignUp(T1* value, T2 alignment) +{ + return AlignDown(PtrAdd(value, alignment - 1), alignment); +} + +////////////////////////////////////////////////////////////////////////// +/// Helper structure used to access an array of elements that don't +/// correspond to a typical word size. +////////////////////////////////////////////////////////////////////////// +template +class BitsArray +{ +private: + static const size_t BITS_PER_WORD = sizeof(size_t) * 8; + static const size_t ELEMENTS_PER_WORD = BITS_PER_WORD / BitsPerElementT; + static const size_t NUM_WORDS = (ArrayLenT + ELEMENTS_PER_WORD - 1) / ELEMENTS_PER_WORD; + static const size_t ELEMENT_MASK = (size_t(1) << BitsPerElementT) - 1; + + static_assert(ELEMENTS_PER_WORD * BitsPerElementT == BITS_PER_WORD, + "Element size must an integral fraction of pointer size"); + + size_t m_words[NUM_WORDS] = {}; + +public: + + T operator[] (size_t elementIndex) const + { + size_t word = m_words[elementIndex / ELEMENTS_PER_WORD]; + word >>= ((elementIndex % ELEMENTS_PER_WORD) * BitsPerElementT); + return T(word & ELEMENT_MASK); + } +}; diff --git a/src/gallium/drivers/swr/rasterizer/jitter/JitManager.cpp b/src/gallium/drivers/swr/rasterizer/jitter/JitManager.cpp new file mode 100644 index 00000000000..734c89792f0 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/JitManager.cpp @@ -0,0 +1,313 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file JitManager.cpp +* +* @brief Implementation if the Jit Manager. +* +* Notes: +* +******************************************************************************/ +#if defined(_WIN32) +#pragma warning(disable: 4800 4146 4244 4267 4355 4996) +#endif + +#include "jit_api.h" +#include "JitManager.h" +#include "fetch_jit.h" + +#if defined(_WIN32) +#include "llvm/ADT/Triple.h" +#endif +#include "llvm/IR/Function.h" +#include "llvm/Support/DynamicLibrary.h" + +#include "llvm/Support/MemoryBuffer.h" +#include "llvm/Support/SourceMgr.h" + +#include "llvm/Analysis/CFGPrinter.h" +#include "llvm/IRReader/IRReader.h" + +#include "core/state.h" +#include "common/containers.hpp" + +#include "state_llvm.h" + +#include +#if defined(_WIN32) +#include +#include + +#define INTEL_OUTPUT_DIR "c:\\Intel" +#define SWR_OUTPUT_DIR INTEL_OUTPUT_DIR "\\SWR" +#define JITTER_OUTPUT_DIR SWR_OUTPUT_DIR "\\Jitter" +#endif + +using namespace llvm; + +////////////////////////////////////////////////////////////////////////// +/// @brief Contructor for JitManager. +/// @param simdWidth - SIMD width to be used in generated program. +JitManager::JitManager(uint32_t simdWidth, const char *arch) + : mContext(), mBuilder(mContext), mIsModuleFinalized(true), mJitNumber(0), mVWidth(simdWidth), mArch(arch) +{ + InitializeNativeTarget(); + InitializeNativeTargetAsmPrinter(); + InitializeNativeTargetDisassembler(); + + TargetOptions tOpts; + tOpts.AllowFPOpFusion = FPOpFusion::Fast; + tOpts.NoInfsFPMath = false; + tOpts.NoNaNsFPMath = false; + tOpts.UnsafeFPMath = true; +#if defined(_DEBUG) +#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR < 7 + tOpts.NoFramePointerElim = true; +#endif +#endif + + //tOpts.PrintMachineCode = true; + + std::stringstream fnName("JitModule", std::ios_base::in | std::ios_base::out | std::ios_base::ate); + fnName << mJitNumber++; + std::unique_ptr newModule(new Module(fnName.str(), mContext)); + mpCurrentModule = newModule.get(); + + auto &&EB = EngineBuilder(std::move(newModule)); + EB.setTargetOptions(tOpts); + EB.setOptLevel(CodeGenOpt::Aggressive); + + StringRef hostCPUName; + + // force JIT to use the same CPU arch as the rest of swr + if(mArch.AVX512F()) + { + assert(0 && "Implement AVX512 jitter"); + hostCPUName = sys::getHostCPUName(); + if (mVWidth == 0) + { + mVWidth = 16; + } + } + else if(mArch.AVX2()) + { + hostCPUName = StringRef("core-avx2"); + if (mVWidth == 0) + { + mVWidth = 8; + } + } + else if(mArch.AVX()) + { + if (mArch.F16C()) + { + hostCPUName = StringRef("core-avx-i"); + } + else + { + hostCPUName = StringRef("corei7-avx"); + } + if (mVWidth == 0) + { + mVWidth = 8; + } + } + else + { + hostCPUName = sys::getHostCPUName(); + if (mVWidth == 0) + { + mVWidth = 8; // 4? + } + } + + EB.setMCPU(hostCPUName); + +#if defined(_WIN32) + // Needed for MCJIT on windows + Triple hostTriple(sys::getProcessTriple()); + hostTriple.setObjectFormat(Triple::ELF); + mpCurrentModule->setTargetTriple(hostTriple.getTriple()); +#endif // _WIN32 + + mpExec = EB.create(); + +#if LLVM_USE_INTEL_JITEVENTS + JITEventListener *vTune = JITEventListener::createIntelJITEventListener(); + mpExec->RegisterJITEventListener(vTune); +#endif + + mFP32Ty = Type::getFloatTy(mContext); // float type + mInt8Ty = Type::getInt8Ty(mContext); + mInt32Ty = Type::getInt32Ty(mContext); // int type + mInt64Ty = Type::getInt64Ty(mContext); // int type + mV4FP32Ty = StructType::get(mContext, std::vector(4, mFP32Ty), false); // vector4 float type (represented as structure) + mV4Int32Ty = StructType::get(mContext, std::vector(4, mInt32Ty), false); // vector4 int type + + // fetch function signature + // typedef void(__cdecl *PFN_FETCH_FUNC)(SWR_FETCH_CONTEXT& fetchInfo, simdvertex& out); + std::vector fsArgs; + fsArgs.push_back(PointerType::get(Gen_SWR_FETCH_CONTEXT(this), 0)); + fsArgs.push_back(PointerType::get(Gen_simdvertex(this), 0)); + + mFetchShaderTy = FunctionType::get(Type::getVoidTy(mContext), fsArgs, false); + + mSimtFP32Ty = VectorType::get(mFP32Ty, mVWidth); + mSimtInt32Ty = VectorType::get(mInt32Ty, mVWidth); + + mSimdVectorTy = StructType::get(mContext, std::vector(4, mSimtFP32Ty), false); + mSimdVectorInt32Ty = StructType::get(mContext, std::vector(4, mSimtInt32Ty), false); + +#if defined(_WIN32) + // explicitly instantiate used symbols from potentially staticly linked libs + sys::DynamicLibrary::AddSymbol("exp2f", &exp2f); + sys::DynamicLibrary::AddSymbol("log2f", &log2f); + sys::DynamicLibrary::AddSymbol("sinf", &sinf); + sys::DynamicLibrary::AddSymbol("cosf", &cosf); + sys::DynamicLibrary::AddSymbol("powf", &powf); +#endif + +#if defined(_WIN32) + if (KNOB_DUMP_SHADER_IR) + { + CreateDirectory(INTEL_OUTPUT_DIR, NULL); + CreateDirectory(SWR_OUTPUT_DIR, NULL); + CreateDirectory(JITTER_OUTPUT_DIR, NULL); + } + + ///@todo Figure out a better solution for this. + // Redirect stdin, stdout, and stderr to attached console. + freopen("CONIN$", "r", stdin); + freopen("CONOUT$", "w", stdout); + freopen("CONOUT$", "w", stderr); +#endif +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Create new LLVM module. +void JitManager::SetupNewModule() +{ + SWR_ASSERT(mIsModuleFinalized == true && "Current module is not finalized!"); + + std::stringstream fnName("JitModule", std::ios_base::in | std::ios_base::out | std::ios_base::ate); + fnName << mJitNumber++; + std::unique_ptr newModule(new Module(fnName.str(), mContext)); + mpCurrentModule = newModule.get(); +#if defined(_WIN32) + // Needed for MCJIT on windows + Triple hostTriple(sys::getProcessTriple()); + hostTriple.setObjectFormat(Triple::ELF); + newModule->setTargetTriple(hostTriple.getTriple()); +#endif // _WIN32 + + mpExec->addModule(std::move(newModule)); + mIsModuleFinalized = false; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Create new LLVM module from IR. +bool JitManager::SetupModuleFromIR(const uint8_t *pIR) +{ + std::unique_ptr pMem = MemoryBuffer::getMemBuffer(StringRef((const char*)pIR), ""); + + SMDiagnostic Err; + std::unique_ptr newModule = parseIR(pMem.get()->getMemBufferRef(), Err, mContext); + + if (newModule == nullptr) + { + SWR_ASSERT(0, "Parse failed! Check Err for details."); + return false; + } + + mpCurrentModule = newModule.get(); +#if defined(_WIN32) + // Needed for MCJIT on windows + Triple hostTriple(sys::getProcessTriple()); + hostTriple.setObjectFormat(Triple::ELF); + newModule->setTargetTriple(hostTriple.getTriple()); +#endif // _WIN32 + + mpExec->addModule(std::move(newModule)); + mIsModuleFinalized = false; + + return true; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Dump function to file. +void JitManager::DumpToFile(Function *f, const char *fileName) +{ + if (KNOB_DUMP_SHADER_IR) + { +#if defined(_WIN32) + DWORD pid = GetCurrentProcessId(); + TCHAR procname[MAX_PATH]; + GetModuleFileName(NULL, procname, MAX_PATH); + const char* pBaseName = strrchr(procname, '\\'); + std::stringstream outDir; + outDir << JITTER_OUTPUT_DIR << pBaseName << "_" << pid << std::ends; + CreateDirectory(outDir.str().c_str(), NULL); +#endif + + std::error_code EC; + const char *funcName = f->getName().data(); + char fName[256]; +#if defined(_WIN32) + sprintf(fName, "%s\\%s.%s.ll", outDir.str().c_str(), funcName, fileName); +#else + sprintf(fName, "%s.%s.ll", funcName, fileName); +#endif + raw_fd_ostream fd(fName, EC, llvm::sys::fs::F_None); + Module* pModule = f->getParent(); + pModule->print(fd, nullptr); + +#if defined(_WIN32) + sprintf(fName, "%s\\cfg.%s.%s.dot", outDir.str().c_str(), funcName, fileName); +#else + sprintf(fName, "cfg.%s.%s.dot", funcName, fileName); +#endif + fd.flush(); + + raw_fd_ostream fd_cfg(fName, EC, llvm::sys::fs::F_Text); + WriteGraph(fd_cfg, (const Function*)f); + + fd_cfg.flush(); + } +} + +extern "C" +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Create JIT context. + /// @param simdWidth - SIMD width to be used in generated program. + HANDLE JITCALL JitCreateContext(uint32_t targetSimdWidth, const char* arch) + { + return new JitManager(targetSimdWidth, arch); + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Destroy JIT context. + void JITCALL JitDestroyContext(HANDLE hJitContext) + { + delete reinterpret_cast(hJitContext); + } +} diff --git a/src/gallium/drivers/swr/rasterizer/jitter/JitManager.h b/src/gallium/drivers/swr/rasterizer/jitter/JitManager.h new file mode 100644 index 00000000000..c974a611224 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/JitManager.h @@ -0,0 +1,186 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file JitManager.h +* +* @brief JitManager contains the LLVM data structures used for JIT generation +* +* Notes: +* +******************************************************************************/ +#pragma once + +#include "common/os.h" +#include "common/isa.hpp" + +#if defined(_WIN32) +#pragma warning(disable : 4146 4244 4267 4800 4996) +#endif + +// llvm 3.7+ reuses "DEBUG" as an enum value +#pragma push_macro("DEBUG") +#undef DEBUG + +#include "llvm/IR/DataLayout.h" +#include "llvm/IR/Instructions.h" +#include "llvm/IR/LLVMContext.h" +#include "llvm/IR/Module.h" +#include "llvm/IR/Type.h" +#include "llvm/IR/IRBuilder.h" +#include "llvm/IR/IntrinsicInst.h" + +#include "llvm/Config/llvm-config.h" +#ifndef LLVM_VERSION_MAJOR +#include "llvm/Config/config.h" +#endif + +#include "llvm/IR/Verifier.h" +#include "llvm/ExecutionEngine/MCJIT.h" +#include "llvm/Support/FileSystem.h" +#define LLVM_F_NONE sys::fs::F_None + +#include "llvm/Analysis/Passes.h" + +#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 6 +#include "llvm/PassManager.h" +#else +#include "llvm/IR/LegacyPassManager.h" +using namespace llvm::legacy; +#endif + +#include "llvm/CodeGen/Passes.h" +#include "llvm/ExecutionEngine/ExecutionEngine.h" +#include "llvm/Support/raw_ostream.h" +#include "llvm/Support/TargetSelect.h" +#include "llvm/Transforms/IPO.h" +#include "llvm/Transforms/Scalar.h" +#include "llvm/Support/Host.h" + + +#pragma pop_macro("DEBUG") + +using namespace llvm; +////////////////////////////////////////////////////////////////////////// +/// JitInstructionSet +/// @brief Subclass of InstructionSet that allows users to override +/// the reporting of support for certain ISA features. This allows capping +/// the jitted code to a certain feature level, e.g. jit AVX level code on +/// a platform that supports AVX2. +////////////////////////////////////////////////////////////////////////// +class JitInstructionSet : public InstructionSet +{ +public: + JitInstructionSet(const char* requestedIsa) : isaRequest(requestedIsa) + { + std::transform(isaRequest.begin(), isaRequest.end(), isaRequest.begin(), ::tolower); + + if(isaRequest == "avx") + { + bForceAVX = true; + bForceAVX2 = false; + bForceAVX512 = false; + } + else if(isaRequest == "avx2") + { + bForceAVX = false; + bForceAVX2 = true; + bForceAVX512 = false; + } + #if 0 + else if(isaRequest == "avx512") + { + bForceAVX = false; + bForceAVX2 = false; + bForceAVX512 = true; + } + #endif + }; + + bool AVX2(void) { return bForceAVX ? 0 : InstructionSet::AVX2(); } + bool AVX512F(void) { return (bForceAVX | bForceAVX2) ? 0 : InstructionSet::AVX512F(); } + bool BMI2(void) { return bForceAVX ? 0 : InstructionSet::BMI2(); } + +private: + bool bForceAVX = false; + bool bForceAVX2 = false; + bool bForceAVX512 = false; + std::string isaRequest; +}; + + + +struct JitLLVMContext : LLVMContext +{ +}; + + +////////////////////////////////////////////////////////////////////////// +/// JitManager +////////////////////////////////////////////////////////////////////////// +struct JitManager +{ + JitManager(uint32_t w, const char *arch); + ~JitManager(){}; + + JitLLVMContext mContext; ///< LLVM compiler + IRBuilder<> mBuilder; ///< LLVM IR Builder + ExecutionEngine* mpExec; + + // Need to be rebuilt after a JIT and before building new IR + Module* mpCurrentModule; + bool mIsModuleFinalized; + uint32_t mJitNumber; + + uint32_t mVWidth; + + // Built in types. + Type* mInt8Ty; + Type* mInt32Ty; + Type* mInt64Ty; + Type* mFP32Ty; + StructType* mV4FP32Ty; + StructType* mV4Int32Ty; + + // helper scalar function types + FunctionType* mUnaryFPTy; + FunctionType* mBinaryFPTy; + FunctionType* mTrinaryFPTy; + FunctionType* mUnaryIntTy; + FunctionType* mBinaryIntTy; + FunctionType* mTrinaryIntTy; + + Type* mSimtFP32Ty; + Type* mSimtInt32Ty; + + Type* mSimdVectorInt32Ty; + Type* mSimdVectorTy; + + // fetch shader types + FunctionType* mFetchShaderTy; + + JitInstructionSet mArch; + + void SetupNewModule(); + bool SetupModuleFromIR(const uint8_t *pIR); + + static void DumpToFile(Function *f, const char *fileName); +}; diff --git a/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp new file mode 100644 index 00000000000..954524afd3a --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.cpp @@ -0,0 +1,772 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file blend_jit.cpp +* +* @brief Implementation of the blend jitter +* +* Notes: +* +******************************************************************************/ +#include "jit_api.h" +#include "blend_jit.h" +#include "builder.h" +#include "state_llvm.h" +#include "common/containers.hpp" +#include "llvm/IR/DataLayout.h" + +#include + +// components with bit-widths <= the QUANTIZE_THRESHOLD will be quantized +#define QUANTIZE_THRESHOLD 2 + +////////////////////////////////////////////////////////////////////////// +/// Interface to Jitting a blend shader +////////////////////////////////////////////////////////////////////////// +struct BlendJit : public Builder +{ + BlendJit(JitManager* pJitMgr) : Builder(pJitMgr){}; + + template + void GenerateBlendFactor(SWR_BLEND_FACTOR factor, Value* constColor[4], Value* src[4], Value* src1[4], Value* dst[4], Value* result[4]) + { + Value* out[4]; + + switch (factor) + { + case BLENDFACTOR_ONE: + out[0] = out[1] = out[2] = out[3] = VIMMED1(1.0f); + break; + case BLENDFACTOR_SRC_COLOR: + out[0] = src[0]; + out[1] = src[1]; + out[2] = src[2]; + out[3] = src[3]; + break; + case BLENDFACTOR_SRC_ALPHA: + out[0] = out[1] = out[2] = out[3] = src[3]; + break; + case BLENDFACTOR_DST_ALPHA: + out[0] = out[1] = out[2] = out[3] = dst[3]; + break; + case BLENDFACTOR_DST_COLOR: + out[0] = dst[0]; + out[1] = dst[1]; + out[2] = dst[2]; + out[3] = dst[3]; + break; + case BLENDFACTOR_SRC_ALPHA_SATURATE: + out[0] = out[1] = out[2] = VMINPS(src[3], FSUB(VIMMED1(1.0f), dst[3])); + out[3] = VIMMED1(1.0f); + break; + case BLENDFACTOR_CONST_COLOR: + out[0] = constColor[0]; + out[1] = constColor[1]; + out[2] = constColor[2]; + out[3] = constColor[3]; + break; + case BLENDFACTOR_CONST_ALPHA: + out[0] = out[1] = out[2] = out[3] = constColor[3]; + break; + case BLENDFACTOR_SRC1_COLOR: + out[0] = src1[0]; + out[1] = src1[1]; + out[2] = src1[2]; + out[3] = src1[3]; + break; + case BLENDFACTOR_SRC1_ALPHA: + out[0] = out[1] = out[2] = out[3] = src1[3]; + break; + case BLENDFACTOR_ZERO: + out[0] = out[1] = out[2] = out[3] = VIMMED1(0.0f); + break; + case BLENDFACTOR_INV_SRC_COLOR: + out[0] = FSUB(VIMMED1(1.0f), src[0]); + out[1] = FSUB(VIMMED1(1.0f), src[1]); + out[2] = FSUB(VIMMED1(1.0f), src[2]); + out[3] = FSUB(VIMMED1(1.0f), src[3]); + break; + case BLENDFACTOR_INV_SRC_ALPHA: + out[0] = out[1] = out[2] = out[3] = FSUB(VIMMED1(1.0f), src[3]); + break; + case BLENDFACTOR_INV_DST_ALPHA: + out[0] = out[1] = out[2] = out[3] = FSUB(VIMMED1(1.0f), dst[3]); + break; + case BLENDFACTOR_INV_DST_COLOR: + out[0] = FSUB(VIMMED1(1.0f), dst[0]); + out[1] = FSUB(VIMMED1(1.0f), dst[1]); + out[2] = FSUB(VIMMED1(1.0f), dst[2]); + out[3] = FSUB(VIMMED1(1.0f), dst[3]); + break; + case BLENDFACTOR_INV_CONST_COLOR: + out[0] = FSUB(VIMMED1(1.0f), constColor[0]); + out[1] = FSUB(VIMMED1(1.0f), constColor[1]); + out[2] = FSUB(VIMMED1(1.0f), constColor[2]); + out[3] = FSUB(VIMMED1(1.0f), constColor[3]); + break; + case BLENDFACTOR_INV_CONST_ALPHA: + out[0] = out[1] = out[2] = out[3] = FSUB(VIMMED1(1.0f), constColor[3]); + break; + case BLENDFACTOR_INV_SRC1_COLOR: + out[0] = FSUB(VIMMED1(1.0f), src1[0]); + out[1] = FSUB(VIMMED1(1.0f), src1[1]); + out[2] = FSUB(VIMMED1(1.0f), src1[2]); + out[3] = FSUB(VIMMED1(1.0f), src1[3]); + break; + case BLENDFACTOR_INV_SRC1_ALPHA: + out[0] = out[1] = out[2] = out[3] = FSUB(VIMMED1(1.0f), src1[3]); + break; + default: + SWR_ASSERT(false, "Unsupported blend factor: %d", factor); + out[0] = out[1] = out[2] = out[3] = VIMMED1(0.0f); + break; + } + + if (Color) + { + result[0] = out[0]; + result[1] = out[1]; + result[2] = out[2]; + } + + if (Alpha) + { + result[3] = out[3]; + } + } + + void Clamp(SWR_FORMAT format, Value* src[4]) + { + const SWR_FORMAT_INFO& info = GetFormatInfo(format); + SWR_TYPE type = info.type[0]; + + switch (type) + { + case SWR_TYPE_FLOAT: + break; + + case SWR_TYPE_UNORM: + src[0] = VMINPS(VMAXPS(src[0], VIMMED1(0.0f)), VIMMED1(1.0f)); + src[1] = VMINPS(VMAXPS(src[1], VIMMED1(0.0f)), VIMMED1(1.0f)); + src[2] = VMINPS(VMAXPS(src[2], VIMMED1(0.0f)), VIMMED1(1.0f)); + src[3] = VMINPS(VMAXPS(src[3], VIMMED1(0.0f)), VIMMED1(1.0f)); + break; + + case SWR_TYPE_SNORM: + src[0] = VMINPS(VMAXPS(src[0], VIMMED1(-1.0f)), VIMMED1(1.0f)); + src[1] = VMINPS(VMAXPS(src[1], VIMMED1(-1.0f)), VIMMED1(1.0f)); + src[2] = VMINPS(VMAXPS(src[2], VIMMED1(-1.0f)), VIMMED1(1.0f)); + src[3] = VMINPS(VMAXPS(src[3], VIMMED1(-1.0f)), VIMMED1(1.0f)); + break; + + default: SWR_ASSERT(false, "Unsupport format type: %d", type); + } + } + + void ApplyDefaults(SWR_FORMAT format, Value* src[4]) + { + const SWR_FORMAT_INFO& info = GetFormatInfo(format); + + bool valid[] = { false, false, false, false }; + for (uint32_t c = 0; c < info.numComps; ++c) + { + valid[info.swizzle[c]] = true; + } + + for (uint32_t c = 0; c < 4; ++c) + { + if (!valid[c]) + { + src[c] = BITCAST(VIMMED1((int)info.defaults[c]), mSimdFP32Ty); + } + } + } + + void ApplyUnusedDefaults(SWR_FORMAT format, Value* src[4]) + { + const SWR_FORMAT_INFO& info = GetFormatInfo(format); + + for (uint32_t c = 0; c < info.numComps; ++c) + { + if (info.type[c] == SWR_TYPE_UNUSED) + { + src[info.swizzle[c]] = BITCAST(VIMMED1((int)info.defaults[info.swizzle[c]]), mSimdFP32Ty); + } + } + } + + void Quantize(SWR_FORMAT format, Value* src[4]) + { + const SWR_FORMAT_INFO& info = GetFormatInfo(format); + for (uint32_t c = 0; c < info.numComps; ++c) + { + if (info.bpc[c] <= QUANTIZE_THRESHOLD) + { + uint32_t swizComp = info.swizzle[c]; + float factor = (float)((1 << info.bpc[c]) - 1); + switch (info.type[c]) + { + case SWR_TYPE_UNORM: + src[swizComp] = FADD(FMUL(src[swizComp], VIMMED1(factor)), VIMMED1(0.5f)); + src[swizComp] = VROUND(src[swizComp], C(_MM_FROUND_TO_ZERO)); + src[swizComp] = FMUL(src[swizComp], VIMMED1(1.0f /factor)); + break; + default: SWR_ASSERT(false, "Unsupported format type: %d", info.type[c]); + } + } + } + } + + template + void BlendFunc(SWR_BLEND_OP blendOp, Value* src[4], Value* srcFactor[4], Value* dst[4], Value* dstFactor[4], Value* result[4]) + { + Value* out[4]; + Value* srcBlend[4]; + Value* dstBlend[4]; + for (uint32_t i = 0; i < 4; ++i) + { + srcBlend[i] = FMUL(src[i], srcFactor[i]); + dstBlend[i] = FMUL(dst[i], dstFactor[i]); + } + + switch (blendOp) + { + case BLENDOP_ADD: + out[0] = FADD(srcBlend[0], dstBlend[0]); + out[1] = FADD(srcBlend[1], dstBlend[1]); + out[2] = FADD(srcBlend[2], dstBlend[2]); + out[3] = FADD(srcBlend[3], dstBlend[3]); + break; + + case BLENDOP_SUBTRACT: + out[0] = FSUB(srcBlend[0], dstBlend[0]); + out[1] = FSUB(srcBlend[1], dstBlend[1]); + out[2] = FSUB(srcBlend[2], dstBlend[2]); + out[3] = FSUB(srcBlend[3], dstBlend[3]); + break; + + case BLENDOP_REVSUBTRACT: + out[0] = FSUB(dstBlend[0], srcBlend[0]); + out[1] = FSUB(dstBlend[1], srcBlend[1]); + out[2] = FSUB(dstBlend[2], srcBlend[2]); + out[3] = FSUB(dstBlend[3], srcBlend[3]); + break; + + case BLENDOP_MIN: + out[0] = VMINPS(src[0], dst[0]); + out[1] = VMINPS(src[1], dst[1]); + out[2] = VMINPS(src[2], dst[2]); + out[3] = VMINPS(src[3], dst[3]); + break; + + case BLENDOP_MAX: + out[0] = VMAXPS(src[0], dst[0]); + out[1] = VMAXPS(src[1], dst[1]); + out[2] = VMAXPS(src[2], dst[2]); + out[3] = VMAXPS(src[3], dst[3]); + break; + + default: + SWR_ASSERT(false, "Unsupported blend operation: %d", blendOp); + out[0] = out[1] = out[2] = out[3] = VIMMED1(0.0f); + break; + } + + if (Color) + { + result[0] = out[0]; + result[1] = out[1]; + result[2] = out[2]; + } + + if (Alpha) + { + result[3] = out[3]; + } + } + + void LogicOpFunc(SWR_LOGIC_OP logicOp, Value* src[4], Value* dst[4], Value* result[4]) + { + // Op: (s == PS output, d = RT contents) + switch(logicOp) + { + case LOGICOP_CLEAR: + result[0] = VIMMED1(0); + result[1] = VIMMED1(0); + result[2] = VIMMED1(0); + result[3] = VIMMED1(0); + break; + + case LOGICOP_NOR: + // ~(s | d) + result[0] = XOR(OR(src[0], dst[0]), VIMMED1(0xFFFFFFFF)); + result[1] = XOR(OR(src[1], dst[1]), VIMMED1(0xFFFFFFFF)); + result[2] = XOR(OR(src[2], dst[2]), VIMMED1(0xFFFFFFFF)); + result[3] = XOR(OR(src[3], dst[3]), VIMMED1(0xFFFFFFFF)); + break; + + case LOGICOP_AND_INVERTED: + // ~s & d + // todo: use avx andnot instr when I can find the intrinsic to call + result[0] = AND(XOR(src[0], VIMMED1(0xFFFFFFFF)), dst[0]); + result[1] = AND(XOR(src[1], VIMMED1(0xFFFFFFFF)), dst[1]); + result[2] = AND(XOR(src[2], VIMMED1(0xFFFFFFFF)), dst[2]); + result[3] = AND(XOR(src[3], VIMMED1(0xFFFFFFFF)), dst[3]); + break; + + case LOGICOP_COPY_INVERTED: + // ~s + result[0] = XOR(src[0], VIMMED1(0xFFFFFFFF)); + result[1] = XOR(src[1], VIMMED1(0xFFFFFFFF)); + result[2] = XOR(src[2], VIMMED1(0xFFFFFFFF)); + result[3] = XOR(src[3], VIMMED1(0xFFFFFFFF)); + break; + + case LOGICOP_AND_REVERSE: + // s & ~d + // todo: use avx andnot instr when I can find the intrinsic to call + result[0] = AND(XOR(dst[0], VIMMED1(0xFFFFFFFF)), src[0]); + result[1] = AND(XOR(dst[1], VIMMED1(0xFFFFFFFF)), src[1]); + result[2] = AND(XOR(dst[2], VIMMED1(0xFFFFFFFF)), src[2]); + result[3] = AND(XOR(dst[3], VIMMED1(0xFFFFFFFF)), src[3]); + break; + + case LOGICOP_INVERT: + // ~d + result[0] = XOR(dst[0], VIMMED1(0xFFFFFFFF)); + result[1] = XOR(dst[1], VIMMED1(0xFFFFFFFF)); + result[2] = XOR(dst[2], VIMMED1(0xFFFFFFFF)); + result[3] = XOR(dst[3], VIMMED1(0xFFFFFFFF)); + break; + + case LOGICOP_XOR: + // s ^ d + result[0] = XOR(src[0], dst[0]); + result[1] = XOR(src[1], dst[1]); + result[2] = XOR(src[2], dst[2]); + result[3] = XOR(src[3], dst[3]); + break; + + case LOGICOP_NAND: + // ~(s & d) + result[0] = XOR(AND(src[0], dst[0]), VIMMED1(0xFFFFFFFF)); + result[1] = XOR(AND(src[1], dst[1]), VIMMED1(0xFFFFFFFF)); + result[2] = XOR(AND(src[2], dst[2]), VIMMED1(0xFFFFFFFF)); + result[3] = XOR(AND(src[3], dst[3]), VIMMED1(0xFFFFFFFF)); + break; + + case LOGICOP_AND: + // s & d + result[0] = AND(src[0], dst[0]); + result[1] = AND(src[1], dst[1]); + result[2] = AND(src[2], dst[2]); + result[3] = AND(src[3], dst[3]); + break; + + case LOGICOP_EQUIV: + // ~(s ^ d) + result[0] = XOR(XOR(src[0], dst[0]), VIMMED1(0xFFFFFFFF)); + result[1] = XOR(XOR(src[1], dst[1]), VIMMED1(0xFFFFFFFF)); + result[2] = XOR(XOR(src[2], dst[2]), VIMMED1(0xFFFFFFFF)); + result[3] = XOR(XOR(src[3], dst[3]), VIMMED1(0xFFFFFFFF)); + break; + + case LOGICOP_NOOP: + result[0] = dst[0]; + result[1] = dst[1]; + result[2] = dst[2]; + result[3] = dst[3]; + break; + + case LOGICOP_OR_INVERTED: + // ~s | d + result[0] = OR(XOR(src[0], VIMMED1(0xFFFFFFFF)), dst[0]); + result[1] = OR(XOR(src[1], VIMMED1(0xFFFFFFFF)), dst[1]); + result[2] = OR(XOR(src[2], VIMMED1(0xFFFFFFFF)), dst[2]); + result[3] = OR(XOR(src[3], VIMMED1(0xFFFFFFFF)), dst[3]); + break; + + case LOGICOP_COPY: + result[0] = src[0]; + result[1] = src[1]; + result[2] = src[2]; + result[3] = src[3]; + break; + + case LOGICOP_OR_REVERSE: + // s | ~d + result[0] = OR(XOR(dst[0], VIMMED1(0xFFFFFFFF)), src[0]); + result[1] = OR(XOR(dst[1], VIMMED1(0xFFFFFFFF)), src[1]); + result[2] = OR(XOR(dst[2], VIMMED1(0xFFFFFFFF)), src[2]); + result[3] = OR(XOR(dst[3], VIMMED1(0xFFFFFFFF)), src[3]); + break; + + case LOGICOP_OR: + // s | d + result[0] = OR(src[0], dst[0]); + result[1] = OR(src[1], dst[1]); + result[2] = OR(src[2], dst[2]); + result[3] = OR(src[3], dst[3]); + break; + + case LOGICOP_SET: + result[0] = VIMMED1(0xFFFFFFFF); + result[1] = VIMMED1(0xFFFFFFFF); + result[2] = VIMMED1(0xFFFFFFFF); + result[3] = VIMMED1(0xFFFFFFFF); + break; + + default: + SWR_ASSERT(false, "Unsupported logic operation: %d", logicOp); + result[0] = result[1] = result[2] = result[3] = VIMMED1(0.0f); + break; + } + } + + void AlphaTest(const BLEND_COMPILE_STATE& state, Value* pBlendState, Value* pAlpha, Value* ppMask) + { + // load uint32_t reference + Value* pRef = VBROADCAST(LOAD(pBlendState, { 0, SWR_BLEND_STATE_alphaTestReference })); + + Value* pTest = nullptr; + if (state.alphaTestFormat == ALPHA_TEST_UNORM8) + { + // convert float alpha to unorm8 + Value* pAlphaU8 = FMUL(pAlpha, VIMMED1(256.0f)); + pAlphaU8 = FP_TO_UI(pAlphaU8, mSimdInt32Ty); + + // compare + switch (state.alphaTestFunction) + { + case ZFUNC_ALWAYS: pTest = VIMMED1(true); break; + case ZFUNC_NEVER: pTest = VIMMED1(false); break; + case ZFUNC_LT: pTest = ICMP_ULT(pAlphaU8, pRef); break; + case ZFUNC_EQ: pTest = ICMP_EQ(pAlphaU8, pRef); break; + case ZFUNC_LE: pTest = ICMP_ULE(pAlphaU8, pRef); break; + case ZFUNC_GT: pTest = ICMP_UGT(pAlphaU8, pRef); break; + case ZFUNC_NE: pTest = ICMP_NE(pAlphaU8, pRef); break; + case ZFUNC_GE: pTest = ICMP_UGE(pAlphaU8, pRef); break; + default: + SWR_ASSERT(false, "Invalid alpha test function"); + break; + } + } + else + { + // cast ref to float + pRef = BITCAST(pRef, mSimdFP32Ty); + + // compare + switch (state.alphaTestFunction) + { + case ZFUNC_ALWAYS: pTest = VIMMED1(true); break; + case ZFUNC_NEVER: pTest = VIMMED1(false); break; + case ZFUNC_LT: pTest = FCMP_OLT(pAlpha, pRef); break; + case ZFUNC_EQ: pTest = FCMP_OEQ(pAlpha, pRef); break; + case ZFUNC_LE: pTest = FCMP_OLE(pAlpha, pRef); break; + case ZFUNC_GT: pTest = FCMP_OGT(pAlpha, pRef); break; + case ZFUNC_NE: pTest = FCMP_ONE(pAlpha, pRef); break; + case ZFUNC_GE: pTest = FCMP_OGE(pAlpha, pRef); break; + default: + SWR_ASSERT(false, "Invalid alpha test function"); + break; + } + } + + // load current mask + Value* pMask = LOAD(ppMask); + + // convert to int1 mask + pMask = MASK(pMask); + + // and with alpha test result + pMask = AND(pMask, pTest); + + // convert back to vector mask + pMask = VMASK(pMask); + + // store new mask + STORE(pMask, ppMask); + } + + Function* Create(const BLEND_COMPILE_STATE& state) + { + static std::size_t jitNum = 0; + + std::stringstream fnName("BlendShader", std::ios_base::in | std::ios_base::out | std::ios_base::ate); + fnName << jitNum++; + + // blend function signature + //typedef void(*PFN_BLEND_JIT_FUNC)(const SWR_BLEND_STATE*, simdvector&, simdvector&, uint32_t, BYTE*, simdvector&, simdscalari*, simdscalari*); + + std::vector args{ + PointerType::get(Gen_SWR_BLEND_STATE(JM()), 0), // SWR_BLEND_STATE* + PointerType::get(mSimdFP32Ty, 0), // simdvector& src + PointerType::get(mSimdFP32Ty, 0), // simdvector& src1 + Type::getInt32Ty(JM()->mContext), // sampleNum + PointerType::get(mSimdFP32Ty, 0), // uint8_t* pDst + PointerType::get(mSimdFP32Ty, 0), // simdvector& result + PointerType::get(mSimdInt32Ty, 0), // simdscalari* oMask + PointerType::get(mSimdInt32Ty, 0), // simdscalari* pMask + }; + + FunctionType* fTy = FunctionType::get(IRB()->getVoidTy(), args, false); + Function* blendFunc = Function::Create(fTy, GlobalValue::ExternalLinkage, fnName.str(), JM()->mpCurrentModule); + + BasicBlock* entry = BasicBlock::Create(JM()->mContext, "entry", blendFunc); + + IRB()->SetInsertPoint(entry); + + // arguments + auto argitr = blendFunc->getArgumentList().begin(); + Value* pBlendState = &*argitr++; + pBlendState->setName("pBlendState"); + Value* pSrc = &*argitr++; + pSrc->setName("src"); + Value* pSrc1 = &*argitr++; + pSrc1->setName("src1"); + Value* sampleNum = &*argitr++; + sampleNum->setName("sampleNum"); + Value* pDst = &*argitr++; + pDst->setName("pDst"); + Value* pResult = &*argitr++; + pResult->setName("result"); + Value* ppoMask = &*argitr++; + ppoMask->setName("ppoMask"); + Value* ppMask = &*argitr++; + ppMask->setName("pMask"); + + static_assert(KNOB_COLOR_HOT_TILE_FORMAT == R32G32B32A32_FLOAT, "Unsupported hot tile format"); + Value* dst[4]; + Value* constantColor[4]; + Value* src[4]; + Value* src1[4]; + Value* result[4]; + for (uint32_t i = 0; i < 4; ++i) + { + // load hot tile + dst[i] = LOAD(pDst, { i }); + + // load constant color + constantColor[i] = VBROADCAST(LOAD(pBlendState, { 0, SWR_BLEND_STATE_constantColor, i })); + + // load src + src[i] = LOAD(pSrc, { i }); + + // load src1 + src1[i] = LOAD(pSrc1, { i }); + } + Value* currentMask = VIMMED1(-1); + if(state.desc.alphaToCoverageEnable) + { + currentMask = FP_TO_SI(FMUL(src[3], VBROADCAST(C((float)state.desc.numSamples))), mSimdInt32Ty); + } + + // alpha test + if (state.desc.alphaTestEnable) + { + AlphaTest(state, pBlendState, src[3], ppMask); + } + + // color blend + if (state.blendState.blendEnable) + { + // clamp sources + Clamp(state.format, src); + Clamp(state.format, src1); + Clamp(state.format, dst); + Clamp(state.format, constantColor); + + // apply defaults to hottile contents to take into account missing components + ApplyDefaults(state.format, dst); + + // Force defaults for unused 'X' components + ApplyUnusedDefaults(state.format, dst); + + // Quantize low precision components + Quantize(state.format, dst); + + // special case clamping for R11G11B10_float which has no sign bit + if (state.format == R11G11B10_FLOAT) + { + dst[0] = VMAXPS(dst[0], VIMMED1(0.0f)); + dst[1] = VMAXPS(dst[1], VIMMED1(0.0f)); + dst[2] = VMAXPS(dst[2], VIMMED1(0.0f)); + dst[3] = VMAXPS(dst[3], VIMMED1(0.0f)); + } + + Value* srcFactor[4]; + Value* dstFactor[4]; + if (state.desc.independentAlphaBlendEnable) + { + GenerateBlendFactor(state.blendState.sourceBlendFactor, constantColor, src, src1, dst, srcFactor); + GenerateBlendFactor(state.blendState.sourceAlphaBlendFactor, constantColor, src, src1, dst, srcFactor); + + GenerateBlendFactor(state.blendState.destBlendFactor, constantColor, src, src1, dst, dstFactor); + GenerateBlendFactor(state.blendState.destAlphaBlendFactor, constantColor, src, src1, dst, dstFactor); + + BlendFunc(state.blendState.colorBlendFunc, src, srcFactor, dst, dstFactor, result); + BlendFunc(state.blendState.alphaBlendFunc, src, srcFactor, dst, dstFactor, result); + } + else + { + GenerateBlendFactor(state.blendState.sourceBlendFactor, constantColor, src, src1, dst, srcFactor); + GenerateBlendFactor(state.blendState.destBlendFactor, constantColor, src, src1, dst, dstFactor); + + BlendFunc(state.blendState.colorBlendFunc, src, srcFactor, dst, dstFactor, result); + } + + // store results out + for (uint32_t i = 0; i < 4; ++i) + { + STORE(result[i], pResult, { i }); + } + } + + if(state.blendState.logicOpEnable) + { + const SWR_FORMAT_INFO& info = GetFormatInfo(state.format); + SWR_ASSERT(info.type[0] == SWR_TYPE_UINT); + Value* vMask[4]; + for(uint32_t i = 0; i < 4; i++) + { + switch(info.bpc[i]) + { + case 0: vMask[i] = VIMMED1(0x00000000); break; + case 2: vMask[i] = VIMMED1(0x00000003); break; + case 5: vMask[i] = VIMMED1(0x0000001F); break; + case 6: vMask[i] = VIMMED1(0x0000003F); break; + case 8: vMask[i] = VIMMED1(0x000000FF); break; + case 10: vMask[i] = VIMMED1(0x000003FF); break; + case 11: vMask[i] = VIMMED1(0x000007FF); break; + case 16: vMask[i] = VIMMED1(0x0000FFFF); break; + case 24: vMask[i] = VIMMED1(0x00FFFFFF); break; + case 32: vMask[i] = VIMMED1(0xFFFFFFFF); break; + default: + vMask[i] = VIMMED1(0x0); + SWR_ASSERT(0, "Unsupported bpc for logic op\n"); + break; + } + src[i] = BITCAST(src[i], mSimdInt32Ty);//, vMask[i]); + dst[i] = BITCAST(dst[i], mSimdInt32Ty); + } + + LogicOpFunc(state.blendState.logicOpFunc, src, dst, result); + + // store results out + for(uint32_t i = 0; i < 4; ++i) + { + // clear upper bits from PS output not in RT format after doing logic op + result[i] = AND(result[i], vMask[i]); + + STORE(BITCAST(result[i], mSimdFP32Ty), pResult, {i}); + } + } + + if(state.desc.oMaskEnable) + { + assert(!(state.desc.alphaToCoverageEnable)); + // load current mask + Value* oMask = LOAD(ppoMask); + Value* sampleMasked = VBROADCAST(SHL(C(1), sampleNum)); + oMask = AND(oMask, sampleMasked); + currentMask = AND(oMask, currentMask); + } + + if(state.desc.sampleMaskEnable) + { + Value* sampleMask = LOAD(pBlendState, { 0, SWR_BLEND_STATE_sampleMask}); + Value* sampleMasked = SHL(C(1), sampleNum); + sampleMask = AND(sampleMask, sampleMasked); + sampleMask = VBROADCAST(ICMP_SGT(sampleMask, C(0))); + sampleMask = S_EXT(sampleMask, mSimdInt32Ty); + currentMask = AND(sampleMask, currentMask); + } + + if(state.desc.sampleMaskEnable || state.desc.alphaToCoverageEnable || + state.desc.oMaskEnable) + { + // load current mask + Value* pMask = LOAD(ppMask); + currentMask = S_EXT(ICMP_SGT(currentMask, VBROADCAST(C(0))), mSimdInt32Ty); + Value* outputMask = AND(pMask, currentMask); + // store new mask + STORE(outputMask, GEP(ppMask, C(0))); + } + + RET_VOID(); + + JitManager::DumpToFile(blendFunc, ""); + + FunctionPassManager passes(JM()->mpCurrentModule); + passes.add(createBreakCriticalEdgesPass()); + passes.add(createCFGSimplificationPass()); + passes.add(createEarlyCSEPass()); + passes.add(createPromoteMemoryToRegisterPass()); + passes.add(createCFGSimplificationPass()); + passes.add(createEarlyCSEPass()); + passes.add(createInstructionCombiningPass()); + passes.add(createInstructionSimplifierPass()); + passes.add(createConstantPropagationPass()); + passes.add(createSCCPPass()); + passes.add(createAggressiveDCEPass()); + + passes.run(*blendFunc); + + JitManager::DumpToFile(blendFunc, "optimized"); + + return blendFunc; + } +}; + +////////////////////////////////////////////////////////////////////////// +/// @brief JITs from fetch shader IR +/// @param hJitMgr - JitManager handle +/// @param func - LLVM function IR +/// @return PFN_FETCH_FUNC - pointer to fetch code +PFN_BLEND_JIT_FUNC JitBlendFunc(HANDLE hJitMgr, const HANDLE hFunc) +{ + const llvm::Function *func = (const llvm::Function*)hFunc; + JitManager* pJitMgr = reinterpret_cast(hJitMgr); + PFN_BLEND_JIT_FUNC pfnBlend; + pfnBlend = (PFN_BLEND_JIT_FUNC)(pJitMgr->mpExec->getFunctionAddress(func->getName().str())); + // MCJIT finalizes modules the first time you JIT code from them. After finalized, you cannot add new IR to the module + pJitMgr->mIsModuleFinalized = true; + + return pfnBlend; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief JIT compiles blend shader +/// @param hJitMgr - JitManager handle +/// @param state - blend state to build function from +extern "C" PFN_BLEND_JIT_FUNC JITCALL JitCompileBlend(HANDLE hJitMgr, const BLEND_COMPILE_STATE& state) +{ + JitManager* pJitMgr = reinterpret_cast(hJitMgr); + + pJitMgr->SetupNewModule(); + + BlendJit theJit(pJitMgr); + HANDLE hFunc = theJit.Create(state); + + return JitBlendFunc(hJitMgr, hFunc); +} diff --git a/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.h b/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.h new file mode 100644 index 00000000000..057eb92b67e --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/blend_jit.h @@ -0,0 +1,93 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file blend_jit.h +* +* @brief Definition of the blend jitter +* +* Notes: +* +******************************************************************************/ +#pragma once + +#include "common/formats.h" +#include "core/context.h" +#include "core/state.h" + +struct RENDER_TARGET_BLEND_COMPILE_STATE +{ + bool blendEnable; + bool logicOpEnable; + SWR_BLEND_FACTOR sourceAlphaBlendFactor; + SWR_BLEND_FACTOR destAlphaBlendFactor; + SWR_BLEND_FACTOR sourceBlendFactor; + SWR_BLEND_FACTOR destBlendFactor; + SWR_BLEND_OP colorBlendFunc; + SWR_BLEND_OP alphaBlendFunc; + SWR_LOGIC_OP logicOpFunc; +}; + +enum ALPHA_TEST_FORMAT +{ + ALPHA_TEST_UNORM8, + ALPHA_TEST_FLOAT32 +}; + +////////////////////////////////////////////////////////////////////////// +/// BLEND_DESC +////////////////////////////////////////////////////////////////////////// +struct BLEND_DESC +{ + union + { + struct + { + uint32_t alphaTestEnable: 1; + uint32_t independentAlphaBlendEnable: 1; + uint32_t alphaToCoverageEnable: 1; + uint32_t oMaskEnable:1; + uint32_t inputCoverageEnable:1; + uint32_t sampleMaskEnable:1; + uint32_t numSamples:5; + uint32_t _reserved : 21; + }; + uint32_t bits; + }; +}; +#define BLEND_ENABLE_MASK 0x3D // a2c | oMaskEnable | inputCoverageEnable | sampleMaskEnable +////////////////////////////////////////////////////////////////////////// +/// State required for blend jit +////////////////////////////////////////////////////////////////////////// +struct BLEND_COMPILE_STATE +{ + SWR_FORMAT format; // format of render target being blended + RENDER_TARGET_BLEND_COMPILE_STATE blendState; + BLEND_DESC desc; + + SWR_ZFUNCTION alphaTestFunction; + ALPHA_TEST_FORMAT alphaTestFormat; + + bool operator==(const BLEND_COMPILE_STATE& other) const + { + return memcmp(this, &other, sizeof(BLEND_COMPILE_STATE)) == 0; + } +}; diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder.cpp b/src/gallium/drivers/swr/rasterizer/jitter/builder.cpp new file mode 100644 index 00000000000..c15bdf1e756 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/builder.cpp @@ -0,0 +1,71 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file builder.h +* +* @brief Includes all the builder related functionality +* +* Notes: +* +******************************************************************************/ + +#include "builder.h" + +using namespace llvm; + +////////////////////////////////////////////////////////////////////////// +/// @brief Contructor for Builder. +/// @param pJitMgr - JitManager which contains modules, function passes, etc. +Builder::Builder(JitManager *pJitMgr) + : mpJitMgr(pJitMgr) +{ + mpIRBuilder = &pJitMgr->mBuilder; + + mVoidTy = Type::getVoidTy(pJitMgr->mContext); + mFP16Ty = Type::getHalfTy(pJitMgr->mContext); + mFP32Ty = Type::getFloatTy(pJitMgr->mContext); + mDoubleTy = Type::getDoubleTy(pJitMgr->mContext); + mInt1Ty = Type::getInt1Ty(pJitMgr->mContext); + mInt8Ty = Type::getInt8Ty(pJitMgr->mContext); + mInt16Ty = Type::getInt16Ty(pJitMgr->mContext); + mInt32Ty = Type::getInt32Ty(pJitMgr->mContext); + mInt64Ty = Type::getInt64Ty(pJitMgr->mContext); + mV4FP32Ty = StructType::get(pJitMgr->mContext, std::vector(4, mFP32Ty), false); // vector4 float type (represented as structure) + mV4Int32Ty = StructType::get(pJitMgr->mContext, std::vector(4, mInt32Ty), false); // vector4 int type + mSimdInt16Ty = VectorType::get(mInt16Ty, mpJitMgr->mVWidth); + mSimdInt32Ty = VectorType::get(mInt32Ty, mpJitMgr->mVWidth); + mSimdInt64Ty = VectorType::get(mInt64Ty, mpJitMgr->mVWidth); + mSimdFP16Ty = VectorType::get(mFP16Ty, mpJitMgr->mVWidth); + mSimdFP32Ty = VectorType::get(mFP32Ty, mpJitMgr->mVWidth); + + if (sizeof(uint32_t*) == 4) + { + mIntPtrTy = mInt32Ty; + mSimdIntPtrTy = mSimdInt32Ty; + } + else + { + SWR_ASSERT(sizeof(uint32_t*) == 8); + mIntPtrTy = mInt64Ty; + mSimdIntPtrTy = mSimdInt64Ty; + } +} diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder.h b/src/gallium/drivers/swr/rasterizer/jitter/builder.h new file mode 100644 index 00000000000..49216612cc9 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/builder.h @@ -0,0 +1,71 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file builder.h +* +* @brief Includes all the builder related functionality +* +* Notes: +* +******************************************************************************/ +#pragma once + +#include "JitManager.h" +#include "common/formats.h" + +using namespace llvm; + +struct Builder +{ + Builder(JitManager *pJitMgr); + IRBuilder<>* IRB() { return mpIRBuilder; }; + JitManager* JM() { return mpJitMgr; } + + JitManager* mpJitMgr; + IRBuilder<>* mpIRBuilder; + + // Built in types. + Type* mVoidTy; + Type* mInt1Ty; + Type* mInt8Ty; + Type* mInt16Ty; + Type* mInt32Ty; + Type* mInt64Ty; + Type* mIntPtrTy; + Type* mFP16Ty; + Type* mFP32Ty; + Type* mDoubleTy; + Type* mSimdFP16Ty; + Type* mSimdFP32Ty; + Type* mSimdInt16Ty; + Type* mSimdInt32Ty; + Type* mSimdInt64Ty; + Type* mSimdIntPtrTy; + StructType* mV4FP32Ty; + StructType* mV4Int32Ty; + +#include "builder_gen.h" +#include "builder_x86.h" +#include "builder_misc.h" +#include "builder_math.h" + +}; diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_math.h b/src/gallium/drivers/swr/rasterizer/jitter/builder_math.h new file mode 100644 index 00000000000..92867ec9836 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_math.h @@ -0,0 +1,34 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file builder_math.h +* +* @brief math/alu builder functions +* +* Notes: +* +******************************************************************************/ +#pragma once + +Value* VLOG2PS(Value* src); +Value* VPOW24PS(Value* src); +Value* VEXP2PS(Value* src); diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp new file mode 100644 index 00000000000..5394fc7bf5a --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.cpp @@ -0,0 +1,1447 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file builder_misc.cpp +* +* @brief Implementation for miscellaneous builder functions +* +* Notes: +* +******************************************************************************/ +#include "builder.h" +#include "llvm/Support/DynamicLibrary.h" + +void __cdecl CallPrint(const char* fmt, ...); + +////////////////////////////////////////////////////////////////////////// +/// @brief Convert an IEEE 754 32-bit single precision float to an +/// 16 bit float with 5 exponent bits and a variable +/// number of mantissa bits. +/// @param val - 32-bit float +/// @todo Maybe move this outside of this file into a header? +static uint16_t Convert32To16Float(float val) +{ + uint32_t sign, exp, mant; + uint32_t roundBits; + + // Extract the sign, exponent, and mantissa + uint32_t uf = *(uint32_t*)&val; + sign = (uf & 0x80000000) >> 31; + exp = (uf & 0x7F800000) >> 23; + mant = uf & 0x007FFFFF; + + // Check for out of range + if (std::isnan(val)) + { + exp = 0x1F; + mant = 0x200; + sign = 1; // set the sign bit for NANs + } + else if (std::isinf(val)) + { + exp = 0x1f; + mant = 0x0; + } + else if (exp > (0x70 + 0x1E)) // Too big to represent -> max representable value + { + exp = 0x1E; + mant = 0x3FF; + } + else if ((exp <= 0x70) && (exp >= 0x66)) // It's a denorm + { + mant |= 0x00800000; + for (; exp <= 0x70; mant >>= 1, exp++) + ; + exp = 0; + mant = mant >> 13; + } + else if (exp < 0x66) // Too small to represent -> Zero + { + exp = 0; + mant = 0; + } + else + { + // Saves bits that will be shifted off for rounding + roundBits = mant & 0x1FFFu; + // convert exponent and mantissa to 16 bit format + exp = exp - 0x70; + mant = mant >> 13; + + // Essentially RTZ, but round up if off by only 1 lsb + if (roundBits == 0x1FFFu) + { + mant++; + // check for overflow + if ((mant & 0xC00u) != 0) + exp++; + // make sure only the needed bits are used + mant &= 0x3FF; + } + } + + uint32_t tmpVal = (sign << 15) | (exp << 10) | mant; + return (uint16_t)tmpVal; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Convert an IEEE 754 16-bit float to an 32-bit single precision +/// float +/// @param val - 16-bit float +/// @todo Maybe move this outside of this file into a header? +static float ConvertSmallFloatTo32(UINT val) +{ + UINT result; + if ((val & 0x7fff) == 0) + { + result = ((uint32_t)(val & 0x8000)) << 16; + } + else if ((val & 0x7c00) == 0x7c00) + { + result = ((val & 0x3ff) == 0) ? 0x7f800000 : 0x7fc00000; + result |= ((uint32_t)val & 0x8000) << 16; + } + else + { + uint32_t sign = (val & 0x8000) << 16; + uint32_t mant = (val & 0x3ff) << 13; + uint32_t exp = (val >> 10) & 0x1f; + if ((exp == 0) && (mant != 0)) // Adjust exponent and mantissa for denormals + { + mant <<= 1; + while (mant < (0x400 << 13)) + { + exp--; + mant <<= 1; + } + mant &= (0x3ff << 13); + } + exp = ((exp - 15 + 127) & 0xff) << 23; + result = sign | exp | mant; + } + + return *(float*)&result; +} + +Constant *Builder::C(bool i) +{ + return ConstantInt::get(IRB()->getInt1Ty(), (i ? 1 : 0)); +} + +Constant *Builder::C(char i) +{ + return ConstantInt::get(IRB()->getInt8Ty(), i); +} + +Constant *Builder::C(uint8_t i) +{ + return ConstantInt::get(IRB()->getInt8Ty(), i); +} + +Constant *Builder::C(int i) +{ + return ConstantInt::get(IRB()->getInt32Ty(), i); +} + +Constant *Builder::C(int64_t i) +{ + return ConstantInt::get(IRB()->getInt64Ty(), i); +} + +Constant *Builder::C(uint16_t i) +{ + return ConstantInt::get(mInt16Ty,i); +} + +Constant *Builder::C(uint32_t i) +{ + return ConstantInt::get(IRB()->getInt32Ty(), i); +} + +Constant *Builder::C(float i) +{ + return ConstantFP::get(IRB()->getFloatTy(), i); +} + +Constant *Builder::PRED(bool pred) +{ + return ConstantInt::get(IRB()->getInt1Ty(), (pred ? 1 : 0)); +} + +Value *Builder::VIMMED1(int i) +{ + return ConstantVector::getSplat(JM()->mVWidth, cast(C(i))); +} + +Value *Builder::VIMMED1(uint32_t i) +{ + return ConstantVector::getSplat(JM()->mVWidth, cast(C(i))); +} + +Value *Builder::VIMMED1(float i) +{ + return ConstantVector::getSplat(JM()->mVWidth, cast(C(i))); +} + +Value *Builder::VIMMED1(bool i) +{ + return ConstantVector::getSplat(JM()->mVWidth, cast(C(i))); +} + +Value *Builder::VUNDEF_IPTR() +{ + return UndefValue::get(VectorType::get(PointerType::get(mInt32Ty, 0),JM()->mVWidth)); +} + +Value *Builder::VUNDEF_I() +{ + return UndefValue::get(VectorType::get(mInt32Ty, JM()->mVWidth)); +} + +Value *Builder::VUNDEF(Type *ty, uint32_t size) +{ + return UndefValue::get(VectorType::get(ty, size)); +} + +Value *Builder::VUNDEF_F() +{ + return UndefValue::get(VectorType::get(mFP32Ty, JM()->mVWidth)); +} + +Value *Builder::VUNDEF(Type* t) +{ + return UndefValue::get(VectorType::get(t, JM()->mVWidth)); +} + +#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 6 +Value *Builder::VINSERT(Value *vec, Value *val, uint64_t index) +{ + return VINSERT(vec, val, C((int64_t)index)); +} +#endif + +Value *Builder::VBROADCAST(Value *src) +{ + // check if src is already a vector + if (src->getType()->isVectorTy()) + { + return src; + } + + return VECTOR_SPLAT(JM()->mVWidth, src); +} + +uint32_t Builder::IMMED(Value* v) +{ + SWR_ASSERT(isa(v)); + ConstantInt *pValConst = cast(v); + return pValConst->getZExtValue(); +} + +Value *Builder::GEP(Value* ptr, const std::initializer_list &indexList) +{ + std::vector indices; + for (auto i : indexList) + indices.push_back(i); + return GEPA(ptr, indices); +} + +Value *Builder::GEP(Value* ptr, const std::initializer_list &indexList) +{ + std::vector indices; + for (auto i : indexList) + indices.push_back(C(i)); + return GEPA(ptr, indices); +} + +LoadInst *Builder::LOAD(Value *basePtr, const std::initializer_list &indices, const llvm::Twine& name) +{ + std::vector valIndices; + for (auto i : indices) + valIndices.push_back(C(i)); + return LOAD(GEPA(basePtr, valIndices), name); +} + +LoadInst *Builder::LOADV(Value *basePtr, const std::initializer_list &indices, const llvm::Twine& name) +{ + std::vector valIndices; + for (auto i : indices) + valIndices.push_back(i); + return LOAD(GEPA(basePtr, valIndices), name); +} + +StoreInst *Builder::STORE(Value *val, Value *basePtr, const std::initializer_list &indices) +{ + std::vector valIndices; + for (auto i : indices) + valIndices.push_back(C(i)); + return STORE(val, GEPA(basePtr, valIndices)); +} + +StoreInst *Builder::STOREV(Value *val, Value *basePtr, const std::initializer_list &indices) +{ + std::vector valIndices; + for (auto i : indices) + valIndices.push_back(i); + return STORE(val, GEPA(basePtr, valIndices)); +} + +CallInst *Builder::CALL(Value *Callee, const std::initializer_list &argsList) +{ + std::vector args; + for (auto arg : argsList) + args.push_back(arg); + return CALLA(Callee, args); +} + +Value *Builder::VRCP(Value *va) +{ + return FDIV(VIMMED1(1.0f), va); // 1 / a +} + +Value *Builder::VPLANEPS(Value* vA, Value* vB, Value* vC, Value* &vX, Value* &vY) +{ + Value* vOut = FMADDPS(vA, vX, vC); + vOut = FMADDPS(vB, vY, vOut); + return vOut; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Generate an i32 masked load operation in LLVM IR. If not +/// supported on the underlying platform, emulate it with float masked load +/// @param src - base address pointer for the load +/// @param vMask - SIMD wide mask that controls whether to access memory load 0 +Value *Builder::MASKLOADD(Value* src,Value* mask) +{ + Value* vResult; + // use avx2 gather instruction is available + if(JM()->mArch.AVX2()) + { + Function *func = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::x86_avx2_maskload_d_256); + vResult = CALL(func,{src,mask}); + } + else + { + Function *func = Intrinsic::getDeclaration(JM()->mpCurrentModule,Intrinsic::x86_avx_maskload_ps_256); + Value* fMask = BITCAST(mask,VectorType::get(mFP32Ty,JM()->mVWidth)); + vResult = BITCAST(CALL(func,{src,fMask}), VectorType::get(mInt32Ty,JM()->mVWidth)); + } + return vResult; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief insert a JIT call to CallPrint +/// - outputs formatted string to both stdout and VS output window +/// - DEBUG builds only +/// Usage example: +/// PRINT("index %d = 0x%p\n",{C(lane), pIndex}); +/// where C(lane) creates a constant value to print, and pIndex is the Value* +/// result from a GEP, printing out the pointer to memory +/// @param printStr - constant string to print, which includes format specifiers +/// @param printArgs - initializer list of Value*'s to print to std out +CallInst *Builder::PRINT(const std::string &printStr,const std::initializer_list &printArgs) +{ + // push the arguments to CallPrint into a vector + std::vector printCallArgs; + // save room for the format string. we still need to modify it for vectors + printCallArgs.resize(1); + + // search through the format string for special processing + size_t pos = 0; + std::string tempStr(printStr); + pos = tempStr.find('%', pos); + auto v = printArgs.begin(); + + while ((pos != std::string::npos) && (v != printArgs.end())) + { + Value* pArg = *v; + Type* pType = pArg->getType(); + + if (tempStr[pos + 1] == 't') + { + if (pType->isVectorTy()) + { + Type* pContainedType = pType->getContainedType(0); + + std::string vectorFormatStr; + + if (pContainedType->isFloatTy()) + { + tempStr[pos + 1] = 'f'; // Ensure its %f + printCallArgs.push_back(FP_EXT(VEXTRACT(pArg, C(0)), mDoubleTy)); + + for (uint32_t i = 1; i < pType->getVectorNumElements(); ++i) + { + vectorFormatStr += "%f "; + printCallArgs.push_back(FP_EXT(VEXTRACT(pArg, C(i)), mDoubleTy)); + } + } + else if (pContainedType->isIntegerTy()) + { + tempStr[pos + 1] = 'd'; // Ensure its %d + printCallArgs.push_back(VEXTRACT(pArg, C(0))); + + for (uint32_t i = 1; i < pType->getVectorNumElements(); ++i) + { + vectorFormatStr += "%d "; + printCallArgs.push_back(VEXTRACT(pArg, C(i))); + } + } + else + { + SWR_ASSERT(0, "Unsupported tyep"); + } + + tempStr.insert(pos, vectorFormatStr); + pos += vectorFormatStr.size(); + } + else + { + if (pType->isFloatTy()) + { + tempStr[pos + 1] = 'f'; // Ensure its %f + printCallArgs.push_back(FP_EXT(pArg, mDoubleTy)); + } + else if (pType->isIntegerTy()) + { + tempStr[pos + 1] = 'd'; // Ensure its %d + printCallArgs.push_back(pArg); + } + } + } + else if (toupper(tempStr[pos + 1]) == 'X') + { + if (pType->isVectorTy()) + { + tempStr[pos] = '0'; + tempStr.insert(pos + 1, "x%08"); + + printCallArgs.push_back(VEXTRACT(pArg, C(0))); + + std::string vectorFormatStr; + for (uint32_t i = 1; i < pType->getVectorNumElements(); ++i) + { + vectorFormatStr += "0x%08X "; + printCallArgs.push_back(VEXTRACT(pArg, C(i))); + } + + tempStr.insert(pos, vectorFormatStr); + pos += vectorFormatStr.size(); + } + else + { + tempStr[pos] = '0'; + tempStr.insert(pos + 1, "x%08"); + printCallArgs.push_back(pArg); + pos += 3; + } + } + // for %f we need to cast float Values to doubles so that they print out correctly + else if ((tempStr[pos + 1] == 'f') && (pType->isFloatTy())) + { + printCallArgs.push_back(FP_EXT(pArg, Type::getDoubleTy(JM()->mContext))); + pos++; + } + // add special handling for %f and %d format specifiers to make printing llvm vector types easier + else if (pType->isVectorTy()) + { + Type* pContainedType = pType->getContainedType(0); + + if ((tempStr[pos + 1] == 'f') && (pContainedType->isFloatTy())) + { + uint32_t i = 0; + for (; i < (pArg->getType()->getVectorNumElements()) - 1; i++) + { + tempStr.insert(pos, std::string("%f ")); + pos += 3; + printCallArgs.push_back(FP_EXT(VEXTRACT(pArg, C(i)), Type::getDoubleTy(JM()->mContext))); + } + printCallArgs.push_back(FP_EXT(VEXTRACT(pArg, C(i)), Type::getDoubleTy(JM()->mContext))); + } + else if ((tempStr[pos + 1] == 'd') && (pContainedType->isIntegerTy())) + { + uint32_t i = 0; + for (; i < (pArg->getType()->getVectorNumElements()) - 1; i++) + { + tempStr.insert(pos, std::string("%d ")); + pos += 3; + printCallArgs.push_back(VEXTRACT(pArg, C(i))); + } + printCallArgs.push_back(VEXTRACT(pArg, C(i))); + } + else + { + /// not a supported vector to print + /// @todo pointer types too + SWR_ASSERT(0); + } + } + else + { + printCallArgs.push_back(pArg); + } + + // advance to the next arguement + v++; + pos = tempStr.find('%', ++pos); + } + + // create global variable constant string + Constant *constString = ConstantDataArray::getString(JM()->mContext,tempStr,true); + GlobalVariable *gvPtr = new GlobalVariable(constString->getType(),true,GlobalValue::InternalLinkage,constString,"printStr"); + JM()->mpCurrentModule->getGlobalList().push_back(gvPtr); + + // get a pointer to the first character in the constant string array + std::vector geplist{C(0),C(0)}; +#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 6 + Constant *strGEP = ConstantExpr::getGetElementPtr(gvPtr,geplist,false); +#else + Constant *strGEP = ConstantExpr::getGetElementPtr(nullptr, gvPtr,geplist,false); +#endif + + // insert the pointer to the format string in the argument vector + printCallArgs[0] = strGEP; + + // get pointer to CallPrint function and insert decl into the module if needed + std::vector args; + args.push_back(PointerType::get(mInt8Ty,0)); + FunctionType* callPrintTy = FunctionType::get(Type::getVoidTy(JM()->mContext),args,true); + Function *callPrintFn = cast(JM()->mpCurrentModule->getOrInsertFunction("CallPrint", callPrintTy)); + + // if we haven't yet added the symbol to the symbol table + if((sys::DynamicLibrary::SearchForAddressOfSymbol("CallPrint")) == nullptr) + { + sys::DynamicLibrary::AddSymbol("CallPrint", (void *)&CallPrint); + } + + // insert a call to CallPrint + return CALLA(callPrintFn,printCallArgs); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Wrapper around PRINT with initializer list. +CallInst* Builder::PRINT(const std::string &printStr) +{ + return PRINT(printStr, {}); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Generate a masked gather operation in LLVM IR. If not +/// supported on the underlying platform, emulate it with loads +/// @param vSrc - SIMD wide value that will be loaded if mask is invalid +/// @param pBase - Int8* base VB address pointer value +/// @param vIndices - SIMD wide value of VB byte offsets +/// @param vMask - SIMD wide mask that controls whether to access memory or the src values +/// @param scale - value to scale indices by +Value *Builder::GATHERPS(Value* vSrc, Value* pBase, Value* vIndices, Value* vMask, Value* scale) +{ + Value* vGather; + + // use avx2 gather instruction if available + if(JM()->mArch.AVX2()) + { + // force mask to , required by vgather + vMask = BITCAST(vMask, mSimdFP32Ty); + vGather = VGATHERPS(vSrc,pBase,vIndices,vMask,scale); + } + else + { + Value* pStack = STACKSAVE(); + + // store vSrc on the stack. this way we can select between a valid load address and the vSrc address + Value* vSrcPtr = ALLOCA(vSrc->getType()); + STORE(vSrc, vSrcPtr); + + vGather = VUNDEF_F(); + Value *vScaleVec = VBROADCAST(Z_EXT(scale,mInt32Ty)); + Value *vOffsets = MUL(vIndices,vScaleVec); + Value *mask = MASK(vMask); + for(uint32_t i = 0; i < JM()->mVWidth; ++i) + { + // single component byte index + Value *offset = VEXTRACT(vOffsets,C(i)); + // byte pointer to component + Value *loadAddress = GEP(pBase,offset); + loadAddress = BITCAST(loadAddress,PointerType::get(mFP32Ty,0)); + // pointer to the value to load if we're masking off a component + Value *maskLoadAddress = GEP(vSrcPtr,{C(0), C(i)}); + Value *selMask = VEXTRACT(mask,C(i)); + // switch in a safe address to load if we're trying to access a vertex + Value *validAddress = SELECT(selMask, loadAddress, maskLoadAddress); + Value *val = LOAD(validAddress); + vGather = VINSERT(vGather,val,C(i)); + } + STACKRESTORE(pStack); + } + + return vGather; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Generate a masked gather operation in LLVM IR. If not +/// supported on the underlying platform, emulate it with loads +/// @param vSrc - SIMD wide value that will be loaded if mask is invalid +/// @param pBase - Int8* base VB address pointer value +/// @param vIndices - SIMD wide value of VB byte offsets +/// @param vMask - SIMD wide mask that controls whether to access memory or the src values +/// @param scale - value to scale indices by +Value *Builder::GATHERDD(Value* vSrc, Value* pBase, Value* vIndices, Value* vMask, Value* scale) +{ + Value* vGather; + + // use avx2 gather instruction if available + if(JM()->mArch.AVX2()) + { + vGather = VGATHERDD(vSrc, pBase, vIndices, vMask, scale); + } + else + { + Value* pStack = STACKSAVE(); + + // store vSrc on the stack. this way we can select between a valid load address and the vSrc address + Value* vSrcPtr = ALLOCA(vSrc->getType()); + STORE(vSrc, vSrcPtr); + + vGather = VUNDEF_I(); + Value *vScaleVec = VBROADCAST(Z_EXT(scale, mInt32Ty)); + Value *vOffsets = MUL(vIndices, vScaleVec); + Value *mask = MASK(vMask); + for(uint32_t i = 0; i < JM()->mVWidth; ++i) + { + // single component byte index + Value *offset = VEXTRACT(vOffsets, C(i)); + // byte pointer to component + Value *loadAddress = GEP(pBase, offset); + loadAddress = BITCAST(loadAddress, PointerType::get(mInt32Ty, 0)); + // pointer to the value to load if we're masking off a component + Value *maskLoadAddress = GEP(vSrcPtr, {C(0), C(i)}); + Value *selMask = VEXTRACT(mask, C(i)); + // switch in a safe address to load if we're trying to access a vertex + Value *validAddress = SELECT(selMask, loadAddress, maskLoadAddress); + Value *val = LOAD(validAddress, C(0)); + vGather = VINSERT(vGather, val, C(i)); + } + + STACKRESTORE(pStack); + } + return vGather; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief convert x86 mask to llvm mask +Value* Builder::MASK(Value* vmask) +{ + Value* src = BITCAST(vmask, mSimdInt32Ty); + return ICMP_SLT(src, VIMMED1(0)); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief convert llvm mask to x86 mask +Value* Builder::VMASK(Value* mask) +{ + return S_EXT(mask, mSimdInt32Ty); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Generate a VPSHUFB operation in LLVM IR. If not +/// supported on the underlying platform, emulate it +/// @param a - 256bit SIMD(32x8bit) of 8bit integer values +/// @param b - 256bit SIMD(32x8bit) of 8bit integer mask values +/// Byte masks in lower 128 lane of b selects 8 bit values from lower +/// 128bits of a, and vice versa for the upper lanes. If the mask +/// value is negative, '0' is inserted. +Value *Builder::PSHUFB(Value* a, Value* b) +{ + Value* res; + // use avx2 pshufb instruction if available + if(JM()->mArch.AVX2()) + { + res = VPSHUFB(a, b); + } + else + { + Constant* cB = dyn_cast(b); + // number of 8 bit elements in b + uint32_t numElms = cast(cB->getType())->getNumElements(); + // output vector + Value* vShuf = UndefValue::get(VectorType::get(mInt8Ty, numElms)); + + // insert an 8 bit value from the high and low lanes of a per loop iteration + numElms /= 2; + for(uint32_t i = 0; i < numElms; i++) + { + ConstantInt* cLow128b = cast(cB->getAggregateElement(i)); + ConstantInt* cHigh128b = cast(cB->getAggregateElement(i + numElms)); + + // extract values from constant mask + char valLow128bLane = (char)(cLow128b->getSExtValue()); + char valHigh128bLane = (char)(cHigh128b->getSExtValue()); + + Value* insertValLow128b; + Value* insertValHigh128b; + + // if the mask value is negative, insert a '0' in the respective output position + // otherwise, lookup the value at mask position (bits 3..0 of the respective mask byte) in a and insert in output vector + insertValLow128b = (valLow128bLane < 0) ? C((char)0) : VEXTRACT(a, C((valLow128bLane & 0xF))); + insertValHigh128b = (valHigh128bLane < 0) ? C((char)0) : VEXTRACT(a, C((valHigh128bLane & 0xF) + numElms)); + + vShuf = VINSERT(vShuf, insertValLow128b, i); + vShuf = VINSERT(vShuf, insertValHigh128b, (i + numElms)); + } + res = vShuf; + } + return res; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Generate a VPSHUFB operation (sign extend 8 8bit values to 32 +/// bits)in LLVM IR. If not supported on the underlying platform, emulate it +/// @param a - 128bit SIMD lane(16x8bit) of 8bit integer values. Only +/// lower 8 values are used. +Value *Builder::PMOVSXBD(Value* a) +{ + Value* res; + // use avx2 byte sign extend instruction if available + if(JM()->mArch.AVX2()) + { + res = VPMOVSXBD(a); + } + else + { + // VPMOVSXBD output type + Type* v8x32Ty = VectorType::get(mInt32Ty, 8); + // Extract 8 values from 128bit lane and sign extend + res = S_EXT(VSHUFFLE(a, a, C({0, 1, 2, 3, 4, 5, 6, 7})), v8x32Ty); + } + return res; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Generate a VPSHUFB operation (sign extend 8 16bit values to 32 +/// bits)in LLVM IR. If not supported on the underlying platform, emulate it +/// @param a - 128bit SIMD lane(8x16bit) of 16bit integer values. +Value *Builder::PMOVSXWD(Value* a) +{ + Value* res; + // use avx2 word sign extend if available + if(JM()->mArch.AVX2()) + { + res = VPMOVSXWD(a); + } + else + { + // VPMOVSXWD output type + Type* v8x32Ty = VectorType::get(mInt32Ty, 8); + // Extract 8 values from 128bit lane and sign extend + res = S_EXT(VSHUFFLE(a, a, C({0, 1, 2, 3, 4, 5, 6, 7})), v8x32Ty); + } + return res; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Generate a VPERMD operation (shuffle 32 bit integer values +/// across 128 bit lanes) in LLVM IR. If not supported on the underlying +/// platform, emulate it +/// @param a - 256bit SIMD lane(8x32bit) of integer values. +/// @param idx - 256bit SIMD lane(8x32bit) of 3 bit lane index values +Value *Builder::PERMD(Value* a, Value* idx) +{ + Value* res; + // use avx2 permute instruction if available + if(JM()->mArch.AVX2()) + { + // llvm 3.6.0 swapped the order of the args to vpermd + res = VPERMD(idx, a); + } + else + { + res = VSHUFFLE(a, a, idx); + } + return res; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Generate a VCVTPH2PS operation (float16->float32 conversion) +/// in LLVM IR. If not supported on the underlying platform, emulate it +/// @param a - 128bit SIMD lane(8x16bit) of float16 in int16 format. +Value *Builder::CVTPH2PS(Value* a) +{ + if (JM()->mArch.F16C()) + { + return VCVTPH2PS(a); + } + else + { + FunctionType* pFuncTy = FunctionType::get(mFP32Ty, mInt16Ty); + Function* pCvtPh2Ps = cast(JM()->mpCurrentModule->getOrInsertFunction("ConvertSmallFloatTo32", pFuncTy)); + + if (sys::DynamicLibrary::SearchForAddressOfSymbol("ConvertSmallFloatTo32") == nullptr) + { + sys::DynamicLibrary::AddSymbol("ConvertSmallFloatTo32", (void *)&ConvertSmallFloatTo32); + } + + Value* pResult = UndefValue::get(mSimdFP32Ty); + for (uint32_t i = 0; i < JM()->mVWidth; ++i) + { + Value* pSrc = VEXTRACT(a, C(i)); + Value* pConv = CALL(pCvtPh2Ps, std::initializer_list{pSrc}); + pResult = VINSERT(pResult, pConv, C(i)); + } + + return pResult; + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Generate a VCVTPS2PH operation (float32->float16 conversion) +/// in LLVM IR. If not supported on the underlying platform, emulate it +/// @param a - 128bit SIMD lane(8x16bit) of float16 in int16 format. +Value *Builder::CVTPS2PH(Value* a, Value* rounding) +{ + if (JM()->mArch.F16C()) + { + return VCVTPS2PH(a, rounding); + } + else + { + // call scalar C function for now + FunctionType* pFuncTy = FunctionType::get(mInt16Ty, mFP32Ty); + Function* pCvtPs2Ph = cast(JM()->mpCurrentModule->getOrInsertFunction("Convert32To16Float", pFuncTy)); + + if (sys::DynamicLibrary::SearchForAddressOfSymbol("Convert32To16Float") == nullptr) + { + sys::DynamicLibrary::AddSymbol("Convert32To16Float", (void *)&Convert32To16Float); + } + + Value* pResult = UndefValue::get(mSimdInt16Ty); + for (uint32_t i = 0; i < JM()->mVWidth; ++i) + { + Value* pSrc = VEXTRACT(a, C(i)); + Value* pConv = CALL(pCvtPs2Ph, std::initializer_list{pSrc}); + pResult = VINSERT(pResult, pConv, C(i)); + } + + return pResult; + } +} + +Value *Builder::PMAXSD(Value* a, Value* b) +{ + if (JM()->mArch.AVX2()) + { + return VPMAXSD(a, b); + } + else + { + // use 4-wide sse max intrinsic on lower/upper halves of 8-wide sources + Function* pmaxsd = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::x86_sse41_pmaxsd); + + // low 128 + Value* aLo = VEXTRACTI128(a, C((uint8_t)0)); + Value* bLo = VEXTRACTI128(b, C((uint8_t)0)); + Value* resLo = CALL(pmaxsd, {aLo, bLo}); + + // high 128 + Value* aHi = VEXTRACTI128(a, C((uint8_t)1)); + Value* bHi = VEXTRACTI128(b, C((uint8_t)1)); + Value* resHi = CALL(pmaxsd, {aHi, bHi}); + + // combine + Value* result = VINSERTI128(VUNDEF_I(), resLo, C((uint8_t)0)); + result = VINSERTI128(result, resHi, C((uint8_t)1)); + + return result; + } +} + +Value *Builder::PMINSD(Value* a, Value* b) +{ + if (JM()->mArch.AVX2()) + { + return VPMINSD(a, b); + } + else + { + // use 4-wide sse max intrinsic on lower/upper halves of 8-wide sources + Function* pminsd = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::x86_sse41_pminsd); + + // low 128 + Value* aLo = VEXTRACTI128(a, C((uint8_t)0)); + Value* bLo = VEXTRACTI128(b, C((uint8_t)0)); + Value* resLo = CALL(pminsd, {aLo, bLo}); + + // high 128 + Value* aHi = VEXTRACTI128(a, C((uint8_t)1)); + Value* bHi = VEXTRACTI128(b, C((uint8_t)1)); + Value* resHi = CALL(pminsd, {aHi, bHi}); + + // combine + Value* result = VINSERTI128(VUNDEF_I(), resLo, C((uint8_t)0)); + result = VINSERTI128(result, resHi, C((uint8_t)1)); + + return result; + } +} + +void Builder::Gather4(const SWR_FORMAT format, Value* pSrcBase, Value* byteOffsets, + Value* mask, Value* vGatherComponents[], bool bPackedOutput) +{ + const SWR_FORMAT_INFO &info = GetFormatInfo(format); + if(info.type[0] == SWR_TYPE_FLOAT && info.bpc[0] == 32) + { + // ensure our mask is the correct type + mask = BITCAST(mask, mSimdFP32Ty); + GATHER4PS(info, pSrcBase, byteOffsets, mask, vGatherComponents, bPackedOutput); + } + else + { + // ensure our mask is the correct type + mask = BITCAST(mask, mSimdInt32Ty); + GATHER4DD(info, pSrcBase, byteOffsets, mask, vGatherComponents, bPackedOutput); + } +} + +void Builder::GATHER4PS(const SWR_FORMAT_INFO &info, Value* pSrcBase, Value* byteOffsets, + Value* mask, Value* vGatherComponents[], bool bPackedOutput) +{ + switch(info.bpp / info.numComps) + { + case 16: + { + Value* vGatherResult[2]; + Value *vMask; + + // TODO: vGatherMaskedVal + Value* vGatherMaskedVal = VIMMED1((float)0); + + // always have at least one component out of x or y to fetch + + // save mask as it is zero'd out after each gather + vMask = mask; + + vGatherResult[0] = GATHERPS(vGatherMaskedVal, pSrcBase, byteOffsets, vMask, C((char)1)); + // e.g. result of first 8x32bit integer gather for 16bit components + // 256i - 0 1 2 3 4 5 6 7 + // xyxy xyxy xyxy xyxy xyxy xyxy xyxy xyxy + // + + // if we have at least one component out of x or y to fetch + if(info.numComps > 2) + { + // offset base to the next components(zw) in the vertex to gather + pSrcBase = GEP(pSrcBase, C((char)4)); + vMask = mask; + + vGatherResult[1] = GATHERPS(vGatherMaskedVal, pSrcBase, byteOffsets, vMask, C((char)1)); + // e.g. result of second 8x32bit integer gather for 16bit components + // 256i - 0 1 2 3 4 5 6 7 + // zwzw zwzw zwzw zwzw zwzw zwzw zwzw zwzw + // + } + else + { + vGatherResult[1] = vGatherMaskedVal; + } + + // Shuffle gathered components into place, each row is a component + Shuffle16bpcGather4(info, vGatherResult, vGatherComponents, bPackedOutput); + } + break; + case 32: + { + // apply defaults + for (uint32_t i = 0; i < 4; ++i) + { + vGatherComponents[i] = VIMMED1(*(float*)&info.defaults[i]); + } + + for(uint32_t i = 0; i < info.numComps; i++) + { + uint32_t swizzleIndex = info.swizzle[i]; + + // save mask as it is zero'd out after each gather + Value *vMask = mask; + + // Gather a SIMD of components + vGatherComponents[swizzleIndex] = GATHERPS(vGatherComponents[swizzleIndex], pSrcBase, byteOffsets, vMask, C((char)1)); + + // offset base to the next component to gather + pSrcBase = GEP(pSrcBase, C((char)4)); + } + } + break; + default: + SWR_ASSERT(0, "Invalid float format"); + break; + } +} + +void Builder::GATHER4DD(const SWR_FORMAT_INFO &info, Value* pSrcBase, Value* byteOffsets, + Value* mask, Value* vGatherComponents[], bool bPackedOutput) +{ + switch (info.bpp / info.numComps) + { + case 8: + { + Value* vGatherMaskedVal = VIMMED1((int32_t)0); + Value* vGatherResult = GATHERDD(vGatherMaskedVal, pSrcBase, byteOffsets, mask, C((char)1)); + // e.g. result of an 8x32bit integer gather for 8bit components + // 256i - 0 1 2 3 4 5 6 7 + // xyzw xyzw xyzw xyzw xyzw xyzw xyzw xyzw + + Shuffle8bpcGather4(info, vGatherResult, vGatherComponents, bPackedOutput); + } + break; + case 16: + { + Value* vGatherResult[2]; + Value *vMask; + + // TODO: vGatherMaskedVal + Value* vGatherMaskedVal = VIMMED1((int32_t)0); + + // always have at least one component out of x or y to fetch + + // save mask as it is zero'd out after each gather + vMask = mask; + + vGatherResult[0] = GATHERDD(vGatherMaskedVal, pSrcBase, byteOffsets, vMask, C((char)1)); + // e.g. result of first 8x32bit integer gather for 16bit components + // 256i - 0 1 2 3 4 5 6 7 + // xyxy xyxy xyxy xyxy xyxy xyxy xyxy xyxy + // + + // if we have at least one component out of x or y to fetch + if(info.numComps > 2) + { + // offset base to the next components(zw) in the vertex to gather + pSrcBase = GEP(pSrcBase, C((char)4)); + vMask = mask; + + vGatherResult[1] = GATHERDD(vGatherMaskedVal, pSrcBase, byteOffsets, vMask, C((char)1)); + // e.g. result of second 8x32bit integer gather for 16bit components + // 256i - 0 1 2 3 4 5 6 7 + // zwzw zwzw zwzw zwzw zwzw zwzw zwzw zwzw + // + } + else + { + vGatherResult[1] = vGatherMaskedVal; + } + + // Shuffle gathered components into place, each row is a component + Shuffle16bpcGather4(info, vGatherResult, vGatherComponents, bPackedOutput); + + } + break; + case 32: + { + // apply defaults + for (uint32_t i = 0; i < 4; ++i) + { + vGatherComponents[i] = VIMMED1((int)info.defaults[i]); + } + + for(uint32_t i = 0; i < info.numComps; i++) + { + uint32_t swizzleIndex = info.swizzle[i]; + + // save mask as it is zero'd out after each gather + Value *vMask = mask; + + // Gather a SIMD of components + vGatherComponents[swizzleIndex] = GATHERDD(vGatherComponents[swizzleIndex], pSrcBase, byteOffsets, vMask, C((char)1)); + + // offset base to the next component to gather + pSrcBase = GEP(pSrcBase, C((char)4)); + } + } + break; + default: + SWR_ASSERT(0, "unsupported format"); + break; + } +} + +void Builder::Shuffle16bpcGather4(const SWR_FORMAT_INFO &info, Value* vGatherInput[2], Value* vGatherOutput[4], bool bPackedOutput) +{ + // cast types + Type* vGatherTy = VectorType::get(IntegerType::getInt32Ty(JM()->mContext), JM()->mVWidth); + Type* v32x8Ty = VectorType::get(mInt8Ty, JM()->mVWidth * 4); // vwidth is units of 32 bits + + // input could either be float or int vector; do shuffle work in int + vGatherInput[0] = BITCAST(vGatherInput[0], mSimdInt32Ty); + vGatherInput[1] = BITCAST(vGatherInput[1], mSimdInt32Ty); + + if(bPackedOutput) + { + Type* v128bitTy = VectorType::get(IntegerType::getIntNTy(JM()->mContext, 128), JM()->mVWidth / 4); // vwidth is units of 32 bits + + // shuffle mask + Value* vConstMask = C({0, 1, 4, 5, 8, 9, 12, 13, 2, 3, 6, 7, 10, 11, 14, 15, + 0, 1, 4, 5, 8, 9, 12, 13, 2, 3, 6, 7, 10, 11, 14, 15}); + Value* vShufResult = BITCAST(PSHUFB(BITCAST(vGatherInput[0], v32x8Ty), vConstMask), vGatherTy); + // after pshufb: group components together in each 128bit lane + // 256i - 0 1 2 3 4 5 6 7 + // xxxx xxxx yyyy yyyy xxxx xxxx yyyy yyyy + + Value* vi128XY = BITCAST(PERMD(vShufResult, C({0, 1, 4, 5, 2, 3, 6, 7})), v128bitTy); + // after PERMD: move and pack xy components into each 128bit lane + // 256i - 0 1 2 3 4 5 6 7 + // xxxx xxxx xxxx xxxx yyyy yyyy yyyy yyyy + + // do the same for zw components + Value* vi128ZW = nullptr; + if(info.numComps > 2) + { + Value* vShufResult = BITCAST(PSHUFB(BITCAST(vGatherInput[1], v32x8Ty), vConstMask), vGatherTy); + vi128ZW = BITCAST(PERMD(vShufResult, C({0, 1, 4, 5, 2, 3, 6, 7})), v128bitTy); + } + + for(uint32_t i = 0; i < 4; i++) + { + uint32_t swizzleIndex = info.swizzle[i]; + // todo: fixed for packed + Value* vGatherMaskedVal = VIMMED1((int32_t)(info.defaults[i])); + if(i >= info.numComps) + { + // set the default component val + vGatherOutput[swizzleIndex] = vGatherMaskedVal; + continue; + } + + // if x or z, extract 128bits from lane 0, else for y or w, extract from lane 1 + uint32_t lane = ((i == 0) || (i == 2)) ? 0 : 1; + // if x or y, use vi128XY permute result, else use vi128ZW + Value* selectedPermute = (i < 2) ? vi128XY : vi128ZW; + + // extract packed component 128 bit lanes + vGatherOutput[swizzleIndex] = VEXTRACT(selectedPermute, C(lane)); + } + + } + else + { + // pshufb masks for each component + Value* vConstMask[2]; + // x/z shuffle mask + vConstMask[0] = C({0, 1, -1, -1, 4, 5, -1, -1, 8, 9, -1, -1, 12, 13, -1, -1, + 0, 1, -1, -1, 4, 5, -1, -1, 8, 9, -1, -1, 12, 13, -1, -1, }); + + // y/w shuffle mask + vConstMask[1] = C({2, 3, -1, -1, 6, 7, -1, -1, 10, 11, -1, -1, 14, 15, -1, -1, + 2, 3, -1, -1, 6, 7, -1, -1, 10, 11, -1, -1, 14, 15, -1, -1}); + + + // shuffle enabled components into lower word of each 32bit lane, 0 extending to 32 bits + // apply defaults + for (uint32_t i = 0; i < 4; ++i) + { + vGatherOutput[i] = VIMMED1((int32_t)info.defaults[i]); + } + + for(uint32_t i = 0; i < info.numComps; i++) + { + uint32_t swizzleIndex = info.swizzle[i]; + + // select correct constMask for x/z or y/w pshufb + uint32_t selectedMask = ((i == 0) || (i == 2)) ? 0 : 1; + // if x or y, use vi128XY permute result, else use vi128ZW + uint32_t selectedGather = (i < 2) ? 0 : 1; + + vGatherOutput[swizzleIndex] = BITCAST(PSHUFB(BITCAST(vGatherInput[selectedGather], v32x8Ty), vConstMask[selectedMask]), vGatherTy); + // after pshufb mask for x channel; z uses the same shuffle from the second gather + // 256i - 0 1 2 3 4 5 6 7 + // xx00 xx00 xx00 xx00 xx00 xx00 xx00 xx00 + } + } +} + +void Builder::Shuffle8bpcGather4(const SWR_FORMAT_INFO &info, Value* vGatherInput, Value* vGatherOutput[], bool bPackedOutput) +{ + // cast types + Type* vGatherTy = VectorType::get(IntegerType::getInt32Ty(JM()->mContext), JM()->mVWidth); + Type* v32x8Ty = VectorType::get(mInt8Ty, JM()->mVWidth * 4 ); // vwidth is units of 32 bits + + if(bPackedOutput) + { + Type* v128Ty = VectorType::get(IntegerType::getIntNTy(JM()->mContext, 128), JM()->mVWidth / 4); // vwidth is units of 32 bits + // shuffle mask + Value* vConstMask = C({0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15, + 0, 4, 8, 12, 1, 5, 9, 13, 2, 6, 10, 14, 3, 7, 11, 15}); + Value* vShufResult = BITCAST(PSHUFB(BITCAST(vGatherInput, v32x8Ty), vConstMask), vGatherTy); + // after pshufb: group components together in each 128bit lane + // 256i - 0 1 2 3 4 5 6 7 + // xxxx yyyy zzzz wwww xxxx yyyy zzzz wwww + + Value* vi128XY = BITCAST(PERMD(vShufResult, C({0, 4, 0, 0, 1, 5, 0, 0})), v128Ty); + // after PERMD: move and pack xy and zw components in low 64 bits of each 128bit lane + // 256i - 0 1 2 3 4 5 6 7 + // xxxx xxxx dcdc dcdc yyyy yyyy dcdc dcdc (dc - don't care) + + // do the same for zw components + Value* vi128ZW = nullptr; + if(info.numComps > 2) + { + vi128ZW = BITCAST(PERMD(vShufResult, C({2, 6, 0, 0, 3, 7, 0, 0})), v128Ty); + } + + // sign extend all enabled components. If we have a fill vVertexElements, output to current simdvertex + for(uint32_t i = 0; i < 4; i++) + { + uint32_t swizzleIndex = info.swizzle[i]; + // todo: fix for packed + Value* vGatherMaskedVal = VIMMED1((int32_t)(info.defaults[i])); + if(i >= info.numComps) + { + // set the default component val + vGatherOutput[swizzleIndex] = vGatherMaskedVal; + continue; + } + + // if x or z, extract 128bits from lane 0, else for y or w, extract from lane 1 + uint32_t lane = ((i == 0) || (i == 2)) ? 0 : 1; + // if x or y, use vi128XY permute result, else use vi128ZW + Value* selectedPermute = (i < 2) ? vi128XY : vi128ZW; + + // sign extend + vGatherOutput[swizzleIndex] = VEXTRACT(selectedPermute, C(lane)); + } + } + // else zero extend + else{ + // shuffle enabled components into lower byte of each 32bit lane, 0 extending to 32 bits + // apply defaults + for (uint32_t i = 0; i < 4; ++i) + { + vGatherOutput[i] = VIMMED1((int32_t)info.defaults[i]); + } + + for(uint32_t i = 0; i < info.numComps; i++){ + uint32_t swizzleIndex = info.swizzle[i]; + + // pshufb masks for each component + Value* vConstMask; + switch(i) + { + case 0: + // x shuffle mask + vConstMask = C({0, -1, -1, -1, 4, -1, -1, -1, 8, -1, -1, -1, 12, -1, -1, -1, + 0, -1, -1, -1, 4, -1, -1, -1, 8, -1, -1, -1, 12, -1, -1, -1}); + break; + case 1: + // y shuffle mask + vConstMask = C({1, -1, -1, -1, 5, -1, -1, -1, 9, -1, -1, -1, 13, -1, -1, -1, + 1, -1, -1, -1, 5, -1, -1, -1, 9, -1, -1, -1, 13, -1, -1, -1}); + break; + case 2: + // z shuffle mask + vConstMask = C({2, -1, -1, -1, 6, -1, -1, -1, 10, -1, -1, -1, 14, -1, -1, -1, + 2, -1, -1, -1, 6, -1, -1, -1, 10, -1, -1, -1, 14, -1, -1, -1}); + break; + case 3: + // w shuffle mask + vConstMask = C({3, -1, -1, -1, 7, -1, -1, -1, 11, -1, -1, -1, 15, -1, -1, -1, + 3, -1, -1, -1, 7, -1, -1, -1, 11, -1, -1, -1, 15, -1, -1, -1}); + break; + default: + vConstMask = nullptr; + break; + } + + vGatherOutput[swizzleIndex] = BITCAST(PSHUFB(BITCAST(vGatherInput, v32x8Ty), vConstMask), vGatherTy); + // after pshufb for x channel + // 256i - 0 1 2 3 4 5 6 7 + // x000 x000 x000 x000 x000 x000 x000 x000 + } + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief emulates a scatter operation. +/// @param pDst - pointer to destination +/// @param vSrc - vector of src data to scatter +/// @param vOffsets - vector of byte offsets from pDst +/// @param vMask - mask of valid lanes +void Builder::SCATTERPS(Value* pDst, Value* vSrc, Value* vOffsets, Value* vMask) +{ + Value* pStack = STACKSAVE(); + + // allocate tmp stack for masked off lanes + Value* vTmpPtr = ALLOCA(vSrc->getType()->getVectorElementType()); + + Value *mask = MASK(vMask); + for (uint32_t i = 0; i < JM()->mVWidth; ++i) + { + Value *offset = VEXTRACT(vOffsets, C(i)); + // byte pointer to component + Value *storeAddress = GEP(pDst, offset); + storeAddress = BITCAST(storeAddress, PointerType::get(mFP32Ty, 0)); + Value *selMask = VEXTRACT(mask, C(i)); + Value *srcElem = VEXTRACT(vSrc, C(i)); + // switch in a safe address to load if we're trying to access a vertex + Value *validAddress = SELECT(selMask, storeAddress, vTmpPtr); + STORE(srcElem, validAddress); + } + + STACKRESTORE(pStack); +} + +Value* Builder::VABSPS(Value* a) +{ + Value* asInt = BITCAST(a, mSimdInt32Ty); + Value* result = BITCAST(AND(asInt, VIMMED1(0x7fffffff)), mSimdFP32Ty); + return result; +} + +Value *Builder::ICLAMP(Value* src, Value* low, Value* high) +{ + Value *lowCmp = ICMP_SLT(src, low); + Value *ret = SELECT(lowCmp, low, src); + + Value *highCmp = ICMP_SGT(ret, high); + ret = SELECT(highCmp, high, ret); + + return ret; +} + +Value *Builder::FCLAMP(Value* src, Value* low, Value* high) +{ + Value *lowCmp = FCMP_OLT(src, low); + Value *ret = SELECT(lowCmp, low, src); + + Value *highCmp = FCMP_OGT(ret, high); + ret = SELECT(highCmp, high, ret); + + return ret; +} + +Value *Builder::FCLAMP(Value* src, float low, float high) +{ + Value* result = VMAXPS(src, VIMMED1(low)); + result = VMINPS(result, VIMMED1(high)); + + return result; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief save/restore stack, providing ability to push/pop the stack and +/// reduce overall stack requirements for temporary stack use +Value* Builder::STACKSAVE() +{ + Function* pfnStackSave = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::stacksave); +#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 6 + return CALL(pfnStackSave); +#else + return CALLA(pfnStackSave); +#endif +} + +void Builder::STACKRESTORE(Value* pSaved) +{ + Function* pfnStackRestore = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::stackrestore); + CALL(pfnStackRestore, std::initializer_list{pSaved}); +} + +Value *Builder::FMADDPS(Value* a, Value* b, Value* c) +{ + Value* vOut; + // use FMADs if available + if(JM()->mArch.AVX2()) + { + vOut = VFMADDPS(a, b, c); + } + else + { + vOut = FADD(FMUL(a, b), c); + } + return vOut; +} + +Value* Builder::POPCNT(Value* a) +{ + Function* pCtPop = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::ctpop, { a->getType() }); + return CALL(pCtPop, std::initializer_list{a}); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief C functions called by LLVM IR +////////////////////////////////////////////////////////////////////////// + +////////////////////////////////////////////////////////////////////////// +/// @brief called in JIT code, inserted by PRINT +/// output to both stdout and visual studio debug console +void __cdecl CallPrint(const char* fmt, ...) +{ + va_list args; + va_start(args, fmt); + vprintf(fmt, args); + +#if defined( _WIN32 ) + char strBuf[1024]; + vsnprintf_s(strBuf, _TRUNCATE, fmt, args); + OutputDebugString(strBuf); +#endif +} + +Value *Builder::VEXTRACTI128(Value* a, Constant* imm8) +{ +#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 6 + Function *func = + Intrinsic::getDeclaration(JM()->mpCurrentModule, + Intrinsic::x86_avx_vextractf128_si_256); + return CALL(func, {a, imm8}); +#else + bool flag = !imm8->isZeroValue(); + SmallVector idx; + for (unsigned i = 0; i < JM()->mVWidth / 2; i++) { + idx.push_back(C(flag ? i + JM()->mVWidth / 2 : i)); + } + return VSHUFFLE(a, VUNDEF_I(), ConstantVector::get(idx)); +#endif +} + +Value *Builder::VINSERTI128(Value* a, Value* b, Constant* imm8) +{ +#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 6 + Function *func = + Intrinsic::getDeclaration(JM()->mpCurrentModule, + Intrinsic::x86_avx_vinsertf128_si_256); + return CALL(func, {a, b, imm8}); +#else + bool flag = !imm8->isZeroValue(); + SmallVector idx; + for (unsigned i = 0; i < JM()->mVWidth; i++) { + idx.push_back(C(i)); + } + Value *inter = VSHUFFLE(b, VUNDEF_I(), ConstantVector::get(idx)); + + SmallVector idx2; + for (unsigned i = 0; i < JM()->mVWidth / 2; i++) { + idx2.push_back(C(flag ? i : i + JM()->mVWidth)); + } + for (unsigned i = JM()->mVWidth / 2; i < JM()->mVWidth; i++) { + idx2.push_back(C(flag ? i + JM()->mVWidth / 2 : i)); + } + return VSHUFFLE(a, inter, ConstantVector::get(idx2)); +#endif +} diff --git a/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h new file mode 100644 index 00000000000..48e0558c4dd --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/builder_misc.h @@ -0,0 +1,149 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file builder_misc.h +* +* @brief miscellaneous builder functions +* +* Notes: +* +******************************************************************************/ +#pragma once + +Constant *C(bool i); +Constant *C(char i); +Constant *C(uint8_t i); +Constant *C(int i); +Constant *C(int64_t i); +Constant *C(uint16_t i); +Constant *C(uint32_t i); +Constant *C(float i); + +template +Constant *C(const std::initializer_list &constList) +{ + std::vector vConsts; + for(auto i : constList) { + + vConsts.push_back(C((Ty)i)); + } + return ConstantVector::get(vConsts); +} + +Constant *PRED(bool pred); +Value *VIMMED1(int i); +Value *VIMMED1(uint32_t i); +Value *VIMMED1(float i); +Value *VIMMED1(bool i); +Value *VUNDEF(Type* t); +Value *VUNDEF_F(); +Value *VUNDEF_I(); +Value *VUNDEF(Type* ty, uint32_t size); +Value *VUNDEF_IPTR(); +#if LLVM_VERSION_MAJOR == 3 && LLVM_VERSION_MINOR == 6 +Value *VINSERT(Value *vec, Value *val, uint64_t index); +#endif +Value *VBROADCAST(Value *src); +Value *VRCP(Value *va); +Value *VPLANEPS(Value* vA, Value* vB, Value* vC, Value* &vX, Value* &vY); + +uint32_t IMMED(Value* i); + +Value *GEP(Value* ptr, const std::initializer_list &indexList); +Value *GEP(Value* ptr, const std::initializer_list &indexList); +CallInst *CALL(Value *Callee, const std::initializer_list &args); + +LoadInst *LOAD(Value *BasePtr, const std::initializer_list &offset, const llvm::Twine& name = ""); +LoadInst *LOADV(Value *BasePtr, const std::initializer_list &offset, const llvm::Twine& name = ""); +StoreInst *STORE(Value *Val, Value *BasePtr, const std::initializer_list &offset); +StoreInst *STOREV(Value *Val, Value *BasePtr, const std::initializer_list &offset); + +Value *VCMPPS_EQ(Value* a, Value* b) { return VCMPPS(a, b, C((uint8_t)_CMP_EQ_OQ)); } +Value *VCMPPS_LT(Value* a, Value* b) { return VCMPPS(a, b, C((uint8_t)_CMP_LT_OQ)); } +Value *VCMPPS_LE(Value* a, Value* b) { return VCMPPS(a, b, C((uint8_t)_CMP_LE_OQ)); } +Value *VCMPPS_ISNAN(Value* a, Value* b) { return VCMPPS(a, b, C((uint8_t)_CMP_UNORD_Q)); } +Value *VCMPPS_NEQ(Value* a, Value* b) { return VCMPPS(a, b, C((uint8_t)_CMP_NEQ_OQ)); } +Value *VCMPPS_GE(Value* a, Value* b) { return VCMPPS(a, b, C((uint8_t)_CMP_GE_OQ)); } +Value *VCMPPS_GT(Value* a, Value* b) { return VCMPPS(a, b, C((uint8_t)_CMP_GT_OQ)); } +Value *VCMPPS_NOTNAN(Value* a, Value* b){ return VCMPPS(a, b, C((uint8_t)_CMP_ORD_Q)); } + +Value *MASK(Value* vmask); +Value *VMASK(Value* mask); + +////////////////////////////////////////////////////////////////////////// +/// @brief functions that build IR to call x86 intrinsics directly, or +/// emulate them with other instructions if not available on the host +////////////////////////////////////////////////////////////////////////// +Value *MASKLOADD(Value* src, Value* mask); + +void Gather4(const SWR_FORMAT format, Value* pSrcBase, Value* byteOffsets, + Value* mask, Value* vGatherComponents[], bool bPackedOutput); + +Value *GATHERPS(Value* src, Value* pBase, Value* indices, Value* mask, Value* scale); +void GATHER4PS(const SWR_FORMAT_INFO &info, Value* pSrcBase, Value* byteOffsets, + Value* mask, Value* vGatherComponents[], bool bPackedOutput); + +Value *GATHERDD(Value* src, Value* pBase, Value* indices, Value* mask, Value* scale); +void GATHER4DD(const SWR_FORMAT_INFO &info, Value* pSrcBase, Value* byteOffsets, + Value* mask, Value* vGatherComponents[], bool bPackedOutput); + +void SCATTERPS(Value* pDst, Value* vSrc, Value* vOffsets, Value* vMask); + +void Shuffle8bpcGather4(const SWR_FORMAT_INFO &info, Value* vGatherInput, Value* vGatherOutput[], bool bPackedOutput); +void Shuffle16bpcGather4(const SWR_FORMAT_INFO &info, Value* vGatherInput[], Value* vGatherOutput[], bool bPackedOutput); + +Value *PSHUFB(Value* a, Value* b); +Value *PMOVSXBD(Value* a); +Value *PMOVSXWD(Value* a); +Value *PERMD(Value* a, Value* idx); +Value *CVTPH2PS(Value* a); +Value *CVTPS2PH(Value* a, Value* rounding); +Value *PMAXSD(Value* a, Value* b); +Value *PMINSD(Value* a, Value* b); +Value *VABSPS(Value* a); +Value *FMADDPS(Value* a, Value* b, Value* c); + +// LLVM removed VPCMPGTD x86 intrinsic. This emulates that behavior +Value *VPCMPGTD(Value* a, Value* b) +{ + Value* vIndexMask = ICMP_UGT(a,b); + + // need to set the high bit for x86 intrinsic masks + return S_EXT(vIndexMask,VectorType::get(mInt32Ty,JM()->mVWidth)); +} + +Value *ICLAMP(Value* src, Value* low, Value* high); +Value *FCLAMP(Value* src, Value* low, Value* high); +Value *FCLAMP(Value* src, float low, float high); + +CallInst *PRINT(const std::string &printStr); +CallInst *PRINT(const std::string &printStr,const std::initializer_list &printArgs); +Value* STACKSAVE(); +void STACKRESTORE(Value* pSaved); + +Value* POPCNT(Value* a); + +Value* INT3() { return INTERRUPT(C((uint8_t)3)); } + + +Value *VEXTRACTI128(Value* a, Constant* imm8); +Value *VINSERTI128(Value* a, Value* b, Constant* imm8); diff --git a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp new file mode 100644 index 00000000000..c5a180e27cb --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.cpp @@ -0,0 +1,1431 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file fetch_jit.cpp +* +* @brief Implementation of the fetch jitter +* +* Notes: +* +******************************************************************************/ +#include "jit_api.h" +#include "fetch_jit.h" +#include "builder.h" +#include "state_llvm.h" +#include "common/containers.hpp" +#include "llvm/IR/DataLayout.h" +#include +#include + +//#define FETCH_DUMP_VERTEX 1 + +bool isComponentEnabled(ComponentEnable enableMask, uint8_t component); + +enum ConversionType +{ + CONVERT_NONE, + CONVERT_NORMALIZED, + CONVERT_USCALED, + CONVERT_SSCALED, +}; + +////////////////////////////////////////////////////////////////////////// +/// Interface to Jitting a fetch shader +////////////////////////////////////////////////////////////////////////// +struct FetchJit : public Builder +{ + FetchJit(JitManager* pJitMgr) : Builder(pJitMgr){}; + + Function* Create(const FETCH_COMPILE_STATE& fetchState); + Value* GetSimdValid32bitIndices(Value* vIndices, Value* pLastIndex); + Value* GetSimdValid16bitIndices(Value* vIndices, Value* pLastIndex); + Value* GetSimdValid8bitIndices(Value* vIndices, Value* pLastIndex); + + // package up Shuffle*bpcGatherd args into a tuple for convenience + typedef std::tuple Shuffle8bpcArgs; + void Shuffle8bpcGatherd(Shuffle8bpcArgs &args); + + typedef std::tuple Shuffle16bpcArgs; + void Shuffle16bpcGather(Shuffle16bpcArgs &args); + + void StoreVertexElements(Value* pVtxOut, const uint32_t outputElt, const uint32_t numEltsToStore, Value* (&vVertexElements)[4]); + + Value* GenerateCompCtrlVector(const ComponentControl ctrl); + + void JitLoadVertices(const FETCH_COMPILE_STATE &fetchState, Value* fetchInfo, Value* streams, Value* vIndices, Value* pVtxOut); + void JitGatherVertices(const FETCH_COMPILE_STATE &fetchState, Value* fetchInfo, Value* streams, Value* vIndices, Value* pVtxOut); +}; + +Function* FetchJit::Create(const FETCH_COMPILE_STATE& fetchState) +{ + static std::size_t fetchNum = 0; + + std::stringstream fnName("FetchShader", std::ios_base::in | std::ios_base::out | std::ios_base::ate); + fnName << fetchNum++; + + Function* fetch = Function::Create(JM()->mFetchShaderTy, GlobalValue::ExternalLinkage, fnName.str(), JM()->mpCurrentModule); + BasicBlock* entry = BasicBlock::Create(JM()->mContext, "entry", fetch); + + IRB()->SetInsertPoint(entry); + + auto argitr = fetch->getArgumentList().begin(); + + // Fetch shader arguments + Value* fetchInfo = &*argitr; ++argitr; + fetchInfo->setName("fetchInfo"); + Value* pVtxOut = &*argitr; + pVtxOut->setName("vtxOutput"); + // this is just shorthand to tell LLVM to get a pointer to the base address of simdvertex + // index 0(just the pointer to the simdvertex structure + // index 1(which element of the simdvertex structure to offset to(in this case 0) + // so the indices being i32's doesn't matter + // TODO: generated this GEP with a VECTOR structure type so this makes sense + std::vector vtxInputIndices(2, C(0)); + // GEP + pVtxOut = GEP(pVtxOut, C(0)); + pVtxOut = BITCAST(pVtxOut, PointerType::get(VectorType::get(mFP32Ty, JM()->mVWidth), 0)); + + // SWR_FETCH_CONTEXT::pStreams + Value* streams = LOAD(fetchInfo,{0, SWR_FETCH_CONTEXT_pStreams}); + streams->setName("pStreams"); + + // SWR_FETCH_CONTEXT::pIndices + Value* indices = LOAD(fetchInfo,{0, SWR_FETCH_CONTEXT_pIndices}); + indices->setName("pIndices"); + + // SWR_FETCH_CONTEXT::pLastIndex + Value* pLastIndex = LOAD(fetchInfo,{0, SWR_FETCH_CONTEXT_pLastIndex}); + pLastIndex->setName("pLastIndex"); + + + Value* vIndices; + switch(fetchState.indexType) + { + case R8_UINT: + indices = BITCAST(indices, Type::getInt8PtrTy(JM()->mContext, 0)); + if(fetchState.bDisableIndexOOBCheck){ + vIndices = LOAD(BITCAST(indices, PointerType::get(VectorType::get(mInt8Ty, mpJitMgr->mVWidth), 0)), {(uint32_t)0}); + vIndices = Z_EXT(vIndices, mSimdInt32Ty); + } + else{ + pLastIndex = BITCAST(pLastIndex, Type::getInt8PtrTy(JM()->mContext, 0)); + vIndices = GetSimdValid8bitIndices(indices, pLastIndex); + } + break; + case R16_UINT: + indices = BITCAST(indices, Type::getInt16PtrTy(JM()->mContext, 0)); + if(fetchState.bDisableIndexOOBCheck){ + vIndices = LOAD(BITCAST(indices, PointerType::get(VectorType::get(mInt16Ty, mpJitMgr->mVWidth), 0)), {(uint32_t)0}); + vIndices = Z_EXT(vIndices, mSimdInt32Ty); + } + else{ + pLastIndex = BITCAST(pLastIndex, Type::getInt16PtrTy(JM()->mContext, 0)); + vIndices = GetSimdValid16bitIndices(indices, pLastIndex); + } + break; + case R32_UINT: + (fetchState.bDisableIndexOOBCheck) ? vIndices = LOAD(BITCAST(indices, PointerType::get(mSimdInt32Ty,0)),{(uint32_t)0}) + : vIndices = GetSimdValid32bitIndices(indices, pLastIndex); + break; // incoming type is already 32bit int + default: SWR_ASSERT(0, "Unsupported index type"); vIndices = nullptr; break; + } + + // store out vertex IDs + STORE(vIndices, GEP(fetchInfo, { 0, SWR_FETCH_CONTEXT_VertexID })); + + // store out cut mask if enabled + if (fetchState.bEnableCutIndex) + { + Value* vCutIndex = VIMMED1(fetchState.cutIndex); + Value* cutMask = VMASK(ICMP_EQ(vIndices, vCutIndex)); + STORE(cutMask, GEP(fetchInfo, { 0, SWR_FETCH_CONTEXT_CutMask })); + } + + // Fetch attributes from memory and output to a simdvertex struct + // since VGATHER has a perf penalty on HSW vs BDW, allow client to choose which fetch method to use + (fetchState.bDisableVGATHER) ? JitLoadVertices(fetchState, fetchInfo, streams, vIndices, pVtxOut) + : JitGatherVertices(fetchState, fetchInfo, streams, vIndices, pVtxOut); + + RET_VOID(); + + JitManager::DumpToFile(fetch, "src"); + + verifyFunction(*fetch); + + FunctionPassManager setupPasses(JM()->mpCurrentModule); + + ///@todo We don't need the CFG passes for fetch. (e.g. BreakCriticalEdges and CFGSimplification) + setupPasses.add(createBreakCriticalEdgesPass()); + setupPasses.add(createCFGSimplificationPass()); + setupPasses.add(createEarlyCSEPass()); + setupPasses.add(createPromoteMemoryToRegisterPass()); + + setupPasses.run(*fetch); + + JitManager::DumpToFile(fetch, "se"); + + FunctionPassManager optPasses(JM()->mpCurrentModule); + + ///@todo Haven't touched these either. Need to remove some of these and add others. + optPasses.add(createCFGSimplificationPass()); + optPasses.add(createEarlyCSEPass()); + optPasses.add(createInstructionCombiningPass()); + optPasses.add(createInstructionSimplifierPass()); + optPasses.add(createConstantPropagationPass()); + optPasses.add(createSCCPPass()); + optPasses.add(createAggressiveDCEPass()); + + optPasses.run(*fetch); + optPasses.run(*fetch); + + JitManager::DumpToFile(fetch, "opt"); + + return fetch; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Loads attributes from memory using LOADs, shuffling the +/// components into SOA form. +/// *Note* currently does not support component control, +/// component packing, or instancing +/// @param fetchState - info about attributes to be fetched from memory +/// @param streams - value pointer to the current vertex stream +/// @param vIndices - vector value of indices to load +/// @param pVtxOut - value pointer to output simdvertex struct +void FetchJit::JitLoadVertices(const FETCH_COMPILE_STATE &fetchState, Value* fetchInfo, Value* streams, Value* vIndices, Value* pVtxOut) +{ + // Zack shuffles; a variant of the Charleston. + + SWRL::UncheckedFixedVector vectors; + + std::vector pMask(JM()->mVWidth); + for(uint32_t i = 0; i < JM()->mVWidth; ++i) + { + pMask[i] = (C(i < 4 ? i : 4)); + } + Constant* promoteMask = ConstantVector::get(pMask); + Constant* uwvec = UndefValue::get(VectorType::get(mFP32Ty, 4)); + + Value* startVertex = LOAD(fetchInfo, {0, SWR_FETCH_CONTEXT_StartVertex}); + + for(uint32_t nelt = 0; nelt < fetchState.numAttribs; ++nelt) + { + Value* elements[4] = {0}; + const INPUT_ELEMENT_DESC& ied = fetchState.layout[nelt]; + const SWR_FORMAT_INFO &info = GetFormatInfo((SWR_FORMAT)ied.Format); + uint32_t numComponents = info.numComps; + uint32_t bpc = info.bpp / info.numComps; ///@todo Code below assumes all components are same size. Need to fix. + + vectors.clear(); + + // load SWR_VERTEX_BUFFER_STATE::pData + Value *stream = LOAD(streams, {ied.StreamIndex, 2}); + + // load SWR_VERTEX_BUFFER_STATE::pitch + Value *stride = LOAD(streams, {ied.StreamIndex, 1}); + stride = Z_EXT(stride, mInt64Ty); + + // load SWR_VERTEX_BUFFER_STATE::size + Value *size = LOAD(streams, {ied.StreamIndex, 3}); + size = Z_EXT(size, mInt64Ty); + + Value* startVertexOffset = MUL(Z_EXT(startVertex, mInt64Ty), stride); + + // Load from the stream. + for(uint32_t lane = 0; lane < JM()->mVWidth; ++lane) + { + // Get index + Value* index = VEXTRACT(vIndices, C(lane)); + index = Z_EXT(index, mInt64Ty); + + Value* offset = MUL(index, stride); + offset = ADD(offset, C((int64_t)ied.AlignedByteOffset)); + offset = ADD(offset, startVertexOffset); + + if (!fetchState.bDisableIndexOOBCheck) { + // check for out of bound access, including partial OOB, and mask them to 0 + Value *endOffset = ADD(offset, C((int64_t)info.Bpp)); + Value *oob = ICMP_ULE(endOffset, size); + offset = SELECT(oob, offset, ConstantInt::get(mInt64Ty, 0)); + } + + Value* pointer = GEP(stream, offset); + // We use a full-lane, but don't actually care. + Value* vptr = 0; + + // get a pointer to a 4 component attrib in default address space + switch(bpc) + { + case 8: vptr = BITCAST(pointer, PointerType::get(VectorType::get(mInt8Ty, 4), 0)); break; + case 16: vptr = BITCAST(pointer, PointerType::get(VectorType::get(mInt16Ty, 4), 0)); break; + case 32: vptr = BITCAST(pointer, PointerType::get(VectorType::get(mFP32Ty, 4), 0)); break; + default: SWR_ASSERT(false, "Unsupported underlying bpp!"); + } + + // load 4 components of attribute + Value* vec = ALIGNED_LOAD(vptr, 1, false); + + // Convert To FP32 internally + switch(info.type[0]) + { + case SWR_TYPE_UNORM: + switch(bpc) + { + case 8: + vec = UI_TO_FP(vec, VectorType::get(mFP32Ty, 4)); + vec = FMUL(vec, ConstantVector::get(std::vector(4, ConstantFP::get(mFP32Ty, 1.0 / 255.0)))); + break; + case 16: + vec = UI_TO_FP(vec, VectorType::get(mFP32Ty, 4)); + vec = FMUL(vec, ConstantVector::get(std::vector(4, ConstantFP::get(mFP32Ty, 1.0 / 65535.0)))); + break; + default: + SWR_ASSERT(false, "Unsupported underlying type!"); + break; + } + break; + case SWR_TYPE_SNORM: + switch(bpc) + { + case 8: + vec = SI_TO_FP(vec, VectorType::get(mFP32Ty, 4)); + vec = FMUL(vec, ConstantVector::get(std::vector(4, ConstantFP::get(mFP32Ty, 1.0 / 128.0)))); + break; + case 16: + vec = SI_TO_FP(vec, VectorType::get(mFP32Ty, 4)); + vec = FMUL(vec, ConstantVector::get(std::vector(4, ConstantFP::get(mFP32Ty, 1.0 / 32768.0)))); + break; + default: + SWR_ASSERT(false, "Unsupported underlying type!"); + break; + } + break; + case SWR_TYPE_UINT: + // Zero extend uint32_t types. + switch(bpc) + { + case 8: + case 16: + vec = Z_EXT(vec, VectorType::get(mInt32Ty, 4)); + vec = BITCAST(vec, VectorType::get(mFP32Ty, 4)); + break; + case 32: + break; // Pass through unchanged. + default: + SWR_ASSERT(false, "Unsupported underlying type!"); + break; + } + break; + case SWR_TYPE_SINT: + // Sign extend SINT types. + switch(bpc) + { + case 8: + case 16: + vec = S_EXT(vec, VectorType::get(mInt32Ty, 4)); + vec = BITCAST(vec, VectorType::get(mFP32Ty, 4)); + break; + case 32: + break; // Pass through unchanged. + default: + SWR_ASSERT(false, "Unsupported underlying type!"); + break; + } + break; + case SWR_TYPE_FLOAT: + switch(bpc) + { + case 32: + break; // Pass through unchanged. + default: + SWR_ASSERT(false, "Unsupported underlying type!"); + } + break; + case SWR_TYPE_USCALED: + vec = UI_TO_FP(vec, VectorType::get(mFP32Ty, 4)); + break; + case SWR_TYPE_SSCALED: + vec = SI_TO_FP(vec, VectorType::get(mFP32Ty, 4)); + break; + case SWR_TYPE_UNKNOWN: + case SWR_TYPE_UNUSED: + SWR_ASSERT(false, "Unsupported type %d!", info.type[0]); + } + + // promote mask: sse(0,1,2,3) | avx(0,1,2,3,4,4,4,4) + // uwvec: 4 x F32, undef value + Value* wvec = VSHUFFLE(vec, uwvec, promoteMask); + vectors.push_back(wvec); + } + + std::vector v01Mask(JM()->mVWidth); + std::vector v23Mask(JM()->mVWidth); + std::vector v02Mask(JM()->mVWidth); + std::vector v13Mask(JM()->mVWidth); + + // Concatenate the vectors together. + elements[0] = VUNDEF_F(); + elements[1] = VUNDEF_F(); + elements[2] = VUNDEF_F(); + elements[3] = VUNDEF_F(); + for(uint32_t b = 0, num4Wide = JM()->mVWidth / 4; b < num4Wide; ++b) + { + v01Mask[4 * b + 0] = C(0 + 4 * b); + v01Mask[4 * b + 1] = C(1 + 4 * b); + v01Mask[4 * b + 2] = C(0 + 4 * b + JM()->mVWidth); + v01Mask[4 * b + 3] = C(1 + 4 * b + JM()->mVWidth); + + v23Mask[4 * b + 0] = C(2 + 4 * b); + v23Mask[4 * b + 1] = C(3 + 4 * b); + v23Mask[4 * b + 2] = C(2 + 4 * b + JM()->mVWidth); + v23Mask[4 * b + 3] = C(3 + 4 * b + JM()->mVWidth); + + v02Mask[4 * b + 0] = C(0 + 4 * b); + v02Mask[4 * b + 1] = C(2 + 4 * b); + v02Mask[4 * b + 2] = C(0 + 4 * b + JM()->mVWidth); + v02Mask[4 * b + 3] = C(2 + 4 * b + JM()->mVWidth); + + v13Mask[4 * b + 0] = C(1 + 4 * b); + v13Mask[4 * b + 1] = C(3 + 4 * b); + v13Mask[4 * b + 2] = C(1 + 4 * b + JM()->mVWidth); + v13Mask[4 * b + 3] = C(3 + 4 * b + JM()->mVWidth); + + std::vector iMask(JM()->mVWidth); + for(uint32_t i = 0; i < JM()->mVWidth; ++i) + { + if(((4 * b) <= i) && (i < (4 * (b + 1)))) + { + iMask[i] = C(i % 4 + JM()->mVWidth); + } + else + { + iMask[i] = C(i); + } + } + Constant* insertMask = ConstantVector::get(iMask); + elements[0] = VSHUFFLE(elements[0], vectors[4 * b + 0], insertMask); + elements[1] = VSHUFFLE(elements[1], vectors[4 * b + 1], insertMask); + elements[2] = VSHUFFLE(elements[2], vectors[4 * b + 2], insertMask); + elements[3] = VSHUFFLE(elements[3], vectors[4 * b + 3], insertMask); + } + + Value* x0y0x1y1 = VSHUFFLE(elements[0], elements[1], ConstantVector::get(v01Mask)); + Value* x2y2x3y3 = VSHUFFLE(elements[2], elements[3], ConstantVector::get(v01Mask)); + Value* z0w0z1w1 = VSHUFFLE(elements[0], elements[1], ConstantVector::get(v23Mask)); + Value* z2w3z2w3 = VSHUFFLE(elements[2], elements[3], ConstantVector::get(v23Mask)); + elements[0] = VSHUFFLE(x0y0x1y1, x2y2x3y3, ConstantVector::get(v02Mask)); + elements[1] = VSHUFFLE(x0y0x1y1, x2y2x3y3, ConstantVector::get(v13Mask)); + elements[2] = VSHUFFLE(z0w0z1w1, z2w3z2w3, ConstantVector::get(v02Mask)); + elements[3] = VSHUFFLE(z0w0z1w1, z2w3z2w3, ConstantVector::get(v13Mask)); + + switch(numComponents + 1) + { + case 1: elements[0] = VIMMED1(0.0f); + case 2: elements[1] = VIMMED1(0.0f); + case 3: elements[2] = VIMMED1(0.0f); + case 4: elements[3] = VIMMED1(1.0f); + } + + for(uint32_t c = 0; c < 4; ++c) + { + Value* dest = GEP(pVtxOut, C(nelt * 4 + c), "destGEP"); + STORE(elements[c], dest); + } + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Loads attributes from memory using AVX2 GATHER(s) +/// @param fetchState - info about attributes to be fetched from memory +/// @param fetchInfo - first argument passed to fetch shader +/// @param streams - value pointer to the current vertex stream +/// @param vIndices - vector value of indices to gather +/// @param pVtxOut - value pointer to output simdvertex struct +void FetchJit::JitGatherVertices(const FETCH_COMPILE_STATE &fetchState, Value* fetchInfo, + Value* streams, Value* vIndices, Value* pVtxOut) +{ + uint32_t currentVertexElement = 0; + uint32_t outputElt = 0; + Value* vVertexElements[4]; + + Value* startVertex = LOAD(fetchInfo, {0, SWR_FETCH_CONTEXT_StartVertex}); + Value* startInstance = LOAD(fetchInfo, {0, SWR_FETCH_CONTEXT_StartInstance}); + Value* curInstance = LOAD(fetchInfo, {0, SWR_FETCH_CONTEXT_CurInstance}); + Value* vBaseVertex = VBROADCAST(LOAD(fetchInfo, {0, SWR_FETCH_CONTEXT_BaseVertex})); + curInstance->setName("curInstance"); + + for(uint32_t nInputElt = 0; nInputElt < fetchState.numAttribs; ++nInputElt) + { + const INPUT_ELEMENT_DESC& ied = fetchState.layout[nInputElt]; + const SWR_FORMAT_INFO &info = GetFormatInfo((SWR_FORMAT)ied.Format); + uint32_t bpc = info.bpp / info.numComps; ///@todo Code below assumes all components are same size. Need to fix. + + Value *stream = LOAD(streams, {ied.StreamIndex, SWR_VERTEX_BUFFER_STATE_pData}); + + // VGATHER* takes an *i8 src pointer + Value* pStreamBase = BITCAST(stream, PointerType::get(mInt8Ty, 0)); + + Value *stride = LOAD(streams, {ied.StreamIndex, SWR_VERTEX_BUFFER_STATE_pitch}); + Value *vStride = VBROADCAST(stride); + + // max vertex index that is fully in bounds + Value *maxVertex = GEP(streams, {C(ied.StreamIndex), C(SWR_VERTEX_BUFFER_STATE_maxVertex)}); + maxVertex = LOAD(maxVertex); + + Value *vCurIndices; + Value *startOffset; + if(ied.InstanceEnable) + { + Value* stepRate = C(ied.InstanceDataStepRate); + + // prevent a div by 0 for 0 step rate + Value* isNonZeroStep = ICMP_UGT(stepRate, C(0)); + stepRate = SELECT(isNonZeroStep, stepRate, C(1)); + + // calc the current offset into instanced data buffer + Value* calcInstance = UDIV(curInstance, stepRate); + + // if step rate is 0, every instance gets instance 0 + calcInstance = SELECT(isNonZeroStep, calcInstance, C(0)); + + vCurIndices = VBROADCAST(calcInstance); + + startOffset = startInstance; + } + else + { + // offset indices by baseVertex + vCurIndices = ADD(vIndices, vBaseVertex); + + startOffset = startVertex; + } + + // All of the OOB calculations are in vertices, not VB offsets, to prevent having to + // do 64bit address offset calculations. + + // calculate byte offset to the start of the VB + Value* baseOffset = MUL(Z_EXT(startOffset, mInt64Ty), Z_EXT(stride, mInt64Ty)); + pStreamBase = GEP(pStreamBase, baseOffset); + + // if we have a start offset, subtract from max vertex. Used for OOB check + maxVertex = SUB(Z_EXT(maxVertex, mInt64Ty), Z_EXT(startOffset, mInt64Ty)); + Value* neg = ICMP_SLT(maxVertex, C((int64_t)0)); + // if we have a negative value, we're already OOB. clamp at 0. + maxVertex = SELECT(neg, C(0), TRUNC(maxVertex, mInt32Ty)); + + // Load the in bounds size of a partially valid vertex + Value *partialInboundsSize = GEP(streams, {C(ied.StreamIndex), C(SWR_VERTEX_BUFFER_STATE_partialInboundsSize)}); + partialInboundsSize = LOAD(partialInboundsSize); + Value* vPartialVertexSize = VBROADCAST(partialInboundsSize); + Value* vBpp = VBROADCAST(C(info.Bpp)); + Value* vAlignmentOffsets = VBROADCAST(C(ied.AlignedByteOffset)); + + // is the element is <= the partially valid size + Value* vElementInBoundsMask = ICMP_SLE(vBpp, SUB(vPartialVertexSize, vAlignmentOffsets)); + + // are vertices partially OOB? + Value* vMaxVertex = VBROADCAST(maxVertex); + Value* vPartialOOBMask = ICMP_EQ(vCurIndices, vMaxVertex); + + // are vertices are fully in bounds? + Value* vGatherMask = ICMP_ULT(vCurIndices, vMaxVertex); + + // blend in any partially OOB indices that have valid elements + vGatherMask = SELECT(vPartialOOBMask, vElementInBoundsMask, vGatherMask); + vGatherMask = VMASK(vGatherMask); + + // calculate the actual offsets into the VB + Value* vOffsets = MUL(vCurIndices, vStride); + vOffsets = ADD(vOffsets, vAlignmentOffsets); + + // Packing and component control + ComponentEnable compMask = (ComponentEnable)ied.ComponentPacking; + const ComponentControl compCtrl[4] { (ComponentControl)ied.ComponentControl0, (ComponentControl)ied.ComponentControl1, + (ComponentControl)ied.ComponentControl2, (ComponentControl)ied.ComponentControl3}; + + if(info.type[0] == SWR_TYPE_FLOAT) + { + ///@todo: support 64 bit vb accesses + Value* gatherSrc = VIMMED1(0.0f); + + // Gather components from memory to store in a simdvertex structure + switch(bpc) + { + case 16: + { + Value* vGatherResult[2]; + Value *vMask; + + // if we have at least one component out of x or y to fetch + if(isComponentEnabled(compMask, 0) || isComponentEnabled(compMask, 1)){ + // save mask as it is zero'd out after each gather + vMask = vGatherMask; + + vGatherResult[0] = GATHERPS(gatherSrc, pStreamBase, vOffsets, vMask, C((char)1)); + // e.g. result of first 8x32bit integer gather for 16bit components + // 256i - 0 1 2 3 4 5 6 7 + // xyxy xyxy xyxy xyxy xyxy xyxy xyxy xyxy + // + } + + // if we have at least one component out of z or w to fetch + if(isComponentEnabled(compMask, 2) || isComponentEnabled(compMask, 3)){ + // offset base to the next components(zw) in the vertex to gather + pStreamBase = GEP(pStreamBase, C((char)4)); + vMask = vGatherMask; + + vGatherResult[1] = GATHERPS(gatherSrc, pStreamBase, vOffsets, vMask, C((char)1)); + // e.g. result of second 8x32bit integer gather for 16bit components + // 256i - 0 1 2 3 4 5 6 7 + // zwzw zwzw zwzw zwzw zwzw zwzw zwzw zwzw + // + } + + // if we have at least one component to shuffle into place + if(compMask){ + Shuffle16bpcArgs args = std::forward_as_tuple(vGatherResult, pVtxOut, Instruction::CastOps::FPExt, CONVERT_NONE, + currentVertexElement, outputElt, compMask, compCtrl, vVertexElements); + // Shuffle gathered components into place in simdvertex struct + Shuffle16bpcGather(args); // outputs to vVertexElements ref + } + } + break; + case 32: + { + for(uint32_t i = 0; i < 4; i++) + { + if(!isComponentEnabled(compMask, i)){ + // offset base to the next component in the vertex to gather + pStreamBase = GEP(pStreamBase, C((char)4)); + continue; + } + + // if we need to gather the component + if(compCtrl[i] == StoreSrc){ + // save mask as it is zero'd out after each gather + Value *vMask = vGatherMask; + + // Gather a SIMD of vertices + vVertexElements[currentVertexElement++] = GATHERPS(gatherSrc, pStreamBase, vOffsets, vMask, C((char)1)); + } + else{ + vVertexElements[currentVertexElement++] = GenerateCompCtrlVector(compCtrl[i]); + } + + if(currentVertexElement > 3){ + StoreVertexElements(pVtxOut, outputElt++, 4, vVertexElements); + // reset to the next vVertexElement to output + currentVertexElement = 0; + } + + // offset base to the next component in the vertex to gather + pStreamBase = GEP(pStreamBase, C((char)4)); + } + } + break; + default: + SWR_ASSERT(0, "Tried to fetch invalid FP format"); + break; + } + } + else + { + Instruction::CastOps extendCastType = Instruction::CastOps::CastOpsEnd; + ConversionType conversionType = CONVERT_NONE; + + switch(info.type[0]) + { + case SWR_TYPE_UNORM: + conversionType = CONVERT_NORMALIZED; + case SWR_TYPE_UINT: + extendCastType = Instruction::CastOps::ZExt; + break; + case SWR_TYPE_SNORM: + conversionType = CONVERT_NORMALIZED; + case SWR_TYPE_SINT: + extendCastType = Instruction::CastOps::SExt; + break; + case SWR_TYPE_USCALED: + conversionType = CONVERT_USCALED; + extendCastType = Instruction::CastOps::UIToFP; + break; + case SWR_TYPE_SSCALED: + conversionType = CONVERT_SSCALED; + extendCastType = Instruction::CastOps::SIToFP; + break; + default: + break; + } + + // value substituted when component of gather is masked + Value* gatherSrc = VIMMED1(0); + + // Gather components from memory to store in a simdvertex structure + switch (bpc) + { + case 8: + { + // if we have at least one component to fetch + if(compMask){ + Value* vGatherResult = GATHERDD(gatherSrc, pStreamBase, vOffsets, vGatherMask, C((char)1)); + // e.g. result of an 8x32bit integer gather for 8bit components + // 256i - 0 1 2 3 4 5 6 7 + // xyzw xyzw xyzw xyzw xyzw xyzw xyzw xyzw + + Shuffle8bpcArgs args = std::forward_as_tuple(vGatherResult, pVtxOut, extendCastType, conversionType, + currentVertexElement, outputElt, compMask, compCtrl, vVertexElements, info.swizzle); + // Shuffle gathered components into place in simdvertex struct + Shuffle8bpcGatherd(args); // outputs to vVertexElements ref + } + } + break; + case 16: + { + Value* vGatherResult[2]; + Value *vMask; + + // if we have at least one component out of x or y to fetch + if(isComponentEnabled(compMask, 0) || isComponentEnabled(compMask, 1)){ + // save mask as it is zero'd out after each gather + vMask = vGatherMask; + + vGatherResult[0] = GATHERDD(gatherSrc, pStreamBase, vOffsets, vMask, C((char)1)); + // e.g. result of first 8x32bit integer gather for 16bit components + // 256i - 0 1 2 3 4 5 6 7 + // xyxy xyxy xyxy xyxy xyxy xyxy xyxy xyxy + // + } + + // if we have at least one component out of z or w to fetch + if(isComponentEnabled(compMask, 2) || isComponentEnabled(compMask, 3)){ + // offset base to the next components(zw) in the vertex to gather + pStreamBase = GEP(pStreamBase, C((char)4)); + vMask = vGatherMask; + + vGatherResult[1] = GATHERDD(gatherSrc, pStreamBase, vOffsets, vMask, C((char)1)); + // e.g. result of second 8x32bit integer gather for 16bit components + // 256i - 0 1 2 3 4 5 6 7 + // zwzw zwzw zwzw zwzw zwzw zwzw zwzw zwzw + // + } + + // if we have at least one component to shuffle into place + if(compMask){ + Shuffle16bpcArgs args = std::forward_as_tuple(vGatherResult, pVtxOut, extendCastType, conversionType, + currentVertexElement, outputElt, compMask, compCtrl, vVertexElements); + // Shuffle gathered components into place in simdvertex struct + Shuffle16bpcGather(args); // outputs to vVertexElements ref + } + } + break; + case 32: + { + SWR_ASSERT(conversionType == CONVERT_NONE); + + // Gathered components into place in simdvertex struct + for(uint32_t i = 0; i < 4; i++) + { + if(!isComponentEnabled(compMask, i)){ + // offset base to the next component in the vertex to gather + pStreamBase = GEP(pStreamBase, C((char)4)); + continue; + } + + // if we need to gather the component + if(compCtrl[i] == StoreSrc){ + // save mask as it is zero'd out after each gather + Value *vMask = vGatherMask; + + vVertexElements[currentVertexElement++] = GATHERDD(gatherSrc, pStreamBase, vOffsets, vMask, C((char)1)); + + // e.g. result of a single 8x32bit integer gather for 32bit components + // 256i - 0 1 2 3 4 5 6 7 + // xxxx xxxx xxxx xxxx xxxx xxxx xxxx xxxx + } + else{ + vVertexElements[currentVertexElement++] = GenerateCompCtrlVector(compCtrl[i]); + } + + if(currentVertexElement > 3){ + StoreVertexElements(pVtxOut, outputElt++, 4, vVertexElements); + // reset to the next vVertexElement to output + currentVertexElement = 0; + } + + // offset base to the next component in the vertex to gather + pStreamBase = GEP(pStreamBase, C((char)4)); + } + } + break; + } + } + } + + // if we have a partially filled vVertexElement struct, output it + if(currentVertexElement > 0){ + StoreVertexElements(pVtxOut, outputElt++, currentVertexElement+1, vVertexElements); + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Loads a simd of valid indices. OOB indices are set to 0 +/// *Note* have to do 16bit index checking in scalar until we have AVX-512 +/// support +/// @param pIndices - pointer to 8 bit indices +/// @param pLastIndex - pointer to last valid index +Value* FetchJit::GetSimdValid8bitIndices(Value* pIndices, Value* pLastIndex) +{ + // can fit 2 16 bit integers per vWidth lane + Value* vIndices = VUNDEF_I(); + + // store 0 index on stack to be used to conditionally load from if index address is OOB + Value* pZeroIndex = ALLOCA(mInt8Ty); + STORE(C((uint8_t)0), pZeroIndex); + + // Load a SIMD of index pointers + for(int64_t lane = 0; lane < JM()->mVWidth; lane++) + { + // Calculate the address of the requested index + Value *pIndex = GEP(pIndices, C(lane)); + + // check if the address is less than the max index, + Value* mask = ICMP_ULT(pIndex, pLastIndex); + + // if valid, load the index. if not, load 0 from the stack + Value* pValid = SELECT(mask, pIndex, pZeroIndex); + Value *index = LOAD(pValid, "valid index"); + + // zero extended index to 32 bits and insert into the correct simd lane + index = Z_EXT(index, mInt32Ty); + vIndices = VINSERT(vIndices, index, lane); + } + return vIndices; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Loads a simd of valid indices. OOB indices are set to 0 +/// *Note* have to do 16bit index checking in scalar until we have AVX-512 +/// support +/// @param pIndices - pointer to 16 bit indices +/// @param pLastIndex - pointer to last valid index +Value* FetchJit::GetSimdValid16bitIndices(Value* pIndices, Value* pLastIndex) +{ + // can fit 2 16 bit integers per vWidth lane + Value* vIndices = VUNDEF_I(); + + // store 0 index on stack to be used to conditionally load from if index address is OOB + Value* pZeroIndex = ALLOCA(mInt16Ty); + STORE(C((uint16_t)0), pZeroIndex); + + // Load a SIMD of index pointers + for(int64_t lane = 0; lane < JM()->mVWidth; lane++) + { + // Calculate the address of the requested index + Value *pIndex = GEP(pIndices, C(lane)); + + // check if the address is less than the max index, + Value* mask = ICMP_ULT(pIndex, pLastIndex); + + // if valid, load the index. if not, load 0 from the stack + Value* pValid = SELECT(mask, pIndex, pZeroIndex); + Value *index = LOAD(pValid, "valid index"); + + // zero extended index to 32 bits and insert into the correct simd lane + index = Z_EXT(index, mInt32Ty); + vIndices = VINSERT(vIndices, index, lane); + } + return vIndices; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Loads a simd of valid indices. OOB indices are set to 0 +/// @param pIndices - pointer to 32 bit indices +/// @param pLastIndex - pointer to last valid index +Value* FetchJit::GetSimdValid32bitIndices(Value* pIndices, Value* pLastIndex) +{ + DataLayout dL(JM()->mpCurrentModule); + unsigned int ptrSize = dL.getPointerSize() * 8; // ptr size in bits + Value* iLastIndex = PTR_TO_INT(pLastIndex, Type::getIntNTy(JM()->mContext, ptrSize)); + Value* iIndices = PTR_TO_INT(pIndices, Type::getIntNTy(JM()->mContext, ptrSize)); + + // get the number of indices left in the buffer (endPtr - curPtr) / sizeof(index) + Value* numIndicesLeft = SUB(iLastIndex,iIndices); + numIndicesLeft = TRUNC(numIndicesLeft, mInt32Ty); + numIndicesLeft = SDIV(numIndicesLeft, C(4)); + + // create a vector of index counts from the base index ptr passed into the fetch + const std::vector vecIndices {C(0), C(1), C(2), C(3), C(4), C(5), C(6), C(7)}; + Constant* vIndexOffsets = ConstantVector::get(vecIndices); + + // compare index count to the max valid index + // e.g vMaxIndex 4 4 4 4 4 4 4 4 : 4 indices left to load + // vIndexOffsets 0 1 2 3 4 5 6 7 + // ------------------------------ + // vIndexMask -1-1-1-1 0 0 0 0 : offsets < max pass + // vLoadedIndices 0 1 2 3 0 0 0 0 : offsets >= max masked to 0 + Value* vMaxIndex = VBROADCAST(numIndicesLeft); + Value* vIndexMask = VPCMPGTD(vMaxIndex,vIndexOffsets); + + // VMASKLOAD takes an *i8 src pointer + pIndices = BITCAST(pIndices,PointerType::get(mInt8Ty,0)); + + // Load the indices; OOB loads 0 + return MASKLOADD(pIndices,vIndexMask); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Takes a SIMD of gathered 8bpc verts, zero or sign extends, +/// denormalizes if needed, converts to F32 if needed, and positions in +// the proper SIMD rows to be output to the simdvertex structure +/// @param args: (tuple of args, listed below) +/// @param vGatherResult - 8 gathered 8bpc vertices +/// @param pVtxOut - base pointer to output simdvertex struct +/// @param extendType - sign extend or zero extend +/// @param bNormalized - do we need to denormalize? +/// @param currentVertexElement - reference to the current vVertexElement +/// @param outputElt - reference to the current offset from simdvertex we're o +/// @param compMask - component packing mask +/// @param compCtrl - component control val +/// @param vVertexElements[4] - vertex components to output +/// @param swizzle[4] - component swizzle location +void FetchJit::Shuffle8bpcGatherd(Shuffle8bpcArgs &args) +{ + // Unpack tuple args + Value*& vGatherResult = std::get<0>(args); + Value* pVtxOut = std::get<1>(args); + const Instruction::CastOps extendType = std::get<2>(args); + const ConversionType conversionType = std::get<3>(args); + uint32_t ¤tVertexElement = std::get<4>(args); + uint32_t &outputElt = std::get<5>(args); + const ComponentEnable compMask = std::get<6>(args); + const ComponentControl (&compCtrl)[4] = std::get<7>(args); + Value* (&vVertexElements)[4] = std::get<8>(args); + const uint32_t (&swizzle)[4] = std::get<9>(args); + + // cast types + Type* vGatherTy = VectorType::get(IntegerType::getInt32Ty(JM()->mContext), JM()->mVWidth); + Type* v32x8Ty = VectorType::get(mInt8Ty, JM()->mVWidth * 4 ); // vwidth is units of 32 bits + + // have to do extra work for sign extending + if ((extendType == Instruction::CastOps::SExt) || (extendType == Instruction::CastOps::SIToFP)){ + Type* v16x8Ty = VectorType::get(mInt8Ty, JM()->mVWidth * 2); // 8x16bit ints in a 128bit lane + Type* v128Ty = VectorType::get(IntegerType::getIntNTy(JM()->mContext, 128), JM()->mVWidth / 4); // vwidth is units of 32 bits + + // shuffle mask, including any swizzling + const char x = (char)swizzle[0]; const char y = (char)swizzle[1]; + const char z = (char)swizzle[2]; const char w = (char)swizzle[3]; + Value* vConstMask = C({char(x), char(x+4), char(x+8), char(x+12), + char(y), char(y+4), char(y+8), char(y+12), + char(z), char(z+4), char(z+8), char(z+12), + char(w), char(w+4), char(w+8), char(w+12), + char(x), char(x+4), char(x+8), char(x+12), + char(y), char(y+4), char(y+8), char(y+12), + char(z), char(z+4), char(z+8), char(z+12), + char(w), char(w+4), char(w+8), char(w+12)}); + + Value* vShufResult = BITCAST(PSHUFB(BITCAST(vGatherResult, v32x8Ty), vConstMask), vGatherTy); + // after pshufb: group components together in each 128bit lane + // 256i - 0 1 2 3 4 5 6 7 + // xxxx yyyy zzzz wwww xxxx yyyy zzzz wwww + + Value* vi128XY = nullptr; + if(isComponentEnabled(compMask, 0) || isComponentEnabled(compMask, 1)){ + vi128XY = BITCAST(PERMD(vShufResult, C({0, 4, 0, 0, 1, 5, 0, 0})), v128Ty); + // after PERMD: move and pack xy and zw components in low 64 bits of each 128bit lane + // 256i - 0 1 2 3 4 5 6 7 + // xxxx xxxx dcdc dcdc yyyy yyyy dcdc dcdc (dc - don't care) + } + + // do the same for zw components + Value* vi128ZW = nullptr; + if(isComponentEnabled(compMask, 2) || isComponentEnabled(compMask, 3)){ + vi128ZW = BITCAST(PERMD(vShufResult, C({2, 6, 0, 0, 3, 7, 0, 0})), v128Ty); + } + + // init denormalize variables if needed + Instruction::CastOps fpCast; + Value* conversionFactor; + + switch (conversionType) + { + case CONVERT_NORMALIZED: + fpCast = Instruction::CastOps::SIToFP; + conversionFactor = VIMMED1((float)(1.0 / 127.0)); + break; + case CONVERT_SSCALED: + fpCast = Instruction::CastOps::SIToFP; + conversionFactor = VIMMED1((float)(1.0)); + break; + case CONVERT_USCALED: + SWR_ASSERT(0, "Type should not be sign extended!"); + conversionFactor = nullptr; + break; + default: + SWR_ASSERT(conversionType == CONVERT_NONE); + conversionFactor = nullptr; + break; + } + + // sign extend all enabled components. If we have a fill vVertexElements, output to current simdvertex + for(uint32_t i = 0; i < 4; i++){ + if(!isComponentEnabled(compMask, i)){ + continue; + } + + if(compCtrl[i] == ComponentControl::StoreSrc){ + // if x or z, extract 128bits from lane 0, else for y or w, extract from lane 1 + uint32_t lane = ((i == 0) || (i == 2)) ? 0 : 1; + // if x or y, use vi128XY permute result, else use vi128ZW + Value* selectedPermute = (i < 2) ? vi128XY : vi128ZW; + + // sign extend + vVertexElements[currentVertexElement] = PMOVSXBD(BITCAST(VEXTRACT(selectedPermute, C(lane)), v16x8Ty)); + + // denormalize if needed + if(conversionType != CONVERT_NONE){ + vVertexElements[currentVertexElement] = FMUL(CAST(fpCast, vVertexElements[currentVertexElement], mSimdFP32Ty), conversionFactor); + } + currentVertexElement++; + } + else{ + vVertexElements[currentVertexElement++] = GenerateCompCtrlVector(compCtrl[i]); + } + + if(currentVertexElement > 3){ + StoreVertexElements(pVtxOut, outputElt++, 4, vVertexElements); + // reset to the next vVertexElement to output + currentVertexElement = 0; + } + } + } + // else zero extend + else if ((extendType == Instruction::CastOps::ZExt) || (extendType == Instruction::CastOps::UIToFP)) + { + // init denormalize variables if needed + Instruction::CastOps fpCast; + Value* conversionFactor; + + switch (conversionType) + { + case CONVERT_NORMALIZED: + fpCast = Instruction::CastOps::UIToFP; + conversionFactor = VIMMED1((float)(1.0 / 255.0)); + break; + case CONVERT_USCALED: + fpCast = Instruction::CastOps::UIToFP; + conversionFactor = VIMMED1((float)(1.0)); + break; + case CONVERT_SSCALED: + SWR_ASSERT(0, "Type should not be zero extended!"); + conversionFactor = nullptr; + break; + default: + SWR_ASSERT(conversionType == CONVERT_NONE); + conversionFactor = nullptr; + break; + } + + // shuffle enabled components into lower byte of each 32bit lane, 0 extending to 32 bits + for(uint32_t i = 0; i < 4; i++){ + if(!isComponentEnabled(compMask, i)){ + continue; + } + + if(compCtrl[i] == ComponentControl::StoreSrc){ + // pshufb masks for each component + Value* vConstMask; + switch(swizzle[i]){ + case 0: + // x shuffle mask + vConstMask = C({0, -1, -1, -1, 4, -1, -1, -1, 8, -1, -1, -1, 12, -1, -1, -1, + 0, -1, -1, -1, 4, -1, -1, -1, 8, -1, -1, -1, 12, -1, -1, -1}); + break; + case 1: + // y shuffle mask + vConstMask = C({1, -1, -1, -1, 5, -1, -1, -1, 9, -1, -1, -1, 13, -1, -1, -1, + 1, -1, -1, -1, 5, -1, -1, -1, 9, -1, -1, -1, 13, -1, -1, -1}); + break; + case 2: + // z shuffle mask + vConstMask = C({2, -1, -1, -1, 6, -1, -1, -1, 10, -1, -1, -1, 14, -1, -1, -1, + 2, -1, -1, -1, 6, -1, -1, -1, 10, -1, -1, -1, 14, -1, -1, -1}); + break; + case 3: + // w shuffle mask + vConstMask = C({3, -1, -1, -1, 7, -1, -1, -1, 11, -1, -1, -1, 15, -1, -1, -1, + 3, -1, -1, -1, 7, -1, -1, -1, 11, -1, -1, -1, 15, -1, -1, -1}); + break; + default: + vConstMask = nullptr; + break; + } + + vVertexElements[currentVertexElement] = BITCAST(PSHUFB(BITCAST(vGatherResult, v32x8Ty), vConstMask), vGatherTy); + // after pshufb for x channel + // 256i - 0 1 2 3 4 5 6 7 + // x000 x000 x000 x000 x000 x000 x000 x000 + + // denormalize if needed + if (conversionType != CONVERT_NONE){ + vVertexElements[currentVertexElement] = FMUL(CAST(fpCast, vVertexElements[currentVertexElement], mSimdFP32Ty), conversionFactor); + } + currentVertexElement++; + } + else{ + vVertexElements[currentVertexElement++] = GenerateCompCtrlVector(compCtrl[i]); + } + + if(currentVertexElement > 3){ + StoreVertexElements(pVtxOut, outputElt++, 4, vVertexElements); + // reset to the next vVertexElement to output + currentVertexElement = 0; + } + } + } + else + { + SWR_ASSERT(0, "Unsupported conversion type"); + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Takes a SIMD of gathered 16bpc verts, zero or sign extends, +/// denormalizes if needed, converts to F32 if needed, and positions in +// the proper SIMD rows to be output to the simdvertex structure +/// @param args: (tuple of args, listed below) +/// @param vGatherResult[2] - array of gathered 16bpc vertices, 4 per index +/// @param pVtxOut - base pointer to output simdvertex struct +/// @param extendType - sign extend or zero extend +/// @param bNormalized - do we need to denormalize? +/// @param currentVertexElement - reference to the current vVertexElement +/// @param outputElt - reference to the current offset from simdvertex we're o +/// @param compMask - component packing mask +/// @param compCtrl - component control val +/// @param vVertexElements[4] - vertex components to output +void FetchJit::Shuffle16bpcGather(Shuffle16bpcArgs &args) +{ + // Unpack tuple args + Value* (&vGatherResult)[2] = std::get<0>(args); + Value* pVtxOut = std::get<1>(args); + const Instruction::CastOps extendType = std::get<2>(args); + const ConversionType conversionType = std::get<3>(args); + uint32_t ¤tVertexElement = std::get<4>(args); + uint32_t &outputElt = std::get<5>(args); + const ComponentEnable compMask = std::get<6>(args); + const ComponentControl(&compCtrl)[4] = std::get<7>(args); + Value* (&vVertexElements)[4] = std::get<8>(args); + + // cast types + Type* vGatherTy = VectorType::get(IntegerType::getInt32Ty(JM()->mContext), JM()->mVWidth); + Type* v32x8Ty = VectorType::get(mInt8Ty, JM()->mVWidth * 4); // vwidth is units of 32 bits + + // have to do extra work for sign extending + if ((extendType == Instruction::CastOps::SExt) || (extendType == Instruction::CastOps::SIToFP)|| + (extendType == Instruction::CastOps::FPExt)) + { + // is this PP float? + bool bFP = (extendType == Instruction::CastOps::FPExt) ? true : false; + + Type* v8x16Ty = VectorType::get(mInt16Ty, 8); // 8x16bit in a 128bit lane + Type* v128bitTy = VectorType::get(IntegerType::getIntNTy(JM()->mContext, 128), JM()->mVWidth / 4); // vwidth is units of 32 bits + + // shuffle mask + Value* vConstMask = C({0, 1, 4, 5, 8, 9, 12, 13, 2, 3, 6, 7, 10, 11, 14, 15, + 0, 1, 4, 5, 8, 9, 12, 13, 2, 3, 6, 7, 10, 11, 14, 15}); + Value* vi128XY = nullptr; + if(isComponentEnabled(compMask, 0) || isComponentEnabled(compMask, 1)){ + Value* vShufResult = BITCAST(PSHUFB(BITCAST(vGatherResult[0], v32x8Ty), vConstMask), vGatherTy); + // after pshufb: group components together in each 128bit lane + // 256i - 0 1 2 3 4 5 6 7 + // xxxx xxxx yyyy yyyy xxxx xxxx yyyy yyyy + + vi128XY = BITCAST(PERMD(vShufResult, C({0, 1, 4, 5, 2, 3, 6, 7})), v128bitTy); + // after PERMD: move and pack xy components into each 128bit lane + // 256i - 0 1 2 3 4 5 6 7 + // xxxx xxxx xxxx xxxx yyyy yyyy yyyy yyyy + } + + // do the same for zw components + Value* vi128ZW = nullptr; + if(isComponentEnabled(compMask, 2) || isComponentEnabled(compMask, 3)){ + Value* vShufResult = BITCAST(PSHUFB(BITCAST(vGatherResult[1], v32x8Ty), vConstMask), vGatherTy); + vi128ZW = BITCAST(PERMD(vShufResult, C({0, 1, 4, 5, 2, 3, 6, 7})), v128bitTy); + } + + // init denormalize variables if needed + Instruction::CastOps IntToFpCast; + Value* conversionFactor; + + switch (conversionType) + { + case CONVERT_NORMALIZED: + IntToFpCast = Instruction::CastOps::SIToFP; + conversionFactor = VIMMED1((float)(1.0 / 32767.0)); + break; + case CONVERT_SSCALED: + IntToFpCast = Instruction::CastOps::SIToFP; + conversionFactor = VIMMED1((float)(1.0)); + break; + case CONVERT_USCALED: + SWR_ASSERT(0, "Type should not be sign extended!"); + conversionFactor = nullptr; + break; + default: + SWR_ASSERT(conversionType == CONVERT_NONE); + conversionFactor = nullptr; + break; + } + + // sign extend all enabled components. If we have a fill vVertexElements, output to current simdvertex + for(uint32_t i = 0; i < 4; i++){ + if(!isComponentEnabled(compMask, i)){ + continue; + } + + if(compCtrl[i] == ComponentControl::StoreSrc){ + // if x or z, extract 128bits from lane 0, else for y or w, extract from lane 1 + uint32_t lane = ((i == 0) || (i == 2)) ? 0 : 1; + // if x or y, use vi128XY permute result, else use vi128ZW + Value* selectedPermute = (i < 2) ? vi128XY : vi128ZW; + + if(bFP) { + // extract 128 bit lanes to sign extend each component + vVertexElements[currentVertexElement] = CVTPH2PS(BITCAST(VEXTRACT(selectedPermute, C(lane)), v8x16Ty)); + } + else { + // extract 128 bit lanes to sign extend each component + vVertexElements[currentVertexElement] = PMOVSXWD(BITCAST(VEXTRACT(selectedPermute, C(lane)), v8x16Ty)); + + // denormalize if needed + if(conversionType != CONVERT_NONE){ + vVertexElements[currentVertexElement] = FMUL(CAST(IntToFpCast, vVertexElements[currentVertexElement], mSimdFP32Ty), conversionFactor); + } + } + currentVertexElement++; + } + else{ + vVertexElements[currentVertexElement++] = GenerateCompCtrlVector(compCtrl[i]); + } + + if(currentVertexElement > 3){ + StoreVertexElements(pVtxOut, outputElt++, 4, vVertexElements); + // reset to the next vVertexElement to output + currentVertexElement = 0; + } + } + + } + // else zero extend + else if ((extendType == Instruction::CastOps::ZExt) || (extendType == Instruction::CastOps::UIToFP)) + { + // pshufb masks for each component + Value* vConstMask[2]; + if(isComponentEnabled(compMask, 0) || isComponentEnabled(compMask, 2)){ + // x/z shuffle mask + vConstMask[0] = C({0, 1, -1, -1, 4, 5, -1, -1, 8, 9, -1, -1, 12, 13, -1, -1, + 0, 1, -1, -1, 4, 5, -1, -1, 8, 9, -1, -1, 12, 13, -1, -1, }); + } + + if(isComponentEnabled(compMask, 1) || isComponentEnabled(compMask, 3)){ + // y/w shuffle mask + vConstMask[1] = C({2, 3, -1, -1, 6, 7, -1, -1, 10, 11, -1, -1, 14, 15, -1, -1, + 2, 3, -1, -1, 6, 7, -1, -1, 10, 11, -1, -1, 14, 15, -1, -1}); + } + + // init denormalize variables if needed + Instruction::CastOps fpCast; + Value* conversionFactor; + + switch (conversionType) + { + case CONVERT_NORMALIZED: + fpCast = Instruction::CastOps::UIToFP; + conversionFactor = VIMMED1((float)(1.0 / 65535.0)); + break; + case CONVERT_USCALED: + fpCast = Instruction::CastOps::UIToFP; + conversionFactor = VIMMED1((float)(1.0f)); + break; + case CONVERT_SSCALED: + SWR_ASSERT(0, "Type should not be zero extended!"); + conversionFactor = nullptr; + break; + default: + SWR_ASSERT(conversionType == CONVERT_NONE); + conversionFactor = nullptr; + break; + } + + // shuffle enabled components into lower word of each 32bit lane, 0 extending to 32 bits + for(uint32_t i = 0; i < 4; i++){ + if(!isComponentEnabled(compMask, i)){ + continue; + } + + if(compCtrl[i] == ComponentControl::StoreSrc){ + // select correct constMask for x/z or y/w pshufb + uint32_t selectedMask = ((i == 0) || (i == 2)) ? 0 : 1; + // if x or y, use vi128XY permute result, else use vi128ZW + uint32_t selectedGather = (i < 2) ? 0 : 1; + + vVertexElements[currentVertexElement] = BITCAST(PSHUFB(BITCAST(vGatherResult[selectedGather], v32x8Ty), vConstMask[selectedMask]), vGatherTy); + // after pshufb mask for x channel; z uses the same shuffle from the second gather + // 256i - 0 1 2 3 4 5 6 7 + // xx00 xx00 xx00 xx00 xx00 xx00 xx00 xx00 + + // denormalize if needed + if(conversionType != CONVERT_NONE){ + vVertexElements[currentVertexElement] = FMUL(CAST(fpCast, vVertexElements[currentVertexElement], mSimdFP32Ty), conversionFactor); + } + currentVertexElement++; + } + else{ + vVertexElements[currentVertexElement++] = GenerateCompCtrlVector(compCtrl[i]); + } + + if(currentVertexElement > 3){ + StoreVertexElements(pVtxOut, outputElt++, 4, vVertexElements); + // reset to the next vVertexElement to output + currentVertexElement = 0; + } + } + } + else + { + SWR_ASSERT(0, "Unsupported conversion type"); + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Output a simdvertex worth of elements to the current outputElt +/// @param pVtxOut - base address of VIN output struct +/// @param outputElt - simdvertex offset in VIN to write to +/// @param numEltsToStore - number of simdvertex rows to write out +/// @param vVertexElements - LLVM Value*[] simdvertex to write out +void FetchJit::StoreVertexElements(Value* pVtxOut, const uint32_t outputElt, const uint32_t numEltsToStore, Value* (&vVertexElements)[4]) +{ + for(uint32_t c = 0; c < numEltsToStore; ++c) + { + // STORE expects FP32 x vWidth type, just bitcast if needed + if(!vVertexElements[c]->getType()->getScalarType()->isFloatTy()){ +#if FETCH_DUMP_VERTEX + PRINT("vVertexElements[%d]: 0x%x\n", {C(c), vVertexElements[c]}); +#endif + vVertexElements[c] = BITCAST(vVertexElements[c], mSimdFP32Ty); + } +#if FETCH_DUMP_VERTEX + else + { + PRINT("vVertexElements[%d]: %f\n", {C(c), vVertexElements[c]}); + } +#endif + // outputElt * 4 = offsetting by the size of a simdvertex + // + c offsets to a 32bit x vWidth row within the current vertex + Value* dest = GEP(pVtxOut, C(outputElt * 4 + c), "destGEP"); + STORE(vVertexElements[c], dest); + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Generates a constant vector of values based on the +/// ComponentControl value +/// @param ctrl - ComponentControl value +Value* FetchJit::GenerateCompCtrlVector(const ComponentControl ctrl) +{ + switch(ctrl) + { + case NoStore: return VUNDEF_I(); + case Store0: return VIMMED1(0); + case Store1Fp: return VIMMED1(1.0f); + case Store1Int: return VIMMED1(1); + case StoreSrc: + default: SWR_ASSERT(0, "Invalid component control"); return VUNDEF_I(); + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Returns the enable mask for the specified component. +/// @param enableMask - enable bits +/// @param component - component to check if enabled. +bool isComponentEnabled(ComponentEnable enableMask, uint8_t component) +{ + switch (component) + { + // X + case 0: return (enableMask & ComponentEnable::X); + // Y + case 1: return (enableMask & ComponentEnable::Y); + // Z + case 2: return (enableMask & ComponentEnable::Z); + // W + case 3: return (enableMask & ComponentEnable::W); + + default: return false; + } +} + + +////////////////////////////////////////////////////////////////////////// +/// @brief JITs from fetch shader IR +/// @param hJitMgr - JitManager handle +/// @param func - LLVM function IR +/// @return PFN_FETCH_FUNC - pointer to fetch code +PFN_FETCH_FUNC JitFetchFunc(HANDLE hJitMgr, const HANDLE hFunc) +{ + const llvm::Function* func = (const llvm::Function*)hFunc; + JitManager* pJitMgr = reinterpret_cast(hJitMgr); + PFN_FETCH_FUNC pfnFetch; + + pfnFetch = (PFN_FETCH_FUNC)(pJitMgr->mpExec->getFunctionAddress(func->getName().str())); + // MCJIT finalizes modules the first time you JIT code from them. After finalized, you cannot add new IR to the module + pJitMgr->mIsModuleFinalized = true; + +#if defined(KNOB_SWRC_TRACING) + char fName[1024]; + const char *funcName = func->getName().data(); + sprintf(fName, "%s.bin", funcName); + FILE *fd = fopen(fName, "wb"); + fwrite((void *)pfnFetch, 1, 2048, fd); + fclose(fd); +#endif + + return pfnFetch; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief JIT compiles fetch shader +/// @param hJitMgr - JitManager handle +/// @param state - fetch state to build function from +extern "C" PFN_FETCH_FUNC JITCALL JitCompileFetch(HANDLE hJitMgr, const FETCH_COMPILE_STATE& state) +{ + JitManager* pJitMgr = reinterpret_cast(hJitMgr); + + pJitMgr->SetupNewModule(); + + FetchJit theJit(pJitMgr); + HANDLE hFunc = theJit.Create(state); + + return JitFetchFunc(hJitMgr, hFunc); +} diff --git a/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.h b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.h new file mode 100644 index 00000000000..ea3625d2fde --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/fetch_jit.h @@ -0,0 +1,128 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file fetch_jit.h +* +* @brief Definition of the fetch jitter +* +* Notes: +* +******************************************************************************/ +#pragma once + +#include "common/formats.h" +#include "core/state.h" + +////////////////////////////////////////////////////////////////////////// +/// INPUT_ELEMENT_DESC +////////////////////////////////////////////////////////////////////////// +struct INPUT_ELEMENT_DESC +{ + union + { + struct + { + uint32_t AlignedByteOffset : 12; + uint32_t Format : 10; + uint32_t StreamIndex : 6; + uint32_t InstanceEnable : 1; + uint32_t ComponentControl0 : 3; + uint32_t ComponentControl1 : 3; + uint32_t ComponentControl2 : 3; + uint32_t ComponentControl3 : 3; + uint32_t ComponentPacking : 4; + uint32_t _reserved : 19; + }; + uint64_t bits; + }; + uint32_t InstanceDataStepRate; +}; + +// used to set ComponentPacking +enum ComponentEnable +{ + NONE = 0x0, + X = 0x1, + Y = 0x2, + XY = 0x3, + Z = 0x4, + XZ = 0x5, + YZ = 0x6, + XYZ = 0x7, + W = 0x8, + XW = 0x9, + YW = 0xA, + XYW = 0xB, + ZW = 0xC, + XZW = 0xD, + YZW = 0xE, + XYZW = 0xF, +}; + +enum ComponentControl +{ + NoStore = 0, + StoreSrc = 1, + Store0 = 2, + Store1Fp = 3, + Store1Int = 4, +}; + +////////////////////////////////////////////////////////////////////////// +/// State required for fetch shader jit compile. +////////////////////////////////////////////////////////////////////////// +struct FETCH_COMPILE_STATE +{ + uint32_t numAttribs; + INPUT_ELEMENT_DESC layout[KNOB_NUM_ATTRIBUTES]; + SWR_FORMAT indexType; + uint32_t cutIndex{ 0xffffffff }; + + // Options that effect the JIT'd code + bool bDisableVGATHER; // if enabled, FetchJit will generate loads/shuffles instead of VGATHERs + bool bDisableIndexOOBCheck; // if enabled, FetchJit will exclude index OOB check + bool bEnableCutIndex{ false }; // compares indices with the cut index and returns a cut mask + + FETCH_COMPILE_STATE(bool useVGATHER = false, bool indexOOBCheck = false) : + bDisableVGATHER(useVGATHER), bDisableIndexOOBCheck(indexOOBCheck){}; + + bool operator==(const FETCH_COMPILE_STATE &other) const + { + if (numAttribs != other.numAttribs) return false; + if (indexType != other.indexType) return false; + if (bDisableVGATHER != other.bDisableVGATHER) return false; + if (bDisableIndexOOBCheck != other.bDisableIndexOOBCheck) return false; + if (bEnableCutIndex != other.bEnableCutIndex) return false; + if (cutIndex != other.cutIndex) return false; + + for(uint32_t i = 0; i < numAttribs; ++i) + { + if((layout[i].bits != other.layout[i].bits) || + ((layout[i].InstanceEnable == 1) && + (layout[i].InstanceDataStepRate != other.layout[i].InstanceDataStepRate))){ + return false; + } + } + + return true; + } +}; diff --git a/src/gallium/drivers/swr/rasterizer/jitter/jit_api.h b/src/gallium/drivers/swr/rasterizer/jitter/jit_api.h new file mode 100644 index 00000000000..39d63836673 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/jit_api.h @@ -0,0 +1,108 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file jit_api.h +* +* @brief Platform independent JIT interface +* +* Notes: +* +******************************************************************************/ +#pragma once +#include "common/os.h" + +#include "fetch_jit.h" +#include "streamout_jit.h" +#include "blend_jit.h" + +#if defined(_WIN32) +#define EXCEPTION_PRINT_STACK(ret) ret +#endif // _WIN32 + +#if defined(_WIN32) +#define JITCALL __stdcall +#else +#define JITCALL +#endif + +extern "C" +{ + +struct ShaderInfo; + +////////////////////////////////////////////////////////////////////////// +/// Jit Compile Info Input +////////////////////////////////////////////////////////////////////////// +struct JIT_COMPILE_INPUT +{ + SWR_SHADER_TYPE type; + + const void* pIR; ///< Pointer to LLVM IR text. + + bool enableJitSampler; +}; + +////////////////////////////////////////////////////////////////////////// +/// @brief Create JIT context. +HANDLE JITCALL JitCreateContext(uint32_t targetSimdWidth, const char* arch); + +////////////////////////////////////////////////////////////////////////// +/// @brief Destroy JIT context. +void JITCALL JitDestroyContext(HANDLE hJitContext); + +////////////////////////////////////////////////////////////////////////// +/// @brief JIT compile shader. +/// @param hJitContext - Jit Context +/// @param input - Input containing LLVM IR and other information +/// @param output - Output containing information about JIT shader +ShaderInfo* JITCALL JitCompileShader( + HANDLE hJitContext, + const JIT_COMPILE_INPUT& input); + +////////////////////////////////////////////////////////////////////////// +/// @brief JIT destroy shader. +/// @param hJitContext - Jit Context +/// @param pShaderInfo - pointer to shader object. +void JITCALL JitDestroyShader( + HANDLE hJitContext, + ShaderInfo*& pShaderInfo); + +////////////////////////////////////////////////////////////////////////// +/// @brief JIT compiles fetch shader +/// @param hJitContext - Jit Context +/// @param state - Fetch state to build function from +PFN_FETCH_FUNC JITCALL JitCompileFetch(HANDLE hJitContext, const FETCH_COMPILE_STATE& state); + +////////////////////////////////////////////////////////////////////////// +/// @brief JIT compiles streamout shader +/// @param hJitContext - Jit Context +/// @param state - SO state to build function from +PFN_SO_FUNC JITCALL JitCompileStreamout(HANDLE hJitContext, const STREAMOUT_COMPILE_STATE& state); + +////////////////////////////////////////////////////////////////////////// +/// @brief JIT compiles blend shader +/// @param hJitContext - Jit Context +/// @param state - blend state to build function from +PFN_BLEND_JIT_FUNC JITCALL JitCompileBlend(HANDLE hJitContext, const BLEND_COMPILE_STATE& state); + + +}; // extern "C" diff --git a/src/gallium/drivers/swr/rasterizer/jitter/scripts/gen_llvm_ir_macros.py b/src/gallium/drivers/swr/rasterizer/jitter/scripts/gen_llvm_ir_macros.py new file mode 100644 index 00000000000..1814b7c8d5f --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/scripts/gen_llvm_ir_macros.py @@ -0,0 +1,401 @@ +# Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +# +# Permission is hereby granted, free of charge, to any person obtaining a +# copy of this software and associated documentation files (the "Software"), +# to deal in the Software without restriction, including without limitation +# the rights to use, copy, modify, merge, publish, distribute, sublicense, +# and/or sell copies of the Software, and to permit persons to whom the +# Software is furnished to do so, subject to the following conditions: +# +# The above copyright notice and this permission notice (including the next +# paragraph) shall be included in all copies or substantial portions of the +# Software. +# +# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +# IN THE SOFTWARE. + +#!deps/python32/python.exe + +import os, sys, re +import argparse +import json as JSON +import operator + +header = r"""/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file %s +* +* @brief auto-generated file +* +* DO NOT EDIT +* +******************************************************************************/ + +""" + +""" +""" +def gen_file_header(filename): + global header + headerStr = header % filename + return headerStr.splitlines() + + +inst_aliases = { + 'SHUFFLE_VECTOR': 'VSHUFFLE', + 'INSERT_ELEMENT': 'VINSERT', + 'EXTRACT_ELEMENT': 'VEXTRACT', + 'MEM_SET': 'MEMSET', + 'MEM_CPY': 'MEMCPY', + 'MEM_MOVE': 'MEMMOVE', + 'L_SHR': 'LSHR', + 'A_SHR': 'ASHR', + 'BIT_CAST': 'BITCAST', + 'U_DIV': 'UDIV', + 'S_DIV': 'SDIV', + 'U_REM': 'UREM', + 'S_REM': 'SREM', + 'BIN_OP': 'BINOP', +} + +intrinsics = [ + ["VGATHERPS", "x86_avx2_gather_d_ps_256", ["src", "pBase", "indices", "mask", "scale"]], + ["VGATHERDD", "x86_avx2_gather_d_d_256", ["src", "pBase", "indices", "mask", "scale"]], + ["VSQRTPS", "x86_avx_sqrt_ps_256", ["a"]], + ["VRSQRTPS", "x86_avx_rsqrt_ps_256", ["a"]], + ["VRCPPS", "x86_avx_rcp_ps_256", ["a"]], + ["VMINPS", "x86_avx_min_ps_256", ["a", "b"]], + ["VMAXPS", "x86_avx_max_ps_256", ["a", "b"]], + ["VPMINSD", "x86_avx2_pmins_d", ["a", "b"]], + ["VPMAXSD", "x86_avx2_pmaxs_d", ["a", "b"]], + ["VROUND", "x86_avx_round_ps_256", ["a", "rounding"]], + ["VCMPPS", "x86_avx_cmp_ps_256", ["a", "b", "cmpop"]], + ["VBLENDVPS", "x86_avx_blendv_ps_256", ["a", "b", "mask"]], + ["BEXTR_32", "x86_bmi_bextr_32", ["src", "control"]], + ["VMASKLOADD", "x86_avx2_maskload_d_256", ["src", "mask"]], + ["VMASKMOVPS", "x86_avx_maskload_ps_256", ["src", "mask"]], + ["VPSHUFB", "x86_avx2_pshuf_b", ["a", "b"]], + ["VPMOVSXBD", "x86_avx2_pmovsxbd", ["a"]], # sign extend packed 8bit components + ["VPMOVSXWD", "x86_avx2_pmovsxwd", ["a"]], # sign extend packed 16bit components + ["VPERMD", "x86_avx2_permd", ["idx", "a"]], + ["VCVTPH2PS", "x86_vcvtph2ps_256", ["a"]], + ["VCVTPS2PH", "x86_vcvtps2ph_256", ["a", "round"]], + ["VHSUBPS", "x86_avx_hsub_ps_256", ["a", "b"]], + ["VPTESTC", "x86_avx_ptestc_256", ["a", "b"]], + ["VPTESTZ", "x86_avx_ptestz_256", ["a", "b"]], + ["VFMADDPS", "x86_fma_vfmadd_ps_256", ["a", "b", "c"]], + ["VCVTTPS2DQ", "x86_avx_cvtt_ps2dq_256", ["a"]], + ["VMOVMSKPS", "x86_avx_movmsk_ps_256", ["a"]], + ["INTERRUPT", "x86_int", ["a"]], + ] + +def convert_uppercamel(name): + s1 = re.sub('(.)([A-Z][a-z]+)', r'\1_\2', name) + return re.sub('([a-z0-9])([A-Z])', r'\1_\2', s1).upper() + +""" + Given an input file (e.g. IRBuilder.h) generates function dictionary. +""" +def parse_ir_builder(input_file): + + functions = [] + + lines = input_file.readlines() + + idx = 0 + while idx < len(lines) - 1: + line = lines[idx].rstrip() + idx += 1 + + #match = re.search(r"\*Create", line) + match = re.search(r"[\*\s]Create(\w*)\(", line) + if match is not None: + #print("Line: %s" % match.group(1)) + + if re.search(r"^\s*Create", line) is not None: + func_sig = lines[idx-2].rstrip() + line + else: + func_sig = line + + end_of_args = False + while not end_of_args: + end_paren = re.search(r"\)", line) + if end_paren is not None: + end_of_args = True + else: + line = lines[idx].rstrip() + func_sig += line + idx += 1 + + delfunc = re.search(r"LLVM_DELETED_FUNCTION|= delete;", func_sig) + + if not delfunc: + func = re.search(r"(.*?)\*[\n\s]*(Create\w*)\((.*?)\)", func_sig) + if func is not None: + + return_type = func.group(1).lstrip() + '*' + func_name = func.group(2) + arguments = func.group(3) + + func_args = '' + func_args_nodefs = '' + + num_args = arguments.count(',') + + arg_names = [] + num_args = 0 + args = arguments.split(',') + for arg in args: + arg = arg.lstrip() + if arg: + if num_args > 0: + func_args += ', ' + func_args_nodefs += ', ' + func_args += arg + func_args_nodefs += arg.split(' =')[0] + + split_args = arg.split('=') + arg_name = split_args[0].rsplit(None, 1)[-1] + + #print("Before ArgName = %s" % arg_name) + + reg_arg = re.search(r"[\&\*]*(\w*)", arg_name) + if reg_arg: + #print("Arg Name = %s" % reg_arg.group(1)) + arg_names += [reg_arg.group(1)] + + num_args += 1 + + ignore = False + + # The following functions need to be ignored. + if func_name == 'CreateInsertNUWNSWBinOp': + ignore = True + + if func_name == 'CreateMaskedIntrinsic': + ignore = True + + # Convert CamelCase to CAMEL_CASE + func_mod = re.search(r"Create(\w*)", func_name) + if func_mod: + func_mod = func_mod.group(1) + func_mod = convert_uppercamel(func_mod) + if func_mod[0:2] == 'F_' or func_mod[0:2] == 'I_': + func_mod = func_mod[0] + func_mod[2:] + + # Substitute alias based on CAMEL_CASE name. + func_alias = inst_aliases.get(func_mod) + if not func_alias: + func_alias = func_mod + + if func_name == 'CreateCall' or func_name == 'CreateGEP': + arglist = re.search(r'ArrayRef', func_args) + if arglist: + func_alias = func_alias + 'A' + + if not ignore: + functions.append({ + "name": func_name, + "alias": func_alias, + "return": return_type, + "args": func_args, + "args_nodefs": func_args_nodefs, + "arg_names": arg_names + }) + + return functions + +""" + Auto-generates macros for LLVM IR +""" +def generate_gen_h(functions, output_file): + output_lines = gen_file_header(os.path.basename(output_file.name)) + + output_lines += [ + '#pragma once', + '', + '//////////////////////////////////////////////////////////////////////////', + '/// Auto-generated Builder IR declarations', + '//////////////////////////////////////////////////////////////////////////', + ] + + for func in functions: + name = func['name'] + if func['alias']: + name = func['alias'] + output_lines += [ + '%s%s(%s);' % (func['return'], name, func['args']) + ] + + output_file.write('\n'.join(output_lines) + '\n') + +""" + Auto-generates macros for LLVM IR +""" +def generate_gen_cpp(functions, output_file): + output_lines = gen_file_header(os.path.basename(output_file.name)) + + output_lines += [ + '#include \"builder.h\"', + '' + ] + + for func in functions: + name = func['name'] + if func['alias']: + name = func['alias'] + + args = func['arg_names'] + func_args = '' + first_arg = True + for arg in args: + if not first_arg: + func_args += ', ' + func_args += arg + first_arg = False + + output_lines += [ + '//////////////////////////////////////////////////////////////////////////', + '%sBuilder::%s(%s)' % (func['return'], name, func['args_nodefs']), + '{', + ' return IRB()->%s(%s);' % (func['name'], func_args), + '}', + '', + ] + + output_file.write('\n'.join(output_lines) + '\n') + +""" + Auto-generates macros for LLVM IR +""" +def generate_x86_h(output_file): + output_lines = gen_file_header(os.path.basename(output_file.name)) + + output_lines += [ + '#pragma once', + '', + '//////////////////////////////////////////////////////////////////////////', + '/// Auto-generated x86 intrinsics', + '//////////////////////////////////////////////////////////////////////////', + ] + + for inst in intrinsics: + #print("Inst: %s, x86: %s numArgs: %d" % (inst[0], inst[1], len(inst[2]))) + + args = '' + first = True + for arg in inst[2]: + if not first: + args += ', ' + args += ("Value* %s" % arg) + first = False + + output_lines += [ + 'Value *%s(%s);' % (inst[0], args) + ] + + output_file.write('\n'.join(output_lines) + '\n') + +""" + Auto-generates macros for LLVM IR +""" +def generate_x86_cpp(output_file): + output_lines = gen_file_header(os.path.basename(output_file.name)) + + output_lines += [ + '#include \"builder.h\"', + '' + ] + + for inst in intrinsics: + #print("Inst: %s, x86: %s numArgs: %d" % (inst[0], inst[1], len(inst[2]))) + + args = '' + pass_args = '' + first = True + for arg in inst[2]: + if not first: + args += ', ' + pass_args += ', ' + args += ("Value* %s" % arg) + pass_args += arg + first = False + + output_lines += [ + '//////////////////////////////////////////////////////////////////////////', + 'Value *Builder::%s(%s)' % (inst[0], args), + '{', + ' Function *func = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::%s);' % inst[1], + ' return CALL(func, std::initializer_list{%s});' % pass_args, + '}', + '', + ] + + output_file.write('\n'.join(output_lines) + '\n') + +""" + Function which is invoked when this script is started from a command line. + Will present and consume a set of arguments which will tell this script how + to behave +""" +def main(): + + # Parse args... + parser = argparse.ArgumentParser() + parser.add_argument("--input", "-i", type=argparse.FileType('r'), help="Path to IRBuilder.h", required=False) + parser.add_argument("--output", "-o", type=argparse.FileType('w'), help="Path to output file", required=True) + parser.add_argument("--gen_h", "-gen_h", help="Generate builder_gen.h", action="store_true", default=False) + parser.add_argument("--gen_cpp", "-gen_cpp", help="Generate builder_gen.cpp", action="store_true", default=False) + parser.add_argument("--gen_x86_h", "-gen_x86_h", help="Generate x86 intrinsics. No input is needed.", action="store_true", default=False) + parser.add_argument("--gen_x86_cpp", "-gen_x86_cpp", help="Generate x86 intrinsics. No input is needed.", action="store_true", default=False) + args = parser.parse_args() + + if args.input: + functions = parse_ir_builder(args.input) + + if args.gen_h: + generate_gen_h(functions, args.output) + + if args.gen_cpp: + generate_gen_cpp(functions, args.output) + else: + if args.gen_x86_h: + generate_x86_h(args.output) + + if args.gen_x86_cpp: + generate_x86_cpp(args.output) + + if args.gen_h: + print("Need to specify --input for --gen_h!") + + if args.gen_cpp: + print("Need to specify --input for --gen_cpp!") + +if __name__ == '__main__': + main() +# END OF FILE diff --git a/src/gallium/drivers/swr/rasterizer/jitter/scripts/gen_llvm_types.py b/src/gallium/drivers/swr/rasterizer/jitter/scripts/gen_llvm_types.py new file mode 100644 index 00000000000..7bba435467b --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/scripts/gen_llvm_types.py @@ -0,0 +1,341 @@ +# Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +# +# Permission is hereby granted, free of charge, to any person obtaining a +# copy of this software and associated documentation files (the "Software"), +# to deal in the Software without restriction, including without limitation +# the rights to use, copy, modify, merge, publish, distribute, sublicense, +# and/or sell copies of the Software, and to permit persons to whom the +# Software is furnished to do so, subject to the following conditions: +# +# The above copyright notice and this permission notice (including the next +# paragraph) shall be included in all copies or substantial portions of the +# Software. +# +# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +# IN THE SOFTWARE. + +#!deps/python32/python.exe + +import os, sys, re +import argparse +import json as JSON +import operator + +header = r""" +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file %s +* +* @brief auto-generated file +* +* DO NOT EDIT +* +******************************************************************************/ + +#pragma once + +""" + +""" +""" +def gen_file_header(filename): + global header + headerStr = header % filename + return headerStr.splitlines() + +""" +""" +def gen_llvm_type(type, name, postfix_name, is_pointer, is_pointer_pointer, is_array, is_array_array, array_count, array_count1, is_llvm_struct, is_llvm_enum, is_llvm_pfn, output_file): + + llvm_type = '' + + if is_llvm_struct: + if is_pointer or is_pointer_pointer: + llvm_type = 'Type::getInt32Ty(ctx)' + else: + llvm_type = 'ArrayType::get(Type::getInt8Ty(ctx), sizeof(%s))' % type + elif is_llvm_enum: + llvm_type = 'Type::getInt32Ty(ctx)' + elif is_llvm_pfn: + llvm_type = 'PointerType::get(Type::getInt8Ty(ctx), 0)' + else: + if type == "BYTE" or type == "char" or type == "uint8_t" or type == "int8_t" or type == 'bool': + llvm_type = 'Type::getInt8Ty(ctx)' + elif type == 'UINT64' or type == 'INT64' or type == 'uint64_t' or type == 'int64_t': + llvm_type = 'Type::getInt64Ty(ctx)' + elif type == 'UINT16' or type == 'int16_t' or type == 'uint16_t': + llvm_type = 'Type::getInt16Ty(ctx)' + elif type == 'UINT' or type == 'INT' or type == 'int' or type == 'BOOL' or type == 'uint32_t' or type == 'int32_t': + llvm_type = 'Type::getInt32Ty(ctx)' + elif type == 'float' or type == 'FLOAT': + llvm_type = 'Type::getFloatTy(ctx)' + elif type == 'double' or type == 'DOUBLE': + llvm_type = 'Type::getDoubleTy(ctx)' + elif type == 'void' or type == 'VOID': + llvm_type = 'Type::getInt32Ty(ctx)' + elif type == 'HANDLE': + llvm_type = 'PointerType::get(Type::getInt32Ty(ctx), 0)' + elif type == 'simdscalar': + llvm_type = 'VectorType::get(Type::getFloatTy(ctx), pJitMgr->mVWidth)' + elif type == 'simdscalari': + llvm_type = 'VectorType::get(Type::getInt32Ty(ctx), pJitMgr->mVWidth)' + elif type == 'simdvector': + llvm_type = 'ArrayType::get(VectorType::get(Type::getFloatTy(ctx), pJitMgr->mVWidth), 4)' + else: + llvm_type = 'Gen_%s%s(pJitMgr)' % (type, postfix_name) + + if is_pointer: + llvm_type = 'PointerType::get(%s, 0)' % llvm_type + + if is_pointer_pointer: + llvm_type = 'PointerType::get(%s, 0)' % llvm_type + + if is_array_array: + llvm_type = 'ArrayType::get(ArrayType::get(%s, %s), %s)' % (llvm_type, array_count1, array_count) + elif is_array: + llvm_type = 'ArrayType::get(%s, %s)' % (llvm_type, array_count) + + return [' members.push_back( %s ); // %s' % (llvm_type, name)] + +""" +""" +def gen_llvm_types(input_file, output_file): + + output_lines = gen_file_header(os.path.basename(output_file.name)) + + lines = input_file.readlines() + + postfix_name = "" + + for idx in range(len(lines)): + line = lines[idx].rstrip() + + match = re.match(r"(\s*)struct(\s*)(\w+)", line) + if match: + llvm_args = [] + + # Detect start of structure + is_fwd_decl = re.search(r";", line) + + if not is_fwd_decl: + + # Extract the command name + struct_name = match.group(3).strip() + + output_lines += [ + '//////////////////////////////////////////////////////////////////////////', + '/// Generate LLVM type information for %s' % struct_name, + 'INLINE static StructType *Gen_%s%s(JitManager* pJitMgr)' % (struct_name, postfix_name), + '{', + ' LLVMContext& ctx = pJitMgr->mContext;', + ' std::vector members;', + '', + ] + + end_of_struct = False + + while not end_of_struct and idx < len(lines)-1: + idx += 1 + line = lines[idx].rstrip() + + is_llvm_typedef = re.search(r"@llvm_typedef", line) + if is_llvm_typedef is not None: + is_llvm_typedef = True + else: + is_llvm_typedef = False + + ########################################### + # Is field a llvm struct? Tells script to treat type as array of bytes that is size of structure. + is_llvm_struct = re.search(r"@llvm_struct", line) + + if is_llvm_struct is not None: + is_llvm_struct = True + else: + is_llvm_struct = False + + ########################################### + # Is field a llvm enum? Tells script to treat type as an enum and replaced with uint32 type. + is_llvm_enum = re.search(r"@llvm_enum", line) + + if is_llvm_enum is not None: + is_llvm_enum = True + else: + is_llvm_enum = False + + ########################################### + # Is field a llvm function pointer? Tells script to treat type as an enum and replaced with uint32 type. + is_llvm_pfn = re.search(r"@llvm_pfn", line) + + if is_llvm_pfn is not None: + is_llvm_pfn = True + else: + is_llvm_pfn = False + + ########################################### + # Is field const? + is_const = re.search(r"\s+const\s+", line) + + if is_const is not None: + is_const = True + else: + is_const = False + + ########################################### + # Is field a pointer? + is_pointer_pointer = re.search("\*\*", line) + + if is_pointer_pointer is not None: + is_pointer_pointer = True + else: + is_pointer_pointer = False + + ########################################### + # Is field a pointer? + is_pointer = re.search("\*", line) + + if is_pointer is not None: + is_pointer = True + else: + is_pointer = False + + ########################################### + # Is field an array of arrays? + # TODO: Can add this to a list. + is_array_array = re.search("\[(\w*)\]\[(\w*)\]", line) + array_count = '0' + array_count1 = '0' + + if is_array_array is not None: + array_count = is_array_array.group(1) + array_count1 = is_array_array.group(2) + is_array_array = True + else: + is_array_array = False + + ########################################### + # Is field an array? + is_array = re.search("\[(\w*)\]", line) + + if is_array is not None: + array_count = is_array.group(1) + is_array = True + else: + is_array = False + + is_scoped = re.search("::", line) + + if is_scoped is not None: + is_scoped = True + else: + is_scoped = False + + type = None + name = None + if is_const and is_pointer: + + if is_scoped: + field_match = re.match(r"(\s*)(\w+\<*\w*\>*)(\s+)(\w+::)(\w+)(\s*\**\s*)(\w+)", line) + + type = "%s%s" % (field_match.group(4), field_match.group(5)) + name = field_match.group(7) + else: + field_match = re.match(r"(\s*)(\w+\<*\w*\>*)(\s+)(\w+)(\s*\**\s*)(\w+)", line) + + type = field_match.group(4) + name = field_match.group(6) + + elif is_pointer: + field_match = re.match(r"(\s*)(\s+)(\w+\<*\w*\>*)(\s*\**\s*)(\w+)", line) + + if field_match: + type = field_match.group(3) + name = field_match.group(5) + elif is_const: + field_match = re.match(r"(\s*)(\w+\<*\w*\>*)(\s+)(\w+)(\s*)(\w+)", line) + + if field_match: + type = field_match.group(4) + name = field_match.group(6) + else: + if is_scoped: + field_match = re.match(r"\s*(\w+\<*\w*\>*)\s*::\s*(\w+\<*\w*\>*)\s+(\w+)", line) + + if field_match: + type = field_match.group(1) + '::' + field_match.group(2) + name = field_match.group(3) + else: + field_match = re.match(r"(\s*)(\w+\<*\w*\>*)(\s+)(\w+)", line) + + if field_match: + type = field_match.group(2) + name = field_match.group(4) + + if is_llvm_typedef is False: + if type is not None: + output_lines += gen_llvm_type(type, name, postfix_name, is_pointer, is_pointer_pointer, is_array, is_array_array, array_count, array_count1, is_llvm_struct, is_llvm_enum, is_llvm_pfn, output_file) + llvm_args.append(name) + + # Detect end of structure + end_of_struct = re.match(r"(\s*)};", line) + + if (end_of_struct): + output_lines += [ + '', + ' return StructType::get(ctx, members, false);', + '}', + '', + ] + + for i in range(len(llvm_args)): + output_lines.append('static const uint32_t %s%s_%s = %s;' % (struct_name, postfix_name, llvm_args[i], i)) + + output_lines.append('') + + output_file.write('\n'.join(output_lines) + '\n') + +""" + Function which is invoked when this script is started from a command line. + Will present and consume a set of arguments which will tell this script how + to behave +""" +def main(): + + # Parse args... + parser = argparse.ArgumentParser() + parser.add_argument("--input", "-i", type=argparse.FileType('r'), + help="Path to input file containing structs", required=True) + parser.add_argument("--output", "-o", type=argparse.FileType('w'), + help="Path to output file", required=True) + parser.add_argument("--scalar", "-scalar", help="Generates scalar files with all enums", action="store_true", default=False) + args = parser.parse_args() + + gen_llvm_types(args.input, args.output) + +if __name__ == '__main__': + main() +# END OF FILE diff --git a/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp b/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp new file mode 100644 index 00000000000..6c5f22bc47c --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.cpp @@ -0,0 +1,357 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file streamout_jit.cpp +* +* @brief Implementation of the streamout jitter +* +* Notes: +* +******************************************************************************/ +#include "jit_api.h" +#include "streamout_jit.h" +#include "builder.h" +#include "state_llvm.h" +#include "common/containers.hpp" +#include "llvm/IR/DataLayout.h" + +#include +#include + +////////////////////////////////////////////////////////////////////////// +/// Interface to Jitting a fetch shader +////////////////////////////////////////////////////////////////////////// +struct StreamOutJit : public Builder +{ + StreamOutJit(JitManager* pJitMgr) : Builder(pJitMgr){}; + + // returns pointer to SWR_STREAMOUT_BUFFER + Value* getSOBuffer(Value* pSoCtx, uint32_t buffer) + { + return LOAD(pSoCtx, { 0, SWR_STREAMOUT_CONTEXT_pBuffer, buffer }); + } + + + ////////////////////////////////////////////////////////////////////////// + // @brief checks if streamout buffer is oob + // @return true/false + Value* oob(const STREAMOUT_COMPILE_STATE& state, Value* pSoCtx, uint32_t buffer) + { + Value* returnMask = C(false); + + Value* pBuf = getSOBuffer(pSoCtx, buffer); + + // load enable + // @todo bool data types should generate llvm type + Value* enabled = TRUNC(LOAD(pBuf, { 0, SWR_STREAMOUT_BUFFER_enable }), IRB()->getInt1Ty()); + + // load buffer size + Value* bufferSize = LOAD(pBuf, { 0, SWR_STREAMOUT_BUFFER_bufferSize }); + + // load current streamOffset + Value* streamOffset = LOAD(pBuf, { 0, SWR_STREAMOUT_BUFFER_streamOffset }); + + // load buffer pitch + Value* pitch = LOAD(pBuf, { 0, SWR_STREAMOUT_BUFFER_pitch }); + + // buffer is considered oob if in use in a decl but not enabled + returnMask = OR(returnMask, NOT(enabled)); + + // buffer is oob if cannot fit a prims worth of verts + Value* newOffset = ADD(streamOffset, MUL(pitch, C(state.numVertsPerPrim))); + returnMask = OR(returnMask, ICMP_SGT(newOffset, bufferSize)); + + return returnMask; + } + + + ////////////////////////////////////////////////////////////////////////// + // @brief converts scalar bitmask to <4 x i32> suitable for shuffle vector, + // packing the active mask bits + // ex. bitmask 0011 -> (0, 1, 0, 0) + // bitmask 1000 -> (3, 0, 0, 0) + // bitmask 1100 -> (2, 3, 0, 0) + Value* PackMask(uint32_t bitmask) + { + std::vector indices(4, C(0)); + DWORD index; + uint32_t elem = 0; + while (_BitScanForward(&index, bitmask)) + { + indices[elem++] = C((int)index); + bitmask &= ~(1 << index); + } + + return ConstantVector::get(indices); + } + + ////////////////////////////////////////////////////////////////////////// + // @brief convert scalar bitmask to <4xfloat> bitmask + Value* ToMask(uint32_t bitmask) + { + std::vector indices; + for (uint32_t i = 0; i < 4; ++i) + { + if (bitmask & (1 << i)) + { + indices.push_back(C(-1.0f)); + } + else + { + indices.push_back(C(0.0f)); + } + } + return ConstantVector::get(indices); + } + + ////////////////////////////////////////////////////////////////////////// + // @brief processes a single decl from the streamout stream. Reads 4 components from the input + // stream and writes N components to the output buffer given the componentMask or if + // a hole, just increments the buffer pointer + // @param pStream - pointer to current attribute + // @param pOutBuffers - pointers to the current location of each output buffer + // @param decl - input decl + void buildDecl(Value* pStream, Value* pOutBuffers[4], const STREAMOUT_DECL& decl) + { + // @todo add this to x86 macros + Function* maskStore = Intrinsic::getDeclaration(JM()->mpCurrentModule, Intrinsic::x86_avx_maskstore_ps); + + uint32_t numComponents = _mm_popcnt_u32(decl.componentMask); + uint32_t packedMask = (1 << numComponents) - 1; + if (!decl.hole) + { + // increment stream pointer to correct slot + Value* pAttrib = GEP(pStream, C(4 * decl.attribSlot)); + + // load 4 components from stream + Type* simd4Ty = VectorType::get(IRB()->getFloatTy(), 4); + Type* simd4PtrTy = PointerType::get(simd4Ty, 0); + pAttrib = BITCAST(pAttrib, simd4PtrTy); + Value *vattrib = LOAD(pAttrib); + + // shuffle/pack enabled components + Value* vpackedAttrib = VSHUFFLE(vattrib, vattrib, PackMask(decl.componentMask)); + + // store to output buffer + // cast SO buffer to i8*, needed by maskstore + Value* pOut = BITCAST(pOutBuffers[decl.bufferIndex], PointerType::get(mInt8Ty, 0)); + + // cast input to <4xfloat> + Value* src = BITCAST(vpackedAttrib, simd4Ty); + CALL(maskStore, {pOut, ToMask(packedMask), src}); + } + + // increment SO buffer + pOutBuffers[decl.bufferIndex] = GEP(pOutBuffers[decl.bufferIndex], C(numComponents)); + } + + ////////////////////////////////////////////////////////////////////////// + // @brief builds a single vertex worth of data for the given stream + // @param streamState - state for this stream + // @param pCurVertex - pointer to src stream vertex data + // @param pOutBuffer - pointers to up to 4 SO buffers + void buildVertex(const STREAMOUT_STREAM& streamState, Value* pCurVertex, Value* pOutBuffer[4]) + { + for (uint32_t d = 0; d < streamState.numDecls; ++d) + { + const STREAMOUT_DECL& decl = streamState.decl[d]; + buildDecl(pCurVertex, pOutBuffer, decl); + } + } + + void buildStream(const STREAMOUT_COMPILE_STATE& state, const STREAMOUT_STREAM& streamState, Value* pSoCtx, BasicBlock* returnBB, Function* soFunc) + { + // get list of active SO buffers + std::unordered_set activeSOBuffers; + for (uint32_t d = 0; d < streamState.numDecls; ++d) + { + const STREAMOUT_DECL& decl = streamState.decl[d]; + activeSOBuffers.insert(decl.bufferIndex); + } + + // always increment numPrimStorageNeeded + Value *numPrimStorageNeeded = LOAD(pSoCtx, { 0, SWR_STREAMOUT_CONTEXT_numPrimStorageNeeded }); + numPrimStorageNeeded = ADD(numPrimStorageNeeded, C(1)); + STORE(numPrimStorageNeeded, pSoCtx, { 0, SWR_STREAMOUT_CONTEXT_numPrimStorageNeeded }); + + // check OOB on active SO buffers. If any buffer is out of bound, don't write + // the primitive to any buffer + Value* oobMask = C(false); + for (uint32_t buffer : activeSOBuffers) + { + oobMask = OR(oobMask, oob(state, pSoCtx, buffer)); + } + + BasicBlock* validBB = BasicBlock::Create(JM()->mContext, "valid", soFunc); + + // early out if OOB + COND_BR(oobMask, returnBB, validBB); + + IRB()->SetInsertPoint(validBB); + + Value* numPrimsWritten = LOAD(pSoCtx, { 0, SWR_STREAMOUT_CONTEXT_numPrimsWritten }); + numPrimsWritten = ADD(numPrimsWritten, C(1)); + STORE(numPrimsWritten, pSoCtx, { 0, SWR_STREAMOUT_CONTEXT_numPrimsWritten }); + + // compute start pointer for each output buffer + Value* pOutBuffer[4]; + Value* pOutBufferStartVertex[4]; + Value* outBufferPitch[4]; + for (uint32_t b: activeSOBuffers) + { + Value* pBuf = getSOBuffer(pSoCtx, b); + Value* pData = LOAD(pBuf, { 0, SWR_STREAMOUT_BUFFER_pBuffer }); + Value* streamOffset = LOAD(pBuf, { 0, SWR_STREAMOUT_BUFFER_streamOffset }); + pOutBuffer[b] = GEP(pData, streamOffset); + pOutBufferStartVertex[b] = pOutBuffer[b]; + + outBufferPitch[b] = LOAD(pBuf, { 0, SWR_STREAMOUT_BUFFER_pitch }); + } + + // loop over the vertices of the prim + Value* pStreamData = LOAD(pSoCtx, { 0, SWR_STREAMOUT_CONTEXT_pPrimData }); + for (uint32_t v = 0; v < state.numVertsPerPrim; ++v) + { + buildVertex(streamState, pStreamData, pOutBuffer); + + // increment stream and output buffer pointers + // stream verts are always 32*4 dwords apart + pStreamData = GEP(pStreamData, C(KNOB_NUM_ATTRIBUTES * 4)); + + // output buffers offset using pitch in buffer state + for (uint32_t b : activeSOBuffers) + { + pOutBufferStartVertex[b] = GEP(pOutBufferStartVertex[b], outBufferPitch[b]); + pOutBuffer[b] = pOutBufferStartVertex[b]; + } + } + + // update each active buffer's streamOffset + for (uint32_t b : activeSOBuffers) + { + Value* pBuf = getSOBuffer(pSoCtx, b); + Value* streamOffset = LOAD(pBuf, { 0, SWR_STREAMOUT_BUFFER_streamOffset }); + streamOffset = ADD(streamOffset, MUL(C(state.numVertsPerPrim), outBufferPitch[b])); + STORE(streamOffset, pBuf, { 0, SWR_STREAMOUT_BUFFER_streamOffset }); + } + } + + Function* Create(const STREAMOUT_COMPILE_STATE& state) + { + static std::size_t soNum = 0; + + std::stringstream fnName("SOShader", std::ios_base::in | std::ios_base::out | std::ios_base::ate); + fnName << soNum++; + + // SO function signature + // typedef void(__cdecl *PFN_SO_FUNC)(SWR_STREAMOUT_CONTEXT*) + + std::vector args{ + PointerType::get(Gen_SWR_STREAMOUT_CONTEXT(JM()), 0), // SWR_STREAMOUT_CONTEXT* + }; + + FunctionType* fTy = FunctionType::get(IRB()->getVoidTy(), args, false); + Function* soFunc = Function::Create(fTy, GlobalValue::ExternalLinkage, fnName.str(), JM()->mpCurrentModule); + + // create return basic block + BasicBlock* entry = BasicBlock::Create(JM()->mContext, "entry", soFunc); + BasicBlock* returnBB = BasicBlock::Create(JM()->mContext, "return", soFunc); + + IRB()->SetInsertPoint(entry); + + // arguments + auto argitr = soFunc->getArgumentList().begin(); + Value* pSoCtx = &*argitr++; + pSoCtx->setName("pSoCtx"); + + const STREAMOUT_STREAM& streamState = state.stream; + buildStream(state, streamState, pSoCtx, returnBB, soFunc); + + BR(returnBB); + + IRB()->SetInsertPoint(returnBB); + RET_VOID(); + + JitManager::DumpToFile(soFunc, "SoFunc"); + + FunctionPassManager passes(JM()->mpCurrentModule); + passes.add(createBreakCriticalEdgesPass()); + passes.add(createCFGSimplificationPass()); + passes.add(createEarlyCSEPass()); + passes.add(createPromoteMemoryToRegisterPass()); + passes.add(createCFGSimplificationPass()); + passes.add(createEarlyCSEPass()); + passes.add(createInstructionCombiningPass()); + passes.add(createInstructionSimplifierPass()); + passes.add(createConstantPropagationPass()); + passes.add(createSCCPPass()); + passes.add(createAggressiveDCEPass()); + + passes.run(*soFunc); + + JitManager::DumpToFile(soFunc, "SoFunc_optimized"); + + return soFunc; + } +}; + +////////////////////////////////////////////////////////////////////////// +/// @brief JITs from streamout shader IR +/// @param hJitMgr - JitManager handle +/// @param func - LLVM function IR +/// @return PFN_SO_FUNC - pointer to SOS function +PFN_SO_FUNC JitStreamoutFunc(HANDLE hJitMgr, const HANDLE hFunc) +{ + const llvm::Function *func = (const llvm::Function*)hFunc; + JitManager* pJitMgr = reinterpret_cast(hJitMgr); + PFN_SO_FUNC pfnStreamOut; + pfnStreamOut = (PFN_SO_FUNC)(pJitMgr->mpExec->getFunctionAddress(func->getName().str())); + // MCJIT finalizes modules the first time you JIT code from them. After finalized, you cannot add new IR to the module + pJitMgr->mIsModuleFinalized = true; + + return pfnStreamOut; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief JIT compiles streamout shader +/// @param hJitMgr - JitManager handle +/// @param state - SO state to build function from +extern "C" PFN_SO_FUNC JITCALL JitCompileStreamout(HANDLE hJitMgr, const STREAMOUT_COMPILE_STATE& state) +{ + JitManager* pJitMgr = reinterpret_cast(hJitMgr); + + STREAMOUT_COMPILE_STATE soState = state; + if (soState.offsetAttribs) + { + for (uint32_t i = 0; i < soState.stream.numDecls; ++i) + { + soState.stream.decl[i].attribSlot -= soState.offsetAttribs; + } + } + + pJitMgr->SetupNewModule(); + + StreamOutJit theJit(pJitMgr); + HANDLE hFunc = theJit.Create(soState); + + return JitStreamoutFunc(hJitMgr, hFunc); +} diff --git a/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.h b/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.h new file mode 100644 index 00000000000..097f8ab44d9 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/jitter/streamout_jit.h @@ -0,0 +1,94 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file streamout_jit.h +* +* @brief Definition of the streamout jitter +* +* Notes: +* +******************************************************************************/ +#pragma once + +#include "common/formats.h" +#include "core/state.h" + +////////////////////////////////////////////////////////////////////////// +/// STREAMOUT_DECL - Stream decl +////////////////////////////////////////////////////////////////////////// +struct STREAMOUT_DECL +{ + // Buffer that stream maps to. + DWORD bufferIndex; + + // attribute to stream + uint32_t attribSlot; + + // attribute component mask + uint32_t componentMask; + + // indicates this decl is a hole + bool hole; +}; + +////////////////////////////////////////////////////////////////////////// +/// STREAMOUT_STREAM - Stream decls +////////////////////////////////////////////////////////////////////////// +struct STREAMOUT_STREAM +{ + // numnber of decls for this stream + uint32_t numDecls; + + // array of numDecls decls + STREAMOUT_DECL decl[128]; +}; + +////////////////////////////////////////////////////////////////////////// +/// State required for streamout jit +////////////////////////////////////////////////////////////////////////// +struct STREAMOUT_COMPILE_STATE +{ + // number of verts per primitive + uint32_t numVertsPerPrim; + uint32_t offsetAttribs; ///< attrib offset to subtract from all STREAMOUT_DECL::attribSlot values. + + uint64_t streamMask; + + // stream decls + STREAMOUT_STREAM stream; + + bool operator==(const STREAMOUT_COMPILE_STATE &other) const + { + if (numVertsPerPrim != other.numVertsPerPrim) return false; + if (stream.numDecls != other.stream.numDecls) return false; + + for (uint32_t i = 0; i < stream.numDecls; ++i) + { + if (stream.decl[i].bufferIndex != other.stream.decl[i].bufferIndex) return false; + if (stream.decl[i].attribSlot != other.stream.decl[i].attribSlot) return false; + if (stream.decl[i].componentMask != other.stream.decl[i].componentMask) return false; + if (stream.decl[i].hole != other.stream.decl[i].hole) return false; + } + + return true; + } +}; diff --git a/src/gallium/drivers/swr/rasterizer/memory/ClearTile.cpp b/src/gallium/drivers/swr/rasterizer/memory/ClearTile.cpp new file mode 100644 index 00000000000..ad73cd840a7 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/memory/ClearTile.cpp @@ -0,0 +1,287 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file ClearTile.cpp +* +* @brief Functionality for ClearTile. StoreHotTileClear clears a single macro +* tile in the destination. +* +******************************************************************************/ +#include "common/os.h" +#include "core/context.h" +#include "common/formats.h" +#include "memory/TilingFunctions.h" +#include "memory/tilingtraits.h" +#include "memory/Convert.h" + +typedef void(*PFN_STORE_TILES_CLEAR)(const FLOAT*, SWR_SURFACE_STATE*, UINT, UINT); + +////////////////////////////////////////////////////////////////////////// +/// Clear Raster Tile Function Tables. +////////////////////////////////////////////////////////////////////////// +static PFN_STORE_TILES_CLEAR sStoreTilesClearColorTable[NUM_SWR_FORMATS]; + +static PFN_STORE_TILES_CLEAR sStoreTilesClearDepthTable[NUM_SWR_FORMATS]; + +////////////////////////////////////////////////////////////////////////// +/// StoreRasterTileClear +////////////////////////////////////////////////////////////////////////// +template +struct StoreRasterTileClear +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores an 8x8 raster tile to the destination surface. + /// @param pColor - Pointer to clear color. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to raster tile. + INLINE static void StoreClear( + const BYTE* dstFormattedColor, + UINT dstBytesPerPixel, + SWR_SURFACE_STATE* pDstSurface, + UINT x, UINT y) // (x, y) pixel coordinate to start of raster tile. + { + // Compute destination address for raster tile. + BYTE* pDstTile = (BYTE*)pDstSurface->pBaseAddress + + (y * pDstSurface->pitch) + (x * dstBytesPerPixel); + + // start of first row + BYTE* pDst = pDstTile; + UINT dstBytesPerRow = 0; + + // For each raster tile pixel in row 0 (rx, 0) + for (UINT rx = 0; (rx < KNOB_TILE_X_DIM) && ((x + rx) < pDstSurface->width); ++rx) + { + memcpy(pDst, dstFormattedColor, dstBytesPerPixel); + + // Increment pointer to next pixel in row. + pDst += dstBytesPerPixel; + dstBytesPerRow += dstBytesPerPixel; + } + + // start of second row + pDst = pDstTile + pDstSurface->pitch; + + // For each remaining row in the rest of the raster tile + for (UINT ry = 1; (ry < KNOB_TILE_Y_DIM) && ((y + ry) < pDstSurface->height); ++ry) + { + // copy row + memcpy(pDst, pDstTile, dstBytesPerRow); + + // Increment pointer to first pixel in next row. + pDst += pDstSurface->pitch; + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// StoreMacroTileClear - Stores a macro tile clear to its raster tiles. +////////////////////////////////////////////////////////////////////////// +template +struct StoreMacroTileClear +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores a macrotile to the destination surface. + /// @param pColor - Pointer to color to write to pixels. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to macro tile + static void StoreClear( + const FLOAT *pColor, + SWR_SURFACE_STATE* pDstSurface, + UINT x, UINT y) + { + UINT dstBytesPerPixel = (FormatTraits::bpp / 8); + + BYTE dstFormattedColor[16]; // max bpp is 128, so 16 is all we need here for one pixel + + FLOAT srcColor[4]; + + for (UINT comp = 0; comp < FormatTraits::numComps; ++comp) + { + srcColor[comp] = pColor[FormatTraits::swizzle(comp)]; + } + + // using this helper function, but the Tiling Traits is unused inside it so just using a dummy value + ConvertPixelFromFloat(dstFormattedColor, srcColor); + + // Store each raster tile from the hot tile to the destination surface. + // TODO: Put in check for partial coverage on x/y -- SWR_ASSERT if it happens. + // Intent is for this function to only handle full tiles. + for (UINT row = 0; row < KNOB_MACROTILE_Y_DIM; row += KNOB_TILE_Y_DIM) + { + for (UINT col = 0; col < KNOB_MACROTILE_X_DIM; col += KNOB_TILE_X_DIM) + { + StoreRasterTileClear::StoreClear(dstFormattedColor, dstBytesPerPixel, pDstSurface, (x + col), (y + row)); + } + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// @brief Writes clear color to every pixel of a render surface +/// @param hPrivateContext - Handle to private DC +/// @param renderTargetIndex - Index to destination render target +/// @param x, y - Coordinates to raster tile. +/// @param pClearColor - Pointer to clear color +void StoreHotTileClear( + SWR_SURFACE_STATE *pDstSurface, + SWR_RENDERTARGET_ATTACHMENT renderTargetIndex, + UINT x, + UINT y, + const float* pClearColor) +{ + PFN_STORE_TILES_CLEAR pfnStoreTilesClear = NULL; + + SWR_ASSERT(renderTargetIndex != SWR_ATTACHMENT_STENCIL); ///@todo Not supported yet. + + if (renderTargetIndex != SWR_ATTACHMENT_DEPTH) + { + pfnStoreTilesClear = sStoreTilesClearColorTable[pDstSurface->format]; + } + else + { + pfnStoreTilesClear = sStoreTilesClearDepthTable[pDstSurface->format]; + } + + SWR_ASSERT(pfnStoreTilesClear != NULL); + + // Store a macro tile. + /// @todo Once all formats are supported then if check can go away. This is to help us near term to make progress. + if (pfnStoreTilesClear != NULL) + { + pfnStoreTilesClear(pClearColor, pDstSurface, x, y); + } +} + +////////////////////////////////////////////////////////////////////////// +/// INIT_STORE_TILES_TABLE - Helper macro for setting up the tables. +#define INIT_STORE_TILES_CLEAR_COLOR_TABLE() \ + memset(sStoreTilesClearColorTable, 0, sizeof(sStoreTilesClearColorTable)); \ + \ + sStoreTilesClearColorTable[R32G32B32A32_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R32G32B32A32_SINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R32G32B32A32_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R32G32B32X32_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R32G32B32_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R32G32B32_SINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R32G32B32_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16B16A16_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16B16A16_SNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16B16A16_SINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16B16A16_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16B16A16_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R32G32_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R32G32_SINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R32G32_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16B16X16_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16B16X16_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B8G8R8A8_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B8G8R8A8_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R10G10B10A2_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R10G10B10A2_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R10G10B10A2_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8B8A8_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8B8A8_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8B8A8_SNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8B8A8_SINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8B8A8_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16_SNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16_SINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B10G10R10A2_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B10G10R10A2_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R11G11B10_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R32_SINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R32_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R32_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[A32_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B8G8R8X8_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B8G8R8X8_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8B8X8_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8B8X8_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B10G10R10X2_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B5G6R5_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B5G6R5_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B5G5R5A1_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B5G5R5A1_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B4G4R4A4_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B4G4R4A4_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8_SNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8_SINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16_SNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16_SINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[A16_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[A16_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B5G5R5X1_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B5G5R5X1_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8_SNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8_SINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[A8_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[BC1_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[BC2_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[BC3_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[BC4_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[BC5_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[BC1_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[BC2_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[BC3_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8B8_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8B8_SNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[BC4_SNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[BC5_SNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16B16_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16B16_UNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16B16_SNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8B8_UNORM_SRGB] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16B16_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R16G16B16_SINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R10G10B10A2_SNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R10G10B10A2_SINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B10G10R10A2_SNORM] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B10G10R10A2_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[B10G10R10A2_SINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8B8_UINT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearColorTable[R8G8B8_SINT] = StoreMacroTileClear::StoreClear; \ + +////////////////////////////////////////////////////////////////////////// +/// INIT_STORE_TILES_TABLE - Helper macro for setting up the tables. +#define INIT_STORE_TILES_CLEAR_DEPTH_TABLE() \ + memset(sStoreTilesClearDepthTable, 0, sizeof(sStoreTilesClearDepthTable)); \ + \ + sStoreTilesClearDepthTable[R32_FLOAT] = StoreMacroTileClear::StoreClear; \ + sStoreTilesClearDepthTable[R24_UNORM_X8_TYPELESS] = StoreMacroTileClear::StoreClear; \ + +////////////////////////////////////////////////////////////////////////// +/// @brief Sets up tables for ClearTile +void InitSimClearTilesTable() +{ + INIT_STORE_TILES_CLEAR_COLOR_TABLE(); + INIT_STORE_TILES_CLEAR_DEPTH_TABLE(); +} diff --git a/src/gallium/drivers/swr/rasterizer/memory/Convert.h b/src/gallium/drivers/swr/rasterizer/memory/Convert.h new file mode 100644 index 00000000000..0f9e0ad4bd8 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/memory/Convert.h @@ -0,0 +1,698 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file Convert.h +* +* @brief Conversion utility functions +* +******************************************************************************/ +#pragma once + +#if defined(_WIN32) +// disable "potential divide by 0" +#pragma warning(disable: 4723) +#endif + +////////////////////////////////////////////////////////////////////////// +/// @brief Convert an IEEE 754 16-bit float to an 32-bit single precision +/// float +/// @param val - 16-bit float +/// @todo Maybe move this outside of this file into a header? +static float ConvertSmallFloatTo32(UINT val) +{ + UINT result; + if ((val & 0x7fff) == 0) + { + result = ((uint32_t)(val & 0x8000)) << 16; + } + else if ((val & 0x7c00) == 0x7c00) + { + result = ((val & 0x3ff) == 0) ? 0x7f800000 : 0x7fc00000; + result |= ((uint32_t)val & 0x8000) << 16; + } + else + { + uint32_t sign = (val & 0x8000) << 16; + uint32_t mant = (val & 0x3ff) << 13; + uint32_t exp = (val >> 10) & 0x1f; + if ((exp == 0) && (mant != 0)) // Adjust exponent and mantissa for denormals + { + mant <<= 1; + while (mant < (0x400 << 13)) + { + exp--; + mant <<= 1; + } + mant &= (0x3ff << 13); + } + exp = ((exp - 15 + 127) & 0xff) << 23; + result = sign | exp | mant; + } + + return *(float*)&result; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Convert an IEEE 754 32-bit single precision float to an +/// unsigned small float with 5 exponent bits and a variable +/// number of mantissa bits. +/// @param val - 32-bit float +/// @todo Maybe move this outside of this file into a header? +template +static UINT Convert32ToSmallFloat(float val) +{ + uint32_t sign, exp, mant; + uint32_t roundBits; + + // Extract the sign, exponent, and mantissa + UINT uf = *(UINT*)&val; + + sign = (uf & 0x80000000) >> 31; + exp = (uf & 0x7F800000) >> 23; + mant = uf & 0x007FFFFF; + + // 10/11 bit floats are unsigned. Negative values are clamped to 0. + if (sign != 0) + { + exp = mant = 0; + } + // Check for out of range + else if ((exp == 0xFF) && (mant != 0)) // NaN + { + exp = 0x1F; + mant = 1 << numMantissaBits; + } + else if ((exp == 0xFF) && (mant == 0)) // INF + { + exp = 0x1F; + mant = 0; + } + else if (exp > (0x70 + 0x1E)) // Too big to represent + { + exp = 0x1Eu; + mant = (1 << numMantissaBits) - 1; // 0x3F for 6 bit mantissa. + } + else if ((exp <= 0x70) && (exp >= 0x66)) // It's a denorm + { + mant |= 0x00800000; + for (; exp <= 0x70; mant >>= 1, exp++) + ; + exp = 0; + mant = mant >> (23 - numMantissaBits); + } + else if (exp < 0x66) // Too small to represent -> Zero + { + exp = 0; + mant = 0; + } + else + { + // Saves bits that will be shifted off for rounding + roundBits = mant & 0x1FFFu; + // convert exponent and mantissa to 16 bit format + exp = exp - 0x70u; + mant = mant >> (23 - numMantissaBits); + + // Essentially RTZ, but round up if off by only 1 lsb + if (roundBits == 0x1FFFu) + { + mant++; + // check for overflow + if ((mant & (0x3 << numMantissaBits)) != 0) // 0x60 = 0x3 << (num Mantissa Bits) + exp++; + // make sure only the needed bits are used + mant &= (1 << numMantissaBits) - 1; + } + } + + UINT tmpVal = (exp << numMantissaBits) | mant; + return tmpVal; +} + +#if KNOB_ARCH == KNOB_ARCH_AVX +////////////////////////////////////////////////////////////////////////// +/// @brief Convert an IEEE 754 32-bit single precision float to an +/// 16 bit float with 5 exponent bits and a variable +/// number of mantissa bits. +/// @param val - 32-bit float +/// @todo Maybe move this outside of this file into a header? +static uint16_t Convert32To16Float(float val) +{ + uint32_t sign, exp, mant; + uint32_t roundBits; + + // Extract the sign, exponent, and mantissa + uint32_t uf = *(uint32_t*)&val; + sign = (uf & 0x80000000) >> 31; + exp = (uf & 0x7F800000) >> 23; + mant = uf & 0x007FFFFF; + + // Check for out of range + if (std::isnan(val)) + { + exp = 0x1F; + mant = 0x200; + sign = 1; // set the sign bit for NANs + } + else if (std::isinf(val)) + { + exp = 0x1f; + mant = 0x0; + } + else if (exp > (0x70 + 0x1E)) // Too big to represent -> max representable value + { + exp = 0x1E; + mant = 0x3FF; + } + else if ((exp <= 0x70) && (exp >= 0x66)) // It's a denorm + { + mant |= 0x00800000; + for (; exp <= 0x70; mant >>= 1, exp++) + ; + exp = 0; + mant = mant >> 13; + } + else if (exp < 0x66) // Too small to represent -> Zero + { + exp = 0; + mant = 0; + } + else + { + // Saves bits that will be shifted off for rounding + roundBits = mant & 0x1FFFu; + // convert exponent and mantissa to 16 bit format + exp = exp - 0x70; + mant = mant >> 13; + + // Essentially RTZ, but round up if off by only 1 lsb + if (roundBits == 0x1FFFu) + { + mant++; + // check for overflow + if ((mant & 0xC00u) != 0) + exp++; + // make sure only the needed bits are used + mant &= 0x3FF; + } + } + + uint32_t tmpVal = (sign << 15) | (exp << 10) | mant; + return (uint16_t)tmpVal; +} +#endif + +////////////////////////////////////////////////////////////////////////// +/// @brief Retrieve color from hot tile source which is always float. +/// @param pDstPixel - Pointer to destination pixel. +/// @param srcPixel - Pointer to source pixel (pre-swizzled according to dest). +template +static void ConvertPixelFromFloat( + BYTE* pDstPixel, + const float srcPixel[4]) +{ + UINT outColor[4]; // typeless bits + + // Store component + for (UINT comp = 0; comp < FormatTraits::numComps; ++comp) + { + SWR_TYPE type = FormatTraits::GetType(comp); + + float src = srcPixel[comp]; + + switch (type) + { + case SWR_TYPE_UNORM: + { + // Force NaN to 0. IEEE standard, comparisons involving NaN always evaluate to false. + src = (src != src) ? 0.0f : src; + + // Clamp [0, 1] + src = std::max(src, 0.0f); + src = std::min(src, 1.0f); + + // SRGB + if (FormatTraits::isSRGB && comp != 3) + { + src = (src <= 0.0031308f) ? (12.92f * src) : (1.055f * powf(src, (1.0f / 2.4f)) - 0.055f); + } + + // Float scale to integer scale. + UINT scale = (1 << FormatTraits::GetBPC(comp)) - 1; + src = (float)scale * src; + src = roundf(src); + outColor[comp] = (UINT)src; // Drop fractional part. + break; + } + case SWR_TYPE_SNORM: + { + SWR_ASSERT(!FormatTraits::isSRGB); + + // Force NaN to 0. IEEE standard, comparisons involving NaN always evaluate to false. + src = (src != src) ? 0.0f : src; + + // Clamp [-1, 1] + src = std::max(src, -1.0f); + src = std::min(src, 1.0f); + + // Float scale to integer scale. + UINT scale = (1 << (FormatTraits::GetBPC(comp) - 1)) - 1; + src = (float)scale * src; + + // Round + src += (src >= 0) ? 0.5f : -0.5f; + + INT out = (INT)src; + + outColor[comp] = *(UINT*)&out; + + break; + } + case SWR_TYPE_UINT: + { + ///@note The *(UINT*)& is currently necessary as the hot tile appears to always be float. + // However, the number in the hot tile should be unsigned integer. So doing this + // to preserve bits intead of doing a float -> integer conversion. + if (FormatTraits::GetBPC(comp) == 32) + { + outColor[comp] = *(UINT*)&src; + } + else + { + outColor[comp] = *(UINT*)&src; + UINT max = (1 << FormatTraits::GetBPC(comp)) - 1; // 2^numBits - 1 + + outColor[comp] = std::min(max, outColor[comp]); + } + break; + } + case SWR_TYPE_SINT: + { + if (FormatTraits::GetBPC(comp) == 32) + { + outColor[comp] = *(UINT*)&src; + } + else + { + INT out = *(INT*)&src; // Hot tile format is SINT? + INT max = (1 << (FormatTraits::GetBPC(comp) - 1)) - 1; + INT min = -1 - max; + + ///@note The output is unsigned integer (bag of bits) and so performing + // the clamping here based on range of output component. Also, manually adding + // the sign bit in the appropriate spot. Maybe a better way? + out = std::max(out, min); + out = std::min(out, max); + + outColor[comp] = *(UINT*)&out; + } + break; + } + case SWR_TYPE_FLOAT: + { + if (FormatTraits::GetBPC(comp) == 16) + { + // Convert from 32-bit float to 16-bit float using _mm_cvtps_ph + // @todo 16bit float instruction support is orthogonal to avx support. need to + // add check for F16C support instead. +#if KNOB_ARCH == KNOB_ARCH_AVX2 + __m128 src128 = _mm_set1_ps(src); + __m128i srci128 = _mm_cvtps_ph(src128, _MM_FROUND_TRUNC); + UINT value = _mm_extract_epi16(srci128, 0); +#else + UINT value = Convert32To16Float(src); +#endif + + outColor[comp] = value; + } + else if (FormatTraits::GetBPC(comp) == 11) + { + outColor[comp] = Convert32ToSmallFloat<6>(src); + } + else if (FormatTraits::GetBPC(comp) == 10) + { + outColor[comp] = Convert32ToSmallFloat<5>(src); + } + else + { + outColor[comp] = *(UINT*)&src; + } + + break; + } + default: + SWR_ASSERT(0); + break; + } + } + + typename FormatTraits::FormatT* pPixel = (typename FormatTraits::FormatT*)pDstPixel; + + switch (FormatTraits::numComps) + { + case 4: + pPixel->a = outColor[3]; + case 3: + pPixel->b = outColor[2]; + case 2: + pPixel->g = outColor[1]; + case 1: + pPixel->r = outColor[0]; + break; + default: + SWR_ASSERT(0); + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Convert pixel in any format to float32 +/// @param pDstPixel - Pointer to destination pixel. +/// @param srcPixel - Pointer to source pixel +template +INLINE static void ConvertPixelToFloat( + float dstPixel[4], + const BYTE* pSrc) +{ + UINT srcColor[4]; // typeless bits + + // unpack src pixel + typename FormatTraits::FormatT* pPixel = (typename FormatTraits::FormatT*)pSrc; + + // apply format defaults + for (uint32_t comp = 0; comp < 4; ++comp) + { + uint32_t def = FormatTraits::GetDefault(comp); + dstPixel[comp] = *(float*)&def; + } + + // load format data + switch (FormatTraits::numComps) + { + case 4: + srcColor[3] = pPixel->a; + case 3: + srcColor[2] = pPixel->b; + case 2: + srcColor[1] = pPixel->g; + case 1: + srcColor[0] = pPixel->r; + break; + default: + SWR_ASSERT(0); + } + + // Convert components + for (UINT comp = 0; comp < FormatTraits::numComps; ++comp) + { + SWR_TYPE type = FormatTraits::GetType(comp); + + UINT src = srcColor[comp]; + + switch (type) + { + case SWR_TYPE_UNORM: + { + float dst; + if (FormatTraits::isSRGB && comp != 3) + { + dst = *(float*)&srgb8Table[src]; + } + else + { + // component sizes > 16 must use fp divide to maintain ulp requirements + if (FormatTraits::GetBPC(comp) > 16) + { + dst = (float)src / (float)((1 << FormatTraits::GetBPC(comp)) - 1); + } + else + { + const float scale = (1.0f / (float)((1 << FormatTraits::GetBPC(comp)) - 1)); + dst = (float)src * scale; + } + } + dstPixel[FormatTraits::swizzle(comp)] = dst; + break; + } + case SWR_TYPE_SNORM: + { + SWR_ASSERT(!FormatTraits::isSRGB); + + float dst; + if (src == 0x10) + { + dst = -1.0f; + } + else + { + switch (FormatTraits::GetBPC(comp)) + { + case 8: + dst = (float)((int8_t)src); + break; + case 16: + dst = (float)((int16_t)src); + break; + case 32: + dst = (float)((int32_t)src); + break; + default: + assert(0 && "attempted to load from SNORM with unsupported bpc"); + dst = 0.0f; + break; + } + dst = dst * (1.0f / ((1 << (FormatTraits::GetBPC(comp) - 1)) - 1)); + } + dstPixel[FormatTraits::swizzle(comp)] = dst; + break; + } + case SWR_TYPE_UINT: + { + UINT dst = (UINT)src; + dstPixel[FormatTraits::swizzle(comp)] = *(float*)&dst; + break; + } + case SWR_TYPE_SINT: + { + int dst; + switch (FormatTraits::GetBPC(comp)) + { + case 8: + dst = (int8_t)src; + break; + case 16: + dst = (int16_t)src; + break; + case 32: + dst = (int32_t)src; + break; + default: + assert(0 && "attempted to load from SINT with unsupported bpc"); + dst = 0; + break; + } + dstPixel[FormatTraits::swizzle(comp)] = *(float*)&dst; + break; + } + case SWR_TYPE_FLOAT: + { + float dst; + if (FormatTraits::GetBPC(comp) == 16) + { +#if KNOB_ARCH == KNOB_ARCH_AVX2 + // Convert from 16-bit float to 32-bit float using _mm_cvtph_ps + // @todo 16bit float instruction support is orthogonal to avx support. need to + // add check for F16C support instead. + __m128i src128 = _mm_set1_epi32(src); + __m128 res = _mm_cvtph_ps(src128); + _mm_store_ss(&dst, res); +#else + dst = ConvertSmallFloatTo32(src); +#endif + } + else if (FormatTraits::GetBPC(comp) == 11) + { + dst = ConvertSmallFloatTo32(src << 4); + } + else if (FormatTraits::GetBPC(comp) == 10) + { + dst = ConvertSmallFloatTo32(src << 5); + } + else + { + dst = *(float*)&src; + } + + dstPixel[FormatTraits::swizzle(comp)] = *(float*)&dst; + break; + } + default: + SWR_ASSERT(0); + break; + } + } +} + +// non-templated version of conversion functions +INLINE static void ConvertPixelFromFloat( + SWR_FORMAT format, + uint8_t* pDst, + const float srcPixel[4]) +{ + switch (format) + { + case R32G32B32A32_FLOAT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32B32A32_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32B32A32_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32B32X32_FLOAT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32B32A32_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32B32A32_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32B32_FLOAT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32B32_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32B32_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32B32_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32B32_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16A16_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16A16_SNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16A16_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16A16_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16A16_FLOAT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32_FLOAT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32_FLOAT_X8X24_TYPELESS: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16X16_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16X16_FLOAT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16A16_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16A16_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32G32_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32_FLOAT_X8X24_TYPELESS_LD: ConvertPixelFromFloat(pDst, srcPixel); break; + case B8G8R8A8_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case B8G8R8A8_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case R10G10B10A2_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R10G10B10A2_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case R10G10B10A2_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8A8_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8A8_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8A8_SNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8A8_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8A8_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16_SNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16_FLOAT: ConvertPixelFromFloat(pDst, srcPixel); break; + case B10G10R10A2_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case B10G10R10A2_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case R11G11B10_FLOAT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32_FLOAT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R24_UNORM_X8_TYPELESS: ConvertPixelFromFloat(pDst, srcPixel); break; + case R24_UNORM_X8_TYPELESS_LD: ConvertPixelFromFloat(pDst, srcPixel); break; + case A32_FLOAT: ConvertPixelFromFloat(pDst, srcPixel); break; + case B8G8R8X8_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case B8G8R8X8_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8X8_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8X8_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case R9G9B9E5_SHAREDEXP: ConvertPixelFromFloat(pDst, srcPixel); break; + case B10G10R10X2_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R10G10B10X2_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8A8_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8A8_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R32_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case B5G6R5_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case B5G6R5_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case B5G5R5A1_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case B5G5R5A1_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case B4G4R4A4_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case B4G4R4A4_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8_SNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16_SNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16_FLOAT: ConvertPixelFromFloat(pDst, srcPixel); break; + case A16_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case A16_FLOAT: ConvertPixelFromFloat(pDst, srcPixel); break; + case B5G5R5X1_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case B5G5R5X1_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8_SNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case A8_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case YCRCB_SWAPUVY: ConvertPixelFromFloat(pDst, srcPixel); break; + case BC1_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case BC2_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case BC3_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case BC4_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case BC5_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case BC1_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case BC2_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case BC3_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case YCRCB_SWAPUV: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8_SNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case BC4_SNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case BC5_SNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16_FLOAT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16_SNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case BC7_UNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case BC7_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8_UNORM_SRGB: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R16G16B16_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R10G10B10A2_SNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case R10G10B10A2_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R10G10B10A2_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case R10G10B10A2_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case B10G10R10A2_SNORM: ConvertPixelFromFloat(pDst, srcPixel); break; + case B10G10R10A2_USCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case B10G10R10A2_SSCALED: ConvertPixelFromFloat(pDst, srcPixel); break; + case B10G10R10A2_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case B10G10R10A2_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8_UINT: ConvertPixelFromFloat(pDst, srcPixel); break; + case R8G8B8_SINT: ConvertPixelFromFloat(pDst, srcPixel); break; + default: + break; + } +} + + diff --git a/src/gallium/drivers/swr/rasterizer/memory/LoadTile.cpp b/src/gallium/drivers/swr/rasterizer/memory/LoadTile.cpp new file mode 100644 index 00000000000..5d9c0045a8a --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/memory/LoadTile.cpp @@ -0,0 +1,396 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file LoadTile.cpp +* +* @brief Functionality for Load +* +******************************************************************************/ +#include "common/os.h" +#include "common/formats.h" +#include "core/context.h" +#include "core/rdtsc_core.h" +#include "memory/TilingFunctions.h" +#include "memory/tilingtraits.h" +#include "memory/Convert.h" + +typedef void(*PFN_LOAD_TILES)(SWR_SURFACE_STATE*, uint8_t*, uint32_t, uint32_t, uint32_t); + +////////////////////////////////////////////////////////////////////////// +/// Load Raster Tile Function Tables. +////////////////////////////////////////////////////////////////////////// +static PFN_LOAD_TILES sLoadTilesColorTable_SWR_TILE_NONE[NUM_SWR_FORMATS]; +static PFN_LOAD_TILES sLoadTilesDepthTable_SWR_TILE_NONE[NUM_SWR_FORMATS]; + +static PFN_LOAD_TILES sLoadTilesColorTable_SWR_TILE_MODE_YMAJOR[NUM_SWR_FORMATS]; +static PFN_LOAD_TILES sLoadTilesColorTable_SWR_TILE_MODE_XMAJOR[NUM_SWR_FORMATS]; + +static PFN_LOAD_TILES sLoadTilesDepthTable_SWR_TILE_MODE_YMAJOR[NUM_SWR_FORMATS]; + +////////////////////////////////////////////////////////////////////////// +/// LoadRasterTile +////////////////////////////////////////////////////////////////////////// +template +struct LoadRasterTile +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Retrieve color from hot tile source which is always float. + /// @param pSrc - Pointer to raster tile. + /// @param x, y - Coordinates to raster tile. + /// @param output - output color + INLINE static void SetSwizzledDstColor( + const float srcColor[4], + uint32_t x, uint32_t y, + uint8_t* pDst) + { + typedef SimdTile SimdT; + + SimdT* pDstSimdTiles = (SimdT*)pDst; + + // Compute which simd tile we're accessing within 8x8 tile. + // i.e. Compute linear simd tile coordinate given (x, y) in pixel coordinates. + uint32_t simdIndex = (y / SIMD_TILE_Y_DIM) * (KNOB_TILE_X_DIM / SIMD_TILE_X_DIM) + (x / SIMD_TILE_X_DIM); + + SimdT* pSimdTile = &pDstSimdTiles[simdIndex]; + + uint32_t simdOffset = (y % SIMD_TILE_Y_DIM) * SIMD_TILE_X_DIM + (x % SIMD_TILE_X_DIM); + + pSimdTile->SetSwizzledColor(simdOffset, srcColor); + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Loads an 8x8 raster tile from the src surface. + /// @param pSrcSurface - Src surface state + /// @param pDst - Destination hot tile pointer + /// @param x, y - Coordinates to raster tile. + INLINE static void Load( + SWR_SURFACE_STATE* pSrcSurface, + uint8_t* pDst, + uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) // (x, y) pixel coordinate to start of raster tile. + { + uint32_t lodWidth = (pSrcSurface->width == 1) ? 1 : pSrcSurface->width >> pSrcSurface->lod; + uint32_t lodHeight = (pSrcSurface->height == 1) ? 1 : pSrcSurface->height >> pSrcSurface->lod; + + // For each raster tile pixel (rx, ry) + for (uint32_t ry = 0; ry < KNOB_TILE_Y_DIM; ++ry) + { + for (uint32_t rx = 0; rx < KNOB_TILE_X_DIM; ++rx) + { + if (((x + rx) < lodWidth) && + ((y + ry) < lodHeight)) + { + uint8_t* pSrc = (uint8_t*)ComputeSurfaceAddress(x + rx, y + ry, pSrcSurface->arrayIndex + renderTargetArrayIndex, + pSrcSurface->arrayIndex + renderTargetArrayIndex, sampleNum, + pSrcSurface->lod, pSrcSurface); + + float srcColor[4]; + ConvertPixelToFloat(srcColor, pSrc); + + // store pixel to hottile + SetSwizzledDstColor(srcColor, rx, ry, pDst); + } + } + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// LoadMacroTile - Loads a macro tile which consists of raster tiles. +////////////////////////////////////////////////////////////////////////// +template +struct LoadMacroTile +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Load a macrotile to the destination surface. + /// @param pSrc - Pointer to macro tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to macro tile + static void Load( + SWR_SURFACE_STATE* pSrcSurface, + uint8_t *pDstHotTile, + uint32_t x, uint32_t y, uint32_t renderTargetArrayIndex) + { + // Load each raster tile from the hot tile to the destination surface. + for (uint32_t row = 0; row < KNOB_MACROTILE_Y_DIM; row += KNOB_TILE_Y_DIM) + { + for (uint32_t col = 0; col < KNOB_MACROTILE_X_DIM; col += KNOB_TILE_X_DIM) + { + for (uint32_t sampleNum = 0; sampleNum < pSrcSurface->numSamples; sampleNum++) + { + LoadRasterTile::Load(pSrcSurface, pDstHotTile, + (x + col), (y + row), sampleNum, renderTargetArrayIndex); + pDstHotTile += KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * (FormatTraits::bpp / 8); + } + } + } + } +}; + +static void BUCKETS_START(UINT id) +{ +#ifdef KNOB_ENABLE_RDTSC + gBucketMgr.StartBucket(id); +#endif +} + +static void BUCKETS_STOP(UINT id) +{ +#ifdef KNOB_ENABLE_RDTSC + gBucketMgr.StopBucket(id); +#endif +} + +// on demand buckets for load tiles +static std::vector sBuckets(NUM_SWR_FORMATS, -1); +static std::mutex sBucketMutex; + +////////////////////////////////////////////////////////////////////////// +/// @brief Loads a full hottile from a render surface +/// @param hPrivateContext - Handle to private DC +/// @param dstFormat - Format for hot tile. +/// @param renderTargetIndex - Index to src render target +/// @param x, y - Coordinates to raster tile. +/// @param pDstHotTile - Pointer to Hot Tile +void LoadHotTile( + SWR_SURFACE_STATE *pSrcSurface, + SWR_FORMAT dstFormat, + SWR_RENDERTARGET_ATTACHMENT renderTargetIndex, + uint32_t x, uint32_t y, uint32_t renderTargetArrayIndex, + uint8_t *pDstHotTile) +{ + PFN_LOAD_TILES pfnLoadTiles = NULL; + + // don't need to load null surfaces + if (pSrcSurface->type == SURFACE_NULL) + { + return; + } + + // force 0 if requested renderTargetArrayIndex is OOB + if (renderTargetArrayIndex >= pSrcSurface->depth) + { + renderTargetArrayIndex = 0; + } + + if (renderTargetIndex < SWR_ATTACHMENT_DEPTH) + { + switch (pSrcSurface->tileMode) + { + case SWR_TILE_NONE: + pfnLoadTiles = sLoadTilesColorTable_SWR_TILE_NONE[pSrcSurface->format]; + break; + case SWR_TILE_MODE_YMAJOR: + pfnLoadTiles = sLoadTilesColorTable_SWR_TILE_MODE_YMAJOR[pSrcSurface->format]; + break; + case SWR_TILE_MODE_XMAJOR: + pfnLoadTiles = sLoadTilesColorTable_SWR_TILE_MODE_XMAJOR[pSrcSurface->format]; + break; + case SWR_TILE_MODE_WMAJOR: + SWR_ASSERT(pSrcSurface->format == R8_UINT); + pfnLoadTiles = LoadMacroTile, R8_UINT, R8_UINT>::Load; + break; + default: + SWR_ASSERT(0, "Unsupported tiling mode"); + break; + } + } + else if (renderTargetIndex == SWR_ATTACHMENT_DEPTH) + { + // Currently depth can map to linear and tile-y. + switch (pSrcSurface->tileMode) + { + case SWR_TILE_NONE: + pfnLoadTiles = sLoadTilesDepthTable_SWR_TILE_NONE[pSrcSurface->format]; + break; + case SWR_TILE_MODE_YMAJOR: + pfnLoadTiles = sLoadTilesDepthTable_SWR_TILE_MODE_YMAJOR[pSrcSurface->format]; + break; + default: + SWR_ASSERT(0, "Unsupported tiling mode"); + break; + } + } + else + { + SWR_ASSERT(renderTargetIndex == SWR_ATTACHMENT_STENCIL); + SWR_ASSERT(pSrcSurface->format == R8_UINT); + switch (pSrcSurface->tileMode) + { + case SWR_TILE_NONE: + pfnLoadTiles = LoadMacroTile, R8_UINT, R8_UINT>::Load; + break; + case SWR_TILE_MODE_WMAJOR: + pfnLoadTiles = LoadMacroTile, R8_UINT, R8_UINT>::Load; + break; + default: + SWR_ASSERT(0, "Unsupported tiling mode"); + break; + } + } + + if (pfnLoadTiles == nullptr) + { + SWR_ASSERT(false, "Unsupported format for load tile"); + return; + } + + // Load a macro tile. +#ifdef KNOB_ENABLE_RDTSC + if (sBuckets[pSrcSurface->format] == -1) + { + // guard sBuckets update since storetiles is called by multiple threads + sBucketMutex.lock(); + if (sBuckets[pSrcSurface->format] == -1) + { + const SWR_FORMAT_INFO& info = GetFormatInfo(pSrcSurface->format); + BUCKET_DESC desc{ info.name, "", false, 0xffffffff }; + sBuckets[pSrcSurface->format] = gBucketMgr.RegisterBucket(desc); + } + sBucketMutex.unlock(); + } +#endif + + BUCKETS_START(sBuckets[pSrcSurface->format]); + pfnLoadTiles(pSrcSurface, pDstHotTile, x, y, renderTargetArrayIndex); + BUCKETS_STOP(sBuckets[pSrcSurface->format]); +} + +////////////////////////////////////////////////////////////////////////// +/// INIT_LOAD_TILES_TABLE - Helper macro for setting up the tables. +#define INIT_LOAD_TILES_COLOR_TABLE(tilemode) \ + memset(sLoadTilesColorTable_##tilemode, 0, sizeof(sLoadTilesColorTable_##tilemode)); \ + \ + sLoadTilesColorTable_##tilemode[R32G32B32A32_FLOAT] = LoadMacroTile, R32G32B32A32_FLOAT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R32G32B32A32_SINT] = LoadMacroTile, R32G32B32A32_SINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R32G32B32A32_UINT] = LoadMacroTile, R32G32B32A32_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R32G32B32X32_FLOAT] = LoadMacroTile, R32G32B32X32_FLOAT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R32G32B32_FLOAT] = LoadMacroTile, R32G32B32_FLOAT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R32G32B32_SINT] = LoadMacroTile, R32G32B32_SINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R32G32B32_UINT] = LoadMacroTile, R32G32B32_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16B16A16_UNORM] = LoadMacroTile, R16G16B16A16_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16B16A16_SNORM] = LoadMacroTile, R16G16B16A16_SNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16B16A16_SINT] = LoadMacroTile, R16G16B16A16_SINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16B16A16_UINT] = LoadMacroTile, R16G16B16A16_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16B16A16_FLOAT] = LoadMacroTile, R16G16B16A16_FLOAT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R32G32_FLOAT] = LoadMacroTile, R32G32_FLOAT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R32G32_SINT] = LoadMacroTile, R32G32_SINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R32G32_UINT] = LoadMacroTile, R32G32_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16B16X16_UNORM] = LoadMacroTile, R16G16B16X16_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16B16X16_FLOAT] = LoadMacroTile, R16G16B16X16_FLOAT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B8G8R8A8_UNORM] = LoadMacroTile, B8G8R8A8_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B8G8R8A8_UNORM_SRGB] = LoadMacroTile, B8G8R8A8_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R10G10B10A2_UNORM] = LoadMacroTile, R10G10B10A2_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R10G10B10A2_UNORM_SRGB] = LoadMacroTile, R10G10B10A2_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R10G10B10A2_UINT] = LoadMacroTile, R10G10B10A2_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8B8A8_UNORM] = LoadMacroTile, R8G8B8A8_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8B8A8_UNORM_SRGB] = LoadMacroTile, R8G8B8A8_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8B8A8_SNORM] = LoadMacroTile, R8G8B8A8_SNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8B8A8_SINT] = LoadMacroTile, R8G8B8A8_SINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8B8A8_UINT] = LoadMacroTile, R8G8B8A8_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16_UNORM] = LoadMacroTile, R16G16_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16_SNORM] = LoadMacroTile, R16G16_SNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16_SINT] = LoadMacroTile, R16G16_SINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16_UINT] = LoadMacroTile, R16G16_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16_FLOAT] = LoadMacroTile, R16G16_FLOAT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B10G10R10A2_UNORM] = LoadMacroTile, B10G10R10A2_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B10G10R10A2_UNORM_SRGB] = LoadMacroTile, B10G10R10A2_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R11G11B10_FLOAT] = LoadMacroTile, R11G11B10_FLOAT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R32_SINT] = LoadMacroTile, R32_SINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R32_UINT] = LoadMacroTile, R32_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R32_FLOAT] = LoadMacroTile, R32_FLOAT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[A32_FLOAT] = LoadMacroTile, A32_FLOAT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B8G8R8X8_UNORM] = LoadMacroTile, B8G8R8X8_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B8G8R8X8_UNORM_SRGB] = LoadMacroTile, B8G8R8X8_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8B8X8_UNORM] = LoadMacroTile, R8G8B8X8_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8B8X8_UNORM_SRGB] = LoadMacroTile, R8G8B8X8_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B10G10R10X2_UNORM] = LoadMacroTile, B10G10R10X2_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B5G6R5_UNORM] = LoadMacroTile, B5G6R5_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B5G6R5_UNORM_SRGB] = LoadMacroTile, B5G6R5_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B5G5R5A1_UNORM] = LoadMacroTile, B5G5R5A1_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B5G5R5A1_UNORM_SRGB] = LoadMacroTile, B5G5R5A1_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B4G4R4A4_UNORM] = LoadMacroTile, B4G4R4A4_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B4G4R4A4_UNORM_SRGB] = LoadMacroTile, B4G4R4A4_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8_UNORM] = LoadMacroTile, R8G8_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8_SNORM] = LoadMacroTile, R8G8_SNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8_SINT] = LoadMacroTile, R8G8_SINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8_UINT] = LoadMacroTile, R8G8_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16_UNORM] = LoadMacroTile, R16_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16_SNORM] = LoadMacroTile, R16_SNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16_SINT] = LoadMacroTile, R16_SINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16_UINT] = LoadMacroTile, R16_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16_FLOAT] = LoadMacroTile, R16_FLOAT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[A16_UNORM] = LoadMacroTile, A16_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[A16_FLOAT] = LoadMacroTile, A16_FLOAT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B5G5R5X1_UNORM] = LoadMacroTile, B5G5R5X1_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B5G5R5X1_UNORM_SRGB] = LoadMacroTile, B5G5R5X1_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8_UNORM] = LoadMacroTile, R8_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8_SNORM] = LoadMacroTile, R8_SNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8_SINT] = LoadMacroTile, R8_SINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8_UINT] = LoadMacroTile, R8_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[A8_UNORM] = LoadMacroTile, A8_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[BC1_UNORM] = LoadMacroTile, BC1_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[BC2_UNORM] = LoadMacroTile, BC2_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[BC3_UNORM] = LoadMacroTile, BC3_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[BC4_UNORM] = LoadMacroTile, BC4_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[BC5_UNORM] = LoadMacroTile, BC5_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[BC1_UNORM_SRGB] = LoadMacroTile, BC1_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[BC2_UNORM_SRGB] = LoadMacroTile, BC2_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[BC3_UNORM_SRGB] = LoadMacroTile, BC3_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8B8_UNORM] = LoadMacroTile, R8G8B8_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8B8_SNORM] = LoadMacroTile, R8G8B8_SNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[BC4_SNORM] = LoadMacroTile, BC4_SNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[BC5_SNORM] = LoadMacroTile, BC5_SNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16B16_FLOAT] = LoadMacroTile, R16G16B16_FLOAT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16B16_UNORM] = LoadMacroTile, R16G16B16_UNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16B16_SNORM] = LoadMacroTile, R16G16B16_SNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8B8_UNORM_SRGB] = LoadMacroTile, R8G8B8_UNORM_SRGB, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16B16_UINT] = LoadMacroTile, R16G16B16_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R16G16B16_SINT] = LoadMacroTile, R16G16B16_SINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R10G10B10A2_SNORM] = LoadMacroTile, R10G10B10A2_SNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R10G10B10A2_SINT] = LoadMacroTile, R10G10B10A2_SINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B10G10R10A2_SNORM] = LoadMacroTile, B10G10R10A2_SNORM, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B10G10R10A2_UINT] = LoadMacroTile, B10G10R10A2_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[B10G10R10A2_SINT] = LoadMacroTile, B10G10R10A2_SINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8B8_UINT] = LoadMacroTile, R8G8B8_UINT, R32G32B32A32_FLOAT>::Load; \ + sLoadTilesColorTable_##tilemode[R8G8B8_SINT] = LoadMacroTile, R8G8B8_SINT, R32G32B32A32_FLOAT>::Load; \ + +////////////////////////////////////////////////////////////////////////// +/// INIT_LOAD_TILES_TABLE - Helper macro for setting up the tables. +#define INIT_LOAD_TILES_DEPTH_TABLE(tilemode) \ + memset(sLoadTilesDepthTable_##tilemode, 0, sizeof(sLoadTilesDepthTable_##tilemode)); \ + \ + sLoadTilesDepthTable_##tilemode[R16_UNORM] = LoadMacroTile, R16_UNORM, R32_FLOAT>::Load; \ + sLoadTilesDepthTable_##tilemode[R32_FLOAT] = LoadMacroTile, R32_FLOAT, R32_FLOAT>::Load; \ + sLoadTilesDepthTable_##tilemode[R24_UNORM_X8_TYPELESS] = LoadMacroTile, R24_UNORM_X8_TYPELESS, R32_FLOAT>::Load; \ + +////////////////////////////////////////////////////////////////////////// +/// @brief Sets up tables for LoadTile +void InitSimLoadTilesTable() +{ + INIT_LOAD_TILES_COLOR_TABLE(SWR_TILE_NONE); + INIT_LOAD_TILES_DEPTH_TABLE(SWR_TILE_NONE); + + INIT_LOAD_TILES_COLOR_TABLE(SWR_TILE_MODE_YMAJOR); + INIT_LOAD_TILES_COLOR_TABLE(SWR_TILE_MODE_XMAJOR); + + INIT_LOAD_TILES_DEPTH_TABLE(SWR_TILE_MODE_YMAJOR); +} diff --git a/src/gallium/drivers/swr/rasterizer/memory/StoreTile.cpp b/src/gallium/drivers/swr/rasterizer/memory/StoreTile.cpp new file mode 100644 index 00000000000..9ed1d0bd0ec --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/memory/StoreTile.cpp @@ -0,0 +1,1717 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file StoreTile.cpp +* +* @brief Functionality for Store. +* +******************************************************************************/ +#include "common/os.h" +#include "common/formats.h" +#include "core/context.h" +#include "core/rdtsc_core.h" +#include "core/format_conversion.h" + +#include "memory/TilingFunctions.h" +#include "memory/tilingtraits.h" +#include "memory/Convert.h" +#include "core/multisample.h" + +#include +#include + +typedef void(*PFN_STORE_TILES)(uint8_t*, SWR_SURFACE_STATE*, uint32_t, uint32_t, uint32_t); + +////////////////////////////////////////////////////////////////////////// +/// Store Raster Tile Function Tables. +////////////////////////////////////////////////////////////////////////// +static PFN_STORE_TILES sStoreTilesTableColor[SWR_TILE_MODE_COUNT][NUM_SWR_FORMATS] = {}; +static PFN_STORE_TILES sStoreTilesTableDepth[SWR_TILE_MODE_COUNT][NUM_SWR_FORMATS] = {}; +static PFN_STORE_TILES sStoreTilesTableStencil[SWR_TILE_MODE_COUNT][NUM_SWR_FORMATS] = {}; + +////////////////////////////////////////////////////////////////////////// +/// StorePixels +/// @brief Stores a 4x2 (AVX) raster-tile to two rows. +/// @param pSrc - Pointer to source raster tile in SWRZ pixel order +/// @param ppDsts - Array of destination pointers. Each pointer is +/// to a single row of at most 16B. +/// @tparam NumDests - Number of destination pointers. Each pair of +/// pointers is for a 16-byte column of two rows. +////////////////////////////////////////////////////////////////////////// +template +struct StorePixels +{ + static void Store(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests]) = delete; +}; + +////////////////////////////////////////////////////////////////////////// +/// StorePixels (32-bit pixel specialization) +/// @brief Stores a 4x2 (AVX) raster-tile to two rows. +/// @param pSrc - Pointer to source raster tile in SWRZ pixel order +/// @param ppDsts - Array of destination pointers. Each pointer is +/// to a single row of at most 16B. +/// @tparam NumDests - Number of destination pointers. Each pair of +/// pointers is for a 16-byte column of two rows. +////////////////////////////////////////////////////////////////////////// +template <> +struct StorePixels<8, 2> +{ + static void Store(const uint8_t* pSrc, uint8_t* (&ppDsts)[2]) + { + // Each 4-pixel row is 4 bytes. + const uint16_t* pPixSrc = (const uint16_t*)pSrc; + + // Unswizzle from SWR-Z order + uint16_t* pRow = (uint16_t*)ppDsts[0]; + pRow[0] = pPixSrc[0]; + pRow[1] = pPixSrc[2]; + + pRow = (uint16_t*)ppDsts[1]; + pRow[0] = pPixSrc[1]; + pRow[1] = pPixSrc[3]; + } +}; + +////////////////////////////////////////////////////////////////////////// +/// StorePixels (32-bit pixel specialization) +/// @brief Stores a 4x2 (AVX) raster-tile to two rows. +/// @param pSrc - Pointer to source raster tile in SWRZ pixel order +/// @param ppDsts - Array of destination pointers. Each pointer is +/// to a single row of at most 16B. +/// @tparam NumDests - Number of destination pointers. Each pair of +/// pointers is for a 16-byte column of two rows. +////////////////////////////////////////////////////////////////////////// +template <> +struct StorePixels<16, 2> +{ + static void Store(const uint8_t* pSrc, uint8_t* (&ppDsts)[2]) + { + // Each 4-pixel row is 8 bytes. + const uint32_t* pPixSrc = (const uint32_t*)pSrc; + + // Unswizzle from SWR-Z order + uint32_t* pRow = (uint32_t*)ppDsts[0]; + pRow[0] = pPixSrc[0]; + pRow[1] = pPixSrc[2]; + + pRow = (uint32_t*)ppDsts[1]; + pRow[0] = pPixSrc[1]; + pRow[1] = pPixSrc[3]; + } +}; + +////////////////////////////////////////////////////////////////////////// +/// StorePixels (32-bit pixel specialization) +/// @brief Stores a 4x2 (AVX) raster-tile to two rows. +/// @param pSrc - Pointer to source raster tile in SWRZ pixel order +/// @param ppDsts - Array of destination pointers. Each pointer is +/// to a single row of at most 16B. +/// @tparam NumDests - Number of destination pointers. Each pair of +/// pointers is for a 16-byte column of two rows. +////////////////////////////////////////////////////////////////////////// +template <> +struct StorePixels<32, 2> +{ + static void Store(const uint8_t* pSrc, uint8_t* (&ppDsts)[2]) + { + // Each 4-pixel row is 16-bytes + __m128i *pZRow01 = (__m128i*)pSrc; + __m128i vQuad00 = _mm_load_si128(pZRow01); + __m128i vQuad01 = _mm_load_si128(pZRow01 + 1); + + __m128i vRow00 = _mm_unpacklo_epi64(vQuad00, vQuad01); + __m128i vRow10 = _mm_unpackhi_epi64(vQuad00, vQuad01); + + _mm_storeu_si128((__m128i*)ppDsts[0], vRow00); + _mm_storeu_si128((__m128i*)ppDsts[1], vRow10); + } +}; + +////////////////////////////////////////////////////////////////////////// +/// StorePixels (32-bit pixel specialization) +/// @brief Stores a 4x2 (AVX) raster-tile to two rows. +/// @param pSrc - Pointer to source raster tile in SWRZ pixel order +/// @param ppDsts - Array of destination pointers. Each pointer is +/// to a single row of at most 16B. +/// @tparam NumDests - Number of destination pointers. Each pair of +/// pointers is for a 16-byte column of two rows. +////////////////////////////////////////////////////////////////////////// +template <> +struct StorePixels<64, 4> +{ + static void Store(const uint8_t* pSrc, uint8_t* (&ppDsts)[4]) + { + // Each 4-pixel row is 32 bytes. + const __m128i* pPixSrc = (const __m128i*)pSrc; + + // order of pointers match SWR-Z layout + __m128i** pvDsts = (__m128i**)&ppDsts[0]; + *pvDsts[0] = pPixSrc[0]; + *pvDsts[1] = pPixSrc[1]; + *pvDsts[2] = pPixSrc[2]; + *pvDsts[3] = pPixSrc[3]; + } +}; + +////////////////////////////////////////////////////////////////////////// +/// StorePixels (32-bit pixel specialization) +/// @brief Stores a 4x2 (AVX) raster-tile to two rows. +/// @param pSrc - Pointer to source raster tile in SWRZ pixel order +/// @param ppDsts - Array of destination pointers. Each pointer is +/// to a single row of at most 16B. +/// @tparam NumDests - Number of destination pointers. Each pair of +/// pointers is for a 16-byte column of two rows. +////////////////////////////////////////////////////////////////////////// +template <> +struct StorePixels<128, 8> +{ + static void Store(const uint8_t* pSrc, uint8_t* (&ppDsts)[8]) + { + // Each 4-pixel row is 64 bytes. + const __m128i* pPixSrc = (const __m128i*)pSrc; + + // Unswizzle from SWR-Z order + __m128i** pvDsts = (__m128i**)&ppDsts[0]; + *pvDsts[0] = pPixSrc[0]; + *pvDsts[1] = pPixSrc[2]; + *pvDsts[2] = pPixSrc[1]; + *pvDsts[3] = pPixSrc[3]; + *pvDsts[4] = pPixSrc[4]; + *pvDsts[5] = pPixSrc[6]; + *pvDsts[6] = pPixSrc[5]; + *pvDsts[7] = pPixSrc[7]; + } +}; + +////////////////////////////////////////////////////////////////////////// +/// ConvertPixelsSOAtoAOS - Conversion for SIMD pixel (4x2 or 2x2) +////////////////////////////////////////////////////////////////////////// +template +struct ConvertPixelsSOAtoAOS +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Converts a SIMD from the Hot Tile to the destination format + /// and converts from SOA to AOS. + /// @param pSrc - Pointer to raster tile. + /// @param pDst - Pointer to destination surface or deswizzling buffer. + template + INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests]) + { + static const uint32_t MAX_RASTER_TILE_BYTES = 128; // 8 pixels * 16 bytes per pixel + + OSALIGNSIMD(uint8_t) soaTile[MAX_RASTER_TILE_BYTES]; + OSALIGNSIMD(uint8_t) aosTile[MAX_RASTER_TILE_BYTES]; + + // Convert from SrcFormat --> DstFormat + simdvector src; + LoadSOA(pSrc, src); + StoreSOA(src, soaTile); + + // Convert from SOA --> AOS + FormatTraits::TransposeT::Transpose(soaTile, aosTile); + + // Store data into destination + StorePixels::bpp, NumDests>::Store(aosTile, ppDsts); + } +}; + +////////////////////////////////////////////////////////////////////////// +/// ConvertPixelsSOAtoAOS - Conversion for SIMD pixel (4x2 or 2x2) +/// Specialization for no format conversion +////////////////////////////////////////////////////////////////////////// +template +struct ConvertPixelsSOAtoAOS +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Converts a SIMD from the Hot Tile to the destination format + /// and converts from SOA to AOS. + /// @param pSrc - Pointer to raster tile. + /// @param pDst - Pointer to destination surface or deswizzling buffer. + template + INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests]) + { + static const uint32_t MAX_RASTER_TILE_BYTES = 128; // 8 pixels * 16 bytes per pixel + + OSALIGNSIMD(uint8_t) aosTile[MAX_RASTER_TILE_BYTES]; + + // Convert from SOA --> AOS + FormatTraits::TransposeT::Transpose(pSrc, aosTile); + + // Store data into destination + StorePixels::bpp, NumDests>::Store(aosTile, ppDsts); + } +}; + +////////////////////////////////////////////////////////////////////////// +/// ConvertPixelsSOAtoAOS - Specialization conversion for B5G6R6_UNORM +////////////////////////////////////////////////////////////////////////// +template<> +struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, B5G6R5_UNORM > +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Converts a SIMD from the Hot Tile to the destination format + /// and converts from SOA to AOS. + /// @param pSrc - Pointer to raster tile. + /// @param pDst - Pointer to destination surface or deswizzling buffer. + template + INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests]) + { + static const SWR_FORMAT SrcFormat = R32G32B32A32_FLOAT; + static const SWR_FORMAT DstFormat = B5G6R5_UNORM; + static const uint32_t MAX_RASTER_TILE_BYTES = 128; // 8 pixels * 16 bytes per pixel + + OSALIGNSIMD(uint8_t) aosTile[MAX_RASTER_TILE_BYTES]; + + // Load hot-tile + simdvector src, dst; + LoadSOA(pSrc, src); + + // deswizzle + dst.x = src[FormatTraits::swizzle(0)]; + dst.y = src[FormatTraits::swizzle(1)]; + dst.z = src[FormatTraits::swizzle(2)]; + + // clamp + dst.x = Clamp(dst.x, 0); + dst.y = Clamp(dst.y, 1); + dst.z = Clamp(dst.z, 2); + + // normalize + dst.x = Normalize(dst.x, 0); + dst.y = Normalize(dst.y, 1); + dst.z = Normalize(dst.z, 2); + + // pack + simdscalari packed = _simd_castps_si(dst.x); + packed = _simd_or_si(packed, _simd_slli_epi32(_simd_castps_si(dst.y), FormatTraits::GetBPC(0))); + packed = _simd_or_si(packed, _simd_slli_epi32(_simd_castps_si(dst.z), FormatTraits::GetBPC(0) + + FormatTraits::GetBPC(1))); + + // pack low 16 bits of each 32 bit lane to low 128 bits of dst + uint32_t *pPacked = (uint32_t*)&packed; + uint16_t *pAosTile = (uint16_t*)&aosTile[0]; + for (uint32_t t = 0; t < KNOB_SIMD_WIDTH; ++t) + { + *pAosTile++ = *pPacked++; + } + + // Store data into destination + StorePixels::bpp, NumDests>::Store(aosTile, ppDsts); + } +}; + +////////////////////////////////////////////////////////////////////////// +/// ConvertPixelsSOAtoAOS - Conversion for SIMD pixel (4x2 or 2x2) +////////////////////////////////////////////////////////////////////////// +template<> +struct ConvertPixelsSOAtoAOS +{ + static const SWR_FORMAT SrcFormat = R32_FLOAT; + static const SWR_FORMAT DstFormat = R24_UNORM_X8_TYPELESS; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Converts a SIMD from the Hot Tile to the destination format + /// and converts from SOA to AOS. + /// @param pSrc - Pointer to raster tile. + /// @param pDst - Pointer to destination surface or deswizzling buffer. + template + INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests]) + { + static const uint32_t MAX_RASTER_TILE_BYTES = 128; // 8 pixels * 16 bytes per pixel + + OSALIGNSIMD(uint8_t) soaTile[MAX_RASTER_TILE_BYTES]; + OSALIGNSIMD(uint8_t) aosTile[MAX_RASTER_TILE_BYTES]; + + // Convert from SrcFormat --> DstFormat + simdvector src; + LoadSOA(pSrc, src); + StoreSOA(src, soaTile); + + // Convert from SOA --> AOS + FormatTraits::TransposeT::Transpose(soaTile, aosTile); + + // Store data into destination but don't overwrite the X8 bits + // Each 4-pixel row is 16-bytes + __m128i *pZRow01 = (__m128i*)aosTile; + __m128i vQuad00 = _mm_load_si128(pZRow01); + __m128i vQuad01 = _mm_load_si128(pZRow01 + 1); + + __m128i vRow00 = _mm_unpacklo_epi64(vQuad00, vQuad01); + __m128i vRow10 = _mm_unpackhi_epi64(vQuad00, vQuad01); + + __m128i vDst0 = _mm_loadu_si128((const __m128i*)ppDsts[0]); + __m128i vDst1 = _mm_loadu_si128((const __m128i*)ppDsts[1]); + + __m128i vMask = _mm_set1_epi32(0xFFFFFF); + + vDst0 = _mm_andnot_si128(vMask, vDst0); + vDst0 = _mm_or_si128(vDst0, _mm_and_si128(vRow00, vMask)); + vDst1 = _mm_andnot_si128(vMask, vDst1); + vDst1 = _mm_or_si128(vDst1, _mm_and_si128(vRow10, vMask)); + + _mm_storeu_si128((__m128i*)ppDsts[0], vDst0); + _mm_storeu_si128((__m128i*)ppDsts[1], vDst1); + } +}; + +template +INLINE static void FlatConvert(const uint8_t* pSrc, uint8_t* pDst, uint8_t* pDst1) +{ + static const uint32_t offset = sizeof(simdscalar); + + // swizzle rgba -> bgra while we load + simdscalar vComp0 = _simd_load_ps((const float*)(pSrc + (FormatTraits::swizzle(0))*offset)); // float32 rrrrrrrr + simdscalar vComp1 = _simd_load_ps((const float*)(pSrc + (FormatTraits::swizzle(1))*offset)); // float32 gggggggg + simdscalar vComp2 = _simd_load_ps((const float*)(pSrc + (FormatTraits::swizzle(2))*offset)); // float32 bbbbbbbb + simdscalar vComp3 = _simd_load_ps((const float*)(pSrc + (FormatTraits::swizzle(3))*offset)); // float32 aaaaaaaa + + // clamp + vComp0 = _simd_max_ps(vComp0, _simd_setzero_ps()); + vComp0 = _simd_min_ps(vComp0, _simd_set1_ps(1.0f)); + + vComp1 = _simd_max_ps(vComp1, _simd_setzero_ps()); + vComp1 = _simd_min_ps(vComp1, _simd_set1_ps(1.0f)); + + vComp2 = _simd_max_ps(vComp2, _simd_setzero_ps()); + vComp2 = _simd_min_ps(vComp2, _simd_set1_ps(1.0f)); + + vComp3 = _simd_max_ps(vComp3, _simd_setzero_ps()); + vComp3 = _simd_min_ps(vComp3, _simd_set1_ps(1.0f)); + + if (FormatTraits::isSRGB) + { + // Gamma-correct only rgb + vComp0 = FormatTraits::convertSrgb(0, vComp0); + vComp1 = FormatTraits::convertSrgb(1, vComp1); + vComp2 = FormatTraits::convertSrgb(2, vComp2); + } + + // convert float components from 0.0f .. 1.0f to correct scale for 0 .. 255 dest format + vComp0 = _simd_mul_ps(vComp0, _simd_set1_ps(FormatTraits::fromFloat(0))); + vComp1 = _simd_mul_ps(vComp1, _simd_set1_ps(FormatTraits::fromFloat(1))); + vComp2 = _simd_mul_ps(vComp2, _simd_set1_ps(FormatTraits::fromFloat(2))); + vComp3 = _simd_mul_ps(vComp3, _simd_set1_ps(FormatTraits::fromFloat(3))); + + // moving to 8 wide integer vector types + __m256i src0 = _simd_cvtps_epi32(vComp0); // padded byte rrrrrrrr + __m256i src1 = _simd_cvtps_epi32(vComp1); // padded byte gggggggg + __m256i src2 = _simd_cvtps_epi32(vComp2); // padded byte bbbbbbbb + __m256i src3 = _simd_cvtps_epi32(vComp3); // padded byte aaaaaaaa + +#if KNOB_ARCH == KNOB_ARCH_AVX + + // splitting into two sets of 4 wide integer vector types + // because AVX doesn't have instructions to support this operation at 8 wide + __m128i srcLo0 = _mm256_castsi256_si128(src0); // 000r000r000r000r + __m128i srcLo1 = _mm256_castsi256_si128(src1); // 000g000g000g000g + __m128i srcLo2 = _mm256_castsi256_si128(src2); // 000b000b000b000b + __m128i srcLo3 = _mm256_castsi256_si128(src3); // 000a000a000a000a + + __m128i srcHi0 = _mm256_extractf128_si256(src0, 1); // 000r000r000r000r + __m128i srcHi1 = _mm256_extractf128_si256(src1, 1); // 000g000g000g000g + __m128i srcHi2 = _mm256_extractf128_si256(src2, 1); // 000b000b000b000b + __m128i srcHi3 = _mm256_extractf128_si256(src3, 1); // 000a000a000a000a + + srcLo1 = _mm_slli_si128(srcLo1, 1); // 00g000g000g000g0 + srcHi1 = _mm_slli_si128(srcHi1, 1); // 00g000g000g000g0 + srcLo2 = _mm_slli_si128(srcLo2, 2); // 0b000b000b000b00 + srcHi2 = _mm_slli_si128(srcHi2, 2); // 0b000b000b000b00 + srcLo3 = _mm_slli_si128(srcLo3, 3); // a000a000a000a000 + srcHi3 = _mm_slli_si128(srcHi3, 3); // a000a000a000a000 + + srcLo0 = _mm_or_si128(srcLo0, srcLo1); // 00gr00gr00gr00gr + srcLo2 = _mm_or_si128(srcLo2, srcLo3); // ab00ab00ab00ab00 + + srcHi0 = _mm_or_si128(srcHi0, srcHi1); // 00gr00gr00gr00gr + srcHi2 = _mm_or_si128(srcHi2, srcHi3); // ab00ab00ab00ab00 + + srcLo0 = _mm_or_si128(srcLo0, srcLo2); // abgrabgrabgrabgr + srcHi0 = _mm_or_si128(srcHi0, srcHi2); // abgrabgrabgrabgr + + // unpack into rows that get the tiling order correct + __m128i vRow00 = _mm_unpacklo_epi64(srcLo0, srcHi0); // abgrabgrabgrabgrabgrabgrabgrabgr + __m128i vRow10 = _mm_unpackhi_epi64(srcLo0, srcHi0); + + __m256i final = _mm256_castsi128_si256(vRow00); + final = _mm256_insertf128_si256(final, vRow10, 1); + +#elif KNOB_ARCH == KNOB_ARCH_AVX2 + + // logic is as above, only wider + src1 = _mm256_slli_si256(src1, 1); + src2 = _mm256_slli_si256(src2, 2); + src3 = _mm256_slli_si256(src3, 3); + + src0 = _mm256_or_si256(src0, src1); + src2 = _mm256_or_si256(src2, src3); + + __m256i final = _mm256_or_si256(src0, src2); + + // adjust the data to get the tiling order correct 0 1 2 3 -> 0 2 1 3 + final = _mm256_permute4x64_epi64(final, 0xD8); + +#endif + + _mm256_storeu2_m128i((__m128i*)pDst1, (__m128i*)pDst, final); +} + +template +INLINE static void FlatConvertNoAlpha(const uint8_t* pSrc, uint8_t* pDst, uint8_t* pDst1) +{ + static const uint32_t offset = sizeof(simdscalar); + + // swizzle rgba -> bgra while we load + simdscalar vComp0 = _simd_load_ps((const float*)(pSrc + (FormatTraits::swizzle(0))*offset)); // float32 rrrrrrrr + simdscalar vComp1 = _simd_load_ps((const float*)(pSrc + (FormatTraits::swizzle(1))*offset)); // float32 gggggggg + simdscalar vComp2 = _simd_load_ps((const float*)(pSrc + (FormatTraits::swizzle(2))*offset)); // float32 bbbbbbbb + // clamp + vComp0 = _simd_max_ps(vComp0, _simd_setzero_ps()); + vComp0 = _simd_min_ps(vComp0, _simd_set1_ps(1.0f)); + + vComp1 = _simd_max_ps(vComp1, _simd_setzero_ps()); + vComp1 = _simd_min_ps(vComp1, _simd_set1_ps(1.0f)); + + vComp2 = _simd_max_ps(vComp2, _simd_setzero_ps()); + vComp2 = _simd_min_ps(vComp2, _simd_set1_ps(1.0f)); + + if (FormatTraits::isSRGB) + { + // Gamma-correct only rgb + vComp0 = FormatTraits::convertSrgb(0, vComp0); + vComp1 = FormatTraits::convertSrgb(1, vComp1); + vComp2 = FormatTraits::convertSrgb(2, vComp2); + } + + // convert float components from 0.0f .. 1.0f to correct scale for 0 .. 255 dest format + vComp0 = _simd_mul_ps(vComp0, _simd_set1_ps(FormatTraits::fromFloat(0))); + vComp1 = _simd_mul_ps(vComp1, _simd_set1_ps(FormatTraits::fromFloat(1))); + vComp2 = _simd_mul_ps(vComp2, _simd_set1_ps(FormatTraits::fromFloat(2))); + + // moving to 8 wide integer vector types + __m256i src0 = _simd_cvtps_epi32(vComp0); // padded byte rrrrrrrr + __m256i src1 = _simd_cvtps_epi32(vComp1); // padded byte gggggggg + __m256i src2 = _simd_cvtps_epi32(vComp2); // padded byte bbbbbbbb + +#if KNOB_ARCH == KNOB_ARCH_AVX + + // splitting into two sets of 4 wide integer vector types + // because AVX doesn't have instructions to support this operation at 8 wide + __m128i srcLo0 = _mm256_castsi256_si128(src0); // 000r000r000r000r + __m128i srcLo1 = _mm256_castsi256_si128(src1); // 000g000g000g000g + __m128i srcLo2 = _mm256_castsi256_si128(src2); // 000b000b000b000b + + __m128i srcHi0 = _mm256_extractf128_si256(src0, 1); // 000r000r000r000r + __m128i srcHi1 = _mm256_extractf128_si256(src1, 1); // 000g000g000g000g + __m128i srcHi2 = _mm256_extractf128_si256(src2, 1); // 000b000b000b000b + + srcLo1 = _mm_slli_si128(srcLo1, 1); // 00g000g000g000g0 + srcHi1 = _mm_slli_si128(srcHi1, 1); // 00g000g000g000g0 + srcLo2 = _mm_slli_si128(srcLo2, 2); // 0b000b000b000b00 + srcHi2 = _mm_slli_si128(srcHi2, 2); // 0b000b000b000b00 + + srcLo0 = _mm_or_si128(srcLo0, srcLo1); // 00gr00gr00gr00gr + + srcHi0 = _mm_or_si128(srcHi0, srcHi1); // 00gr00gr00gr00gr + + srcLo0 = _mm_or_si128(srcLo0, srcLo2); // 0bgr0bgr0bgr0bgr + srcHi0 = _mm_or_si128(srcHi0, srcHi2); // 0bgr0bgr0bgr0bgr + + // unpack into rows that get the tiling order correct + __m128i vRow00 = _mm_unpacklo_epi64(srcLo0, srcHi0); // 0bgr0bgr0bgr0bgr0bgr0bgr0bgr0bgr + __m128i vRow10 = _mm_unpackhi_epi64(srcLo0, srcHi0); + + __m256i final = _mm256_castsi128_si256(vRow00); + final = _mm256_insertf128_si256(final, vRow10, 1); + +#elif KNOB_ARCH == KNOB_ARCH_AVX2 + + // logic is as above, only wider + src1 = _mm256_slli_si256(src1, 1); + src2 = _mm256_slli_si256(src2, 2); + + src0 = _mm256_or_si256(src0, src1); + + __m256i final = _mm256_or_si256(src0, src2); + + // adjust the data to get the tiling order correct 0 1 2 3 -> 0 2 1 3 + final = _mm256_permute4x64_epi64(final, 0xD8); + +#endif + + _mm256_storeu2_m128i((__m128i*)pDst1, (__m128i*)pDst, final); +} + +template<> +struct ConvertPixelsSOAtoAOS +{ + template + INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests]) + { + FlatConvert(pSrc, ppDsts[0], ppDsts[1]); + } +}; + +template<> +struct ConvertPixelsSOAtoAOS +{ + template + INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests]) + { + FlatConvertNoAlpha(pSrc, ppDsts[0], ppDsts[1]); + } +}; + +template<> +struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, B8G8R8A8_UNORM_SRGB > +{ + template + INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests]) + { + FlatConvert(pSrc, ppDsts[0], ppDsts[1]); + } +}; + +template<> +struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, B8G8R8X8_UNORM_SRGB > +{ + template + INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests]) + { + FlatConvertNoAlpha(pSrc, ppDsts[0], ppDsts[1]); + } +}; + +template<> +struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, R8G8B8A8_UNORM > +{ + template + INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests]) + { + FlatConvert(pSrc, ppDsts[0], ppDsts[1]); + } +}; + +template<> +struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, R8G8B8X8_UNORM > +{ + template + INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests]) + { + FlatConvertNoAlpha(pSrc, ppDsts[0], ppDsts[1]); + } +}; + +template<> +struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, R8G8B8A8_UNORM_SRGB > +{ + template + INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests]) + { + FlatConvert(pSrc, ppDsts[0], ppDsts[1]); + } +}; + +template<> +struct ConvertPixelsSOAtoAOS < R32G32B32A32_FLOAT, R8G8B8X8_UNORM_SRGB > +{ + template + INLINE static void Convert(const uint8_t* pSrc, uint8_t* (&ppDsts)[NumDests]) + { + FlatConvertNoAlpha(pSrc, ppDsts[0], ppDsts[1]); + } +}; + +////////////////////////////////////////////////////////////////////////// +/// StoreRasterTile +////////////////////////////////////////////////////////////////////////// +template +struct StoreRasterTile +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Retrieve color from hot tile source which is always float. + /// @param pSrc - Pointer to raster tile. + /// @param x, y - Coordinates to raster tile. + /// @param output - output color + INLINE static void GetSwizzledSrcColor( + uint8_t* pSrc, + uint32_t x, uint32_t y, + float outputColor[4]) + { + typedef SimdTile SimdT; + + SimdT* pSrcSimdTiles = (SimdT*)pSrc; + + // Compute which simd tile we're accessing within 8x8 tile. + // i.e. Compute linear simd tile coordinate given (x, y) in pixel coordinates. + uint32_t simdIndex = (y / SIMD_TILE_Y_DIM) * (KNOB_TILE_X_DIM / SIMD_TILE_X_DIM) + (x / SIMD_TILE_X_DIM); + + SimdT* pSimdTile = &pSrcSimdTiles[simdIndex]; + + uint32_t simdOffset = (y % SIMD_TILE_Y_DIM) * SIMD_TILE_X_DIM + (x % SIMD_TILE_X_DIM); + + pSimdTile->GetSwizzledColor(simdOffset, outputColor); + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores an 8x8 raster tile to the destination surface. + /// @param pSrc - Pointer to raster tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to raster tile. + INLINE static void Store( + uint8_t *pSrc, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) // (x, y) pixel coordinate to start of raster tile. + { + uint32_t lodWidth = std::max(pDstSurface->width >> pDstSurface->lod, 1U); + uint32_t lodHeight = std::max(pDstSurface->height >> pDstSurface->lod, 1U); + + // For each raster tile pixel (rx, ry) + for (uint32_t ry = 0; ry < KNOB_TILE_Y_DIM; ++ry) + { + for (uint32_t rx = 0; rx < KNOB_TILE_X_DIM; ++rx) + { + // Perform bounds checking. + if (((x + rx) < lodWidth) && + ((y + ry) < lodHeight)) + { + float srcColor[4]; + GetSwizzledSrcColor(pSrc, rx, ry, srcColor); + + uint8_t *pDst = (uint8_t*)ComputeSurfaceAddress((x + rx), (y + ry), + pDstSurface->arrayIndex + renderTargetArrayIndex, pDstSurface->arrayIndex + renderTargetArrayIndex, + sampleNum, pDstSurface->lod, pDstSurface); + ConvertPixelFromFloat(pDst, srcColor); + } + } + } + } +}; + +template +struct OptStoreRasterTile : StoreRasterTile +{}; + +////////////////////////////////////////////////////////////////////////// +/// OptStoreRasterTile - SWR_TILE_MODE_NONE specialization for 8bpp +////////////////////////////////////////////////////////////////////////// +template +struct OptStoreRasterTile< TilingTraits, SrcFormat, DstFormat > +{ + typedef StoreRasterTile, SrcFormat, DstFormat> GenericStoreTile; + static const size_t SRC_BYTES_PER_PIXEL = FormatTraits::bpp / 8; + static const size_t DST_BYTES_PER_PIXEL = FormatTraits::bpp / 8; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores an 8x8 raster tile to the destination surface. + /// @param pSrc - Pointer to raster tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to raster tile. + INLINE static void Store( + uint8_t *pSrc, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) + { + // Punt non-full tiles to generic store + uint32_t lodWidth = std::max(pDstSurface->width >> pDstSurface->lod, 1U); + uint32_t lodHeight = std::max(pDstSurface->height >> pDstSurface->lod, 1U); + if (x + KNOB_TILE_X_DIM > lodWidth || + y + KNOB_TILE_Y_DIM > lodHeight) + { + return GenericStoreTile::Store(pSrc, pDstSurface, x, y, sampleNum, renderTargetArrayIndex); + } + + uint8_t* pDst = (uint8_t*)ComputeSurfaceAddress(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex, + pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface); + uint8_t* ppRows[] = { pDst, pDst + pDstSurface->pitch }; + + for (uint32_t row = 0; row < KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM; ++row) + { + uint8_t* ppStartRows[] = { ppRows[0], ppRows[1] }; + + for (uint32_t col = 0; col < KNOB_TILE_X_DIM / SIMD_TILE_X_DIM; ++col) + { + // Format conversion and convert from SOA to AOS, and store the rows. + ConvertPixelsSOAtoAOS::Convert(pSrc, ppRows); + + ppRows[0] += KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2; + ppRows[1] += KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2;; + pSrc += SRC_BYTES_PER_PIXEL * KNOB_SIMD_WIDTH; + } + + ppRows[0] = ppStartRows[0] + 2 * pDstSurface->pitch; + ppRows[1] = ppStartRows[1] + 2 * pDstSurface->pitch; + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// OptStoreRasterTile - SWR_TILE_MODE_NONE specialization for 16bpp +////////////////////////////////////////////////////////////////////////// +template +struct OptStoreRasterTile< TilingTraits, SrcFormat, DstFormat > +{ + typedef StoreRasterTile, SrcFormat, DstFormat> GenericStoreTile; + static const size_t SRC_BYTES_PER_PIXEL = FormatTraits::bpp / 8; + static const size_t DST_BYTES_PER_PIXEL = FormatTraits::bpp / 8; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores an 8x8 raster tile to the destination surface. + /// @param pSrc - Pointer to raster tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to raster tile. + INLINE static void Store( + uint8_t *pSrc, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) + { + // Punt non-full tiles to generic store + uint32_t lodWidth = std::max(pDstSurface->width >> pDstSurface->lod, 1U); + uint32_t lodHeight = std::max(pDstSurface->height >> pDstSurface->lod, 1U); + if (x + KNOB_TILE_X_DIM > lodWidth || + y + KNOB_TILE_Y_DIM > lodHeight) + { + return GenericStoreTile::Store(pSrc, pDstSurface, x, y, sampleNum, renderTargetArrayIndex); + } + + uint8_t* pDst = (uint8_t*)ComputeSurfaceAddress(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex, + pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface); + uint8_t* ppRows[] = { pDst, pDst + pDstSurface->pitch }; + + for (uint32_t row = 0; row < KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM; ++row) + { + uint8_t* ppStartRows[] = { ppRows[0], ppRows[1] }; + + for (uint32_t col = 0; col < KNOB_TILE_X_DIM / SIMD_TILE_X_DIM; ++col) + { + // Format conversion and convert from SOA to AOS, and store the rows. + ConvertPixelsSOAtoAOS::Convert(pSrc, ppRows); + + ppRows[0] += KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2; + ppRows[1] += KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2;; + pSrc += SRC_BYTES_PER_PIXEL * KNOB_SIMD_WIDTH; + } + + ppRows[0] = ppStartRows[0] + 2 * pDstSurface->pitch; + ppRows[1] = ppStartRows[1] + 2 * pDstSurface->pitch; + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// OptStoreRasterTile - SWR_TILE_MODE_NONE specialization for 32bpp +////////////////////////////////////////////////////////////////////////// +template +struct OptStoreRasterTile< TilingTraits, SrcFormat, DstFormat > +{ + typedef StoreRasterTile, SrcFormat, DstFormat> GenericStoreTile; + static const size_t SRC_BYTES_PER_PIXEL = FormatTraits::bpp / 8; + static const size_t DST_BYTES_PER_PIXEL = FormatTraits::bpp / 8; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores an 8x8 raster tile to the destination surface. + /// @param pSrc - Pointer to raster tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to raster tile. + INLINE static void Store( + uint8_t *pSrc, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) + { + // Punt non-full tiles to generic store + uint32_t lodWidth = std::max(pDstSurface->width >> pDstSurface->lod, 1U); + uint32_t lodHeight = std::max(pDstSurface->height >> pDstSurface->lod, 1U); + if (x + KNOB_TILE_X_DIM > lodWidth || + y + KNOB_TILE_Y_DIM > lodHeight) + { + return GenericStoreTile::Store(pSrc, pDstSurface, x, y, sampleNum, renderTargetArrayIndex); + } + + uint8_t* pDst = (uint8_t*)ComputeSurfaceAddress(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex, + pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface); + uint8_t* ppRows[] = { pDst, pDst + pDstSurface->pitch }; + + for (uint32_t row = 0; row < KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM; ++row) + { + uint8_t* ppStartRows[] = { ppRows[0], ppRows[1] }; + + for (uint32_t col = 0; col < KNOB_TILE_X_DIM / SIMD_TILE_X_DIM; ++col) + { + // Format conversion and convert from SOA to AOS, and store the rows. + ConvertPixelsSOAtoAOS::Convert(pSrc, ppRows); + + ppRows[0] += KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2; + ppRows[1] += KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2;; + pSrc += SRC_BYTES_PER_PIXEL * KNOB_SIMD_WIDTH; + } + + ppRows[0] = ppStartRows[0] + 2 * pDstSurface->pitch; + ppRows[1] = ppStartRows[1] + 2 * pDstSurface->pitch; + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// OptStoreRasterTile - SWR_TILE_MODE_NONE specialization for 64bpp +////////////////////////////////////////////////////////////////////////// +template +struct OptStoreRasterTile< TilingTraits, SrcFormat, DstFormat > +{ + typedef StoreRasterTile, SrcFormat, DstFormat> GenericStoreTile; + static const size_t DST_BYTES_PER_PIXEL = FormatTraits::bpp / 8; + static const size_t SRC_BYTES_PER_PIXEL = FormatTraits::bpp / 8; + static const size_t MAX_DST_COLUMN_BYTES = 16; + static const size_t SRC_COLUMN_BYTES = KNOB_SIMD_WIDTH * SRC_BYTES_PER_PIXEL; + static const size_t DST_COLUMN_BYTES_PER_SRC = KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores an 8x8 raster tile to the destination surface. + /// @param pSrc - Pointer to raster tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to raster tile. + INLINE static void Store( + uint8_t *pSrc, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) + { + // Punt non-full tiles to generic store + uint32_t lodWidth = std::max(pDstSurface->width >> pDstSurface->lod, 1U); + uint32_t lodHeight = std::max(pDstSurface->height >> pDstSurface->lod, 1U); + if (x + KNOB_TILE_X_DIM > lodWidth || + y + KNOB_TILE_Y_DIM > lodHeight) + { + return GenericStoreTile::Store(pSrc, pDstSurface, x, y, sampleNum, renderTargetArrayIndex); + } + + uint8_t* pDst = (uint8_t*)ComputeSurfaceAddress(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex, + pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface); + uint8_t* ppDsts[] = + { + pDst, // row 0, col 0 + pDst + pDstSurface->pitch, // row 1, col 0 + pDst + MAX_DST_COLUMN_BYTES, // row 0, col 1 + pDst + pDstSurface->pitch + MAX_DST_COLUMN_BYTES, // row 1, col 1 + }; + + for (uint32_t row = 0; row < KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM; ++row) + { + uint8_t* ppStartRows[] = + { + ppDsts[0], + ppDsts[1], + ppDsts[2], + ppDsts[3], + }; + + for (uint32_t col = 0; col < KNOB_TILE_X_DIM / SIMD_TILE_X_DIM; ++col) + { + // Format conversion and convert from SOA to AOS, and store the rows. + ConvertPixelsSOAtoAOS::Convert(pSrc, ppDsts); + + ppDsts[0] += DST_COLUMN_BYTES_PER_SRC; + ppDsts[1] += DST_COLUMN_BYTES_PER_SRC; + ppDsts[2] += DST_COLUMN_BYTES_PER_SRC; + ppDsts[3] += DST_COLUMN_BYTES_PER_SRC; + pSrc += SRC_COLUMN_BYTES; + } + + ppDsts[0] = ppStartRows[0] + 2 * pDstSurface->pitch; + ppDsts[1] = ppStartRows[1] + 2 * pDstSurface->pitch; + ppDsts[2] = ppStartRows[2] + 2 * pDstSurface->pitch; + ppDsts[3] = ppStartRows[3] + 2 * pDstSurface->pitch; + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// OptStoreRasterTile - SWR_TILE_MODE_NONE specialization for 128bpp +////////////////////////////////////////////////////////////////////////// +template +struct OptStoreRasterTile< TilingTraits, SrcFormat, DstFormat > +{ + typedef StoreRasterTile, SrcFormat, DstFormat> GenericStoreTile; + static const size_t DST_BYTES_PER_PIXEL = FormatTraits::bpp / 8; + static const size_t SRC_BYTES_PER_PIXEL = FormatTraits::bpp / 8; + static const size_t MAX_DST_COLUMN_BYTES = 16; + static const size_t SRC_COLUMN_BYTES = KNOB_SIMD_WIDTH * SRC_BYTES_PER_PIXEL; + static const size_t DST_COLUMN_BYTES_PER_SRC = KNOB_SIMD_WIDTH * DST_BYTES_PER_PIXEL / 2; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores an 8x8 raster tile to the destination surface. + /// @param pSrc - Pointer to raster tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to raster tile. + INLINE static void Store( + uint8_t *pSrc, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) + { + // Punt non-full tiles to generic store + uint32_t lodWidth = std::max(pDstSurface->width >> pDstSurface->lod, 1U); + uint32_t lodHeight = std::max(pDstSurface->height >> pDstSurface->lod, 1U); + if (x + KNOB_TILE_X_DIM > lodWidth || + y + KNOB_TILE_Y_DIM > lodHeight) + { + return GenericStoreTile::Store(pSrc, pDstSurface, x, y, sampleNum, renderTargetArrayIndex); + } + + uint8_t* pDst = (uint8_t*)ComputeSurfaceAddress(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex, + pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface); + struct DstPtrs + { + uint8_t* ppDsts[8]; + } ptrs; + + // Need 8 pointers, 4 columns of 2 rows each + for (uint32_t y = 0; y < 2; ++y) + { + for (uint32_t x = 0; x < 4; ++x) + { + ptrs.ppDsts[x * 2 + y] = pDst + y * pDstSurface->pitch + x * MAX_DST_COLUMN_BYTES; + } + } + + for (uint32_t row = 0; row < KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM; ++row) + { + DstPtrs startPtrs = ptrs; + + for (uint32_t col = 0; col < KNOB_TILE_X_DIM / SIMD_TILE_X_DIM; ++col) + { + // Format conversion and convert from SOA to AOS, and store the rows. + ConvertPixelsSOAtoAOS::Convert(pSrc, ptrs.ppDsts); + + ptrs.ppDsts[0] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[1] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[2] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[3] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[4] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[5] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[6] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[7] += DST_COLUMN_BYTES_PER_SRC; + pSrc += SRC_COLUMN_BYTES; + } + + ptrs.ppDsts[0] = startPtrs.ppDsts[0] + 2 * pDstSurface->pitch; + ptrs.ppDsts[1] = startPtrs.ppDsts[1] + 2 * pDstSurface->pitch; + ptrs.ppDsts[2] = startPtrs.ppDsts[2] + 2 * pDstSurface->pitch; + ptrs.ppDsts[3] = startPtrs.ppDsts[3] + 2 * pDstSurface->pitch; + ptrs.ppDsts[4] = startPtrs.ppDsts[4] + 2 * pDstSurface->pitch; + ptrs.ppDsts[5] = startPtrs.ppDsts[5] + 2 * pDstSurface->pitch; + ptrs.ppDsts[6] = startPtrs.ppDsts[6] + 2 * pDstSurface->pitch; + ptrs.ppDsts[7] = startPtrs.ppDsts[7] + 2 * pDstSurface->pitch; + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// OptStoreRasterTile - TILE_MODE_YMAJOR specialization for 8bpp +////////////////////////////////////////////////////////////////////////// +template +struct OptStoreRasterTile< TilingTraits, SrcFormat, DstFormat > +{ + typedef StoreRasterTile, SrcFormat, DstFormat> GenericStoreTile; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores an 8x8 raster tile to the destination surface. + /// @param pSrc - Pointer to raster tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to raster tile. + INLINE static void Store( + uint8_t *pSrc, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) + { + static const uint32_t DestRowWidthBytes = 16; // 16B rows + + // Punt non-full tiles to generic store + uint32_t lodWidth = std::max(pDstSurface->width >> pDstSurface->lod, 1U); + uint32_t lodHeight = std::max(pDstSurface->height >> pDstSurface->lod, 1U); + if (x + KNOB_TILE_X_DIM > lodWidth || + y + KNOB_TILE_Y_DIM > lodHeight) + { + return GenericStoreTile::Store(pSrc, pDstSurface, x, y, sampleNum, renderTargetArrayIndex); + } + + // TileY is a column-major tiling mode where each 4KB tile consist of 8 columns of 32 x 16B rows. + // We can compute the offsets to each column within the raster tile once and increment from these. + // There will be 2 x 4-wide columns in an 8x8 raster tile. + uint8_t* pCol0 = (uint8_t*)ComputeSurfaceAddress(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex, + pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface); + + // Increment by a whole SIMD. 4x2 for AVX. 2x2 for SSE. + uint32_t pSrcInc = (FormatTraits::bpp * KNOB_SIMD_WIDTH) / 8; + + // The Hot Tile uses a row-major tiling mode and has a larger memory footprint. So we iterate in a row-major pattern. + for (uint32_t row = 0; row < KNOB_TILE_Y_DIM; row += SIMD_TILE_Y_DIM) + { + uint32_t rowOffset = row * DestRowWidthBytes; + + uint8_t* pRow = pCol0 + rowOffset; + uint8_t* ppDsts[] = { pRow, pRow + DestRowWidthBytes }; + + ConvertPixelsSOAtoAOS::Convert(pSrc, ppDsts); + pSrc += pSrcInc; + + ppDsts[0] += DestRowWidthBytes / 4; + ppDsts[1] += DestRowWidthBytes / 4; + + ConvertPixelsSOAtoAOS::Convert(pSrc, ppDsts); + pSrc += pSrcInc; + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// OptStoreRasterTile - TILE_MODE_YMAJOR specialization for 16bpp +////////////////////////////////////////////////////////////////////////// +template +struct OptStoreRasterTile< TilingTraits, SrcFormat, DstFormat > +{ + typedef StoreRasterTile, SrcFormat, DstFormat> GenericStoreTile; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores an 8x8 raster tile to the destination surface. + /// @param pSrc - Pointer to raster tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to raster tile. + INLINE static void Store( + uint8_t *pSrc, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) + { + static const uint32_t DestRowWidthBytes = 16; // 16B rows + + // Punt non-full tiles to generic store + uint32_t lodWidth = std::max(pDstSurface->width >> pDstSurface->lod, 1U); + uint32_t lodHeight = std::max(pDstSurface->height >> pDstSurface->lod, 1U); + if (x + KNOB_TILE_X_DIM > lodWidth || + y + KNOB_TILE_Y_DIM > lodHeight) + { + return GenericStoreTile::Store(pSrc, pDstSurface, x, y, sampleNum, renderTargetArrayIndex); + } + + // TileY is a column-major tiling mode where each 4KB tile consist of 8 columns of 32 x 16B rows. + // We can compute the offsets to each column within the raster tile once and increment from these. + // There will be 2 x 4-wide columns in an 8x8 raster tile. + uint8_t* pCol0 = (uint8_t*)ComputeSurfaceAddress(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex, + pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface); + + // Increment by a whole SIMD. 4x2 for AVX. 2x2 for SSE. + uint32_t pSrcInc = (FormatTraits::bpp * KNOB_SIMD_WIDTH) / 8; + + // The Hot Tile uses a row-major tiling mode and has a larger memory footprint. So we iterate in a row-major pattern. + for (uint32_t row = 0; row < KNOB_TILE_Y_DIM; row += SIMD_TILE_Y_DIM) + { + uint32_t rowOffset = row * DestRowWidthBytes; + + uint8_t* pRow = pCol0 + rowOffset; + uint8_t* ppDsts[] = { pRow, pRow + DestRowWidthBytes }; + + ConvertPixelsSOAtoAOS::Convert(pSrc, ppDsts); + pSrc += pSrcInc; + + ppDsts[0] += DestRowWidthBytes / 2; + ppDsts[1] += DestRowWidthBytes / 2; + + ConvertPixelsSOAtoAOS::Convert(pSrc, ppDsts); + pSrc += pSrcInc; + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// OptStoreRasterTile - TILE_MODE_XMAJOR specialization for 32bpp +////////////////////////////////////////////////////////////////////////// +template +struct OptStoreRasterTile< TilingTraits, SrcFormat, DstFormat > +{ + typedef StoreRasterTile, SrcFormat, DstFormat> GenericStoreTile; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores an 8x8 raster tile to the destination surface. + /// @param pSrc - Pointer to raster tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to raster tile. + INLINE static void Store( + uint8_t *pSrc, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) + { + static const uint32_t DestRowWidthBytes = 512; // 512B rows + + // Punt non-full tiles to generic store + uint32_t lodWidth = std::max(pDstSurface->width >> pDstSurface->lod, 1U); + uint32_t lodHeight = std::max(pDstSurface->height >> pDstSurface->lod, 1U); + if (x + KNOB_TILE_X_DIM > lodWidth || + y + KNOB_TILE_Y_DIM > lodHeight) + { + return GenericStoreTile::Store(pSrc, pDstSurface, x, y, sampleNum, renderTargetArrayIndex); + } + + // TileX is a row-major tiling mode where each 4KB tile consist of 8 x 512B rows. + // We can compute the offsets to each column within the raster tile once and increment from these. + uint8_t *pRow0 = (uint8_t*)ComputeSurfaceAddress(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex, + pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface); + uint8_t* pRow1 = pRow0 + DestRowWidthBytes; + + for (uint32_t row = 0; row < KNOB_TILE_Y_DIM; row += SIMD_TILE_Y_DIM) + { + for (uint32_t col = 0; col < KNOB_TILE_X_DIM; col += SIMD_TILE_X_DIM) + { + uint32_t xRowOffset = col * (FormatTraits::bpp / 8); + + uint8_t* ppDsts[] = { pRow0 + xRowOffset, pRow1 + xRowOffset }; + ConvertPixelsSOAtoAOS::Convert(pSrc, ppDsts); + + // Increment by a whole SIMD. 4x2 for AVX. 2x2 for SSE. + pSrc += (FormatTraits::bpp * KNOB_SIMD_WIDTH) / 8; + } + + pRow0 += (DestRowWidthBytes * 2); + pRow1 += (DestRowWidthBytes * 2); + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// OptStoreRasterTile - TILE_MODE_YMAJOR specialization for 32bpp +////////////////////////////////////////////////////////////////////////// +template +struct OptStoreRasterTile< TilingTraits, SrcFormat, DstFormat > +{ + typedef StoreRasterTile, SrcFormat, DstFormat> GenericStoreTile; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores an 8x8 raster tile to the destination surface. + /// @param pSrc - Pointer to raster tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to raster tile. + INLINE static void Store( + uint8_t *pSrc, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) + { + static const uint32_t DestRowWidthBytes = 16; // 16B rows + static const uint32_t DestColumnBytes = DestRowWidthBytes * 32; // 16B x 32 rows. + + // Punt non-full tiles to generic store + uint32_t lodWidth = std::max(pDstSurface->width >> pDstSurface->lod, 1U); + uint32_t lodHeight = std::max(pDstSurface->height >> pDstSurface->lod, 1U); + if (x + KNOB_TILE_X_DIM > lodWidth || + y + KNOB_TILE_Y_DIM > lodHeight) + { + return GenericStoreTile::Store(pSrc, pDstSurface, x, y, sampleNum, renderTargetArrayIndex); + } + + // TileY is a column-major tiling mode where each 4KB tile consist of 8 columns of 32 x 16B rows. + // We can compute the offsets to each column within the raster tile once and increment from these. + // There will be 2 x 4-wide columns in an 8x8 raster tile. + uint8_t* pCol0 = (uint8_t*)ComputeSurfaceAddress(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex, + pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface); + + // Increment by a whole SIMD. 4x2 for AVX. 2x2 for SSE. + uint32_t pSrcInc = (FormatTraits::bpp * KNOB_SIMD_WIDTH) / 8; + + // The Hot Tile uses a row-major tiling mode and has a larger memory footprint. So we iterate in a row-major pattern. + for (uint32_t row = 0; row < KNOB_TILE_Y_DIM; row += SIMD_TILE_Y_DIM) + { + uint32_t rowOffset = row * DestRowWidthBytes; + + uint8_t* pRow = pCol0 + rowOffset; + uint8_t* ppDsts[] = { pRow, pRow + DestRowWidthBytes }; + + ConvertPixelsSOAtoAOS::Convert(pSrc, ppDsts); + pSrc += pSrcInc; + + ppDsts[0] += DestColumnBytes; + ppDsts[1] += DestColumnBytes; + + ConvertPixelsSOAtoAOS::Convert(pSrc, ppDsts); + pSrc += pSrcInc; + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// OptStoreRasterTile - TILE_MODE_YMAJOR specialization for 64bpp +////////////////////////////////////////////////////////////////////////// +template +struct OptStoreRasterTile< TilingTraits, SrcFormat, DstFormat > +{ + typedef StoreRasterTile, SrcFormat, DstFormat> GenericStoreTile; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores an 8x8 raster tile to the destination surface. + /// @param pSrc - Pointer to raster tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to raster tile. + INLINE static void Store( + uint8_t *pSrc, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) + { + static const uint32_t DestRowWidthBytes = 16; // 16B rows + static const uint32_t DestColumnBytes = DestRowWidthBytes * 32; // 16B x 32 rows. + + // Punt non-full tiles to generic store + uint32_t lodWidth = std::max(pDstSurface->width >> pDstSurface->lod, 1U); + uint32_t lodHeight = std::max(pDstSurface->height >> pDstSurface->lod, 1U); + if (x + KNOB_TILE_X_DIM > lodWidth || + y + KNOB_TILE_Y_DIM > lodHeight) + { + return GenericStoreTile::Store(pSrc, pDstSurface, x, y, sampleNum, renderTargetArrayIndex); + } + + // TileY is a column-major tiling mode where each 4KB tile consist of 8 columns of 32 x 16B rows. + // We can compute the offsets to each column within the raster tile once and increment from these. + // There will be 2 x 4-wide columns in an 8x8 raster tile. + uint8_t* pCol0 = (uint8_t*)ComputeSurfaceAddress(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex, + pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface); + uint8_t* pCol1 = pCol0 + DestColumnBytes; + + // There are 4 columns, each 2 pixels wide when we have 64bpp pixels. + // Increment by a whole SIMD. 4x2 for AVX. 2x2 for SSE. + uint32_t pSrcInc = (FormatTraits::bpp * KNOB_SIMD_WIDTH) / 8; + + // The Hot Tile uses a row-major tiling mode and has a larger memory footprint. So we iterate in a row-major pattern. + for (uint32_t row = 0; row < KNOB_TILE_Y_DIM; row += SIMD_TILE_Y_DIM) + { + uint32_t rowOffset = row * DestRowWidthBytes; + uint8_t* ppDsts[] = + { + pCol0 + rowOffset, + pCol0 + rowOffset + DestRowWidthBytes, + pCol1 + rowOffset, + pCol1 + rowOffset + DestRowWidthBytes, + }; + + ConvertPixelsSOAtoAOS::Convert(pSrc, ppDsts); + pSrc += pSrcInc; + + ppDsts[0] += DestColumnBytes * 2; + ppDsts[1] += DestColumnBytes * 2; + ppDsts[2] += DestColumnBytes * 2; + ppDsts[3] += DestColumnBytes * 2; + + ConvertPixelsSOAtoAOS::Convert(pSrc, ppDsts); + pSrc += pSrcInc; + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// OptStoreRasterTile - SWR_TILE_MODE_YMAJOR specialization for 128bpp +////////////////////////////////////////////////////////////////////////// +template +struct OptStoreRasterTile< TilingTraits, SrcFormat, DstFormat > +{ + typedef StoreRasterTile, SrcFormat, DstFormat> GenericStoreTile; + + static const size_t TILE_Y_COL_WIDTH_BYTES = 16; + static const size_t TILE_Y_ROWS = 32; + static const size_t TILE_Y_COL_BYTES = TILE_Y_ROWS * TILE_Y_COL_WIDTH_BYTES; + + static const size_t DST_BYTES_PER_PIXEL = FormatTraits::bpp / 8; + static const size_t SRC_BYTES_PER_PIXEL = FormatTraits::bpp / 8; + static const size_t MAX_DST_COLUMN_BYTES = 16; + + static const size_t SRC_COLUMN_BYTES = KNOB_SIMD_WIDTH * SRC_BYTES_PER_PIXEL; + static const size_t DST_COLUMN_BYTES_PER_SRC = TILE_Y_COL_BYTES * 4; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores an 8x8 raster tile to the destination surface. + /// @param pSrc - Pointer to raster tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to raster tile. + INLINE static void Store( + uint8_t *pSrc, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t sampleNum, uint32_t renderTargetArrayIndex) + { + // Punt non-full tiles to generic store + uint32_t lodWidth = std::max(pDstSurface->width >> pDstSurface->lod, 1U); + uint32_t lodHeight = std::max(pDstSurface->height >> pDstSurface->lod, 1U); + if (x + KNOB_TILE_X_DIM > lodWidth || + y + KNOB_TILE_Y_DIM > lodHeight) + { + return GenericStoreTile::Store(pSrc, pDstSurface, x, y, sampleNum, renderTargetArrayIndex); + } + + uint8_t* pDst = (uint8_t*)ComputeSurfaceAddress(x, y, pDstSurface->arrayIndex + renderTargetArrayIndex, + pDstSurface->arrayIndex + renderTargetArrayIndex, sampleNum, pDstSurface->lod, pDstSurface); + struct DstPtrs + { + uint8_t* ppDsts[8]; + } ptrs; + + // Need 8 pointers, 4 columns of 2 rows each + for (uint32_t y = 0; y < 2; ++y) + { + for (uint32_t x = 0; x < 4; ++x) + { + ptrs.ppDsts[x * 2 + y] = pDst + y * TILE_Y_COL_WIDTH_BYTES + x * TILE_Y_COL_BYTES; + } + } + + for (uint32_t row = 0; row < KNOB_TILE_Y_DIM / SIMD_TILE_Y_DIM; ++row) + { + DstPtrs startPtrs = ptrs; + + for (uint32_t col = 0; col < KNOB_TILE_X_DIM / SIMD_TILE_X_DIM; ++col) + { + // Format conversion and convert from SOA to AOS, and store the rows. + ConvertPixelsSOAtoAOS::Convert(pSrc, ptrs.ppDsts); + + ptrs.ppDsts[0] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[1] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[2] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[3] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[4] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[5] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[6] += DST_COLUMN_BYTES_PER_SRC; + ptrs.ppDsts[7] += DST_COLUMN_BYTES_PER_SRC; + pSrc += SRC_COLUMN_BYTES; + } + + ptrs.ppDsts[0] = startPtrs.ppDsts[0] + 2 * TILE_Y_COL_WIDTH_BYTES; + ptrs.ppDsts[1] = startPtrs.ppDsts[1] + 2 * TILE_Y_COL_WIDTH_BYTES; + ptrs.ppDsts[2] = startPtrs.ppDsts[2] + 2 * TILE_Y_COL_WIDTH_BYTES; + ptrs.ppDsts[3] = startPtrs.ppDsts[3] + 2 * TILE_Y_COL_WIDTH_BYTES; + ptrs.ppDsts[4] = startPtrs.ppDsts[4] + 2 * TILE_Y_COL_WIDTH_BYTES; + ptrs.ppDsts[5] = startPtrs.ppDsts[5] + 2 * TILE_Y_COL_WIDTH_BYTES; + ptrs.ppDsts[6] = startPtrs.ppDsts[6] + 2 * TILE_Y_COL_WIDTH_BYTES; + ptrs.ppDsts[7] = startPtrs.ppDsts[7] + 2 * TILE_Y_COL_WIDTH_BYTES; + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// StoreMacroTile - Stores a macro tile which consists of raster tiles. +////////////////////////////////////////////////////////////////////////// +template +struct StoreMacroTile +{ + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores a macrotile to the destination surface using safe implementation. + /// @param pSrc - Pointer to macro tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to macro tile + static void StoreGeneric( + uint8_t *pSrcHotTile, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t renderTargetArrayIndex) + { + // Store each raster tile from the hot tile to the destination surface. + for(uint32_t row = 0; row < KNOB_MACROTILE_Y_DIM; row += KNOB_TILE_Y_DIM) + { + for(uint32_t col = 0; col < KNOB_MACROTILE_X_DIM; col += KNOB_TILE_X_DIM) + { + for(uint32_t sampleNum = 0; sampleNum < pDstSurface->numSamples; sampleNum++) + { + StoreRasterTile::Store (pSrcHotTile, pDstSurface, (x + col), (y + row), sampleNum, + renderTargetArrayIndex); + pSrcHotTile += KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * (FormatTraits::bpp / 8); + } + } + } + } + + typedef void(*PFN_STORE_TILES_INTERNAL)(uint8_t*, SWR_SURFACE_STATE*, uint32_t, uint32_t, uint32_t, uint32_t); + ////////////////////////////////////////////////////////////////////////// + /// @brief Stores a macrotile to the destination surface. + /// @param pSrc - Pointer to macro tile. + /// @param pDstSurface - Destination surface state + /// @param x, y - Coordinates to macro tile + static void Store( + uint8_t *pSrcHotTile, + SWR_SURFACE_STATE* pDstSurface, + uint32_t x, uint32_t y, uint32_t renderTargetArrayIndex) + { + PFN_STORE_TILES_INTERNAL pfnStore[SWR_MAX_NUM_MULTISAMPLES]; + for(uint32_t sampleNum = 0; sampleNum < pDstSurface->numSamples; sampleNum++) + { + size_t dstSurfAddress = (size_t)ComputeSurfaceAddress( + 0, + 0, + pDstSurface->arrayIndex + renderTargetArrayIndex, // z for 3D surfaces + pDstSurface->arrayIndex + renderTargetArrayIndex, // array index for 2D arrays + sampleNum, + pDstSurface->lod, + pDstSurface); + + // Only support generic store-tile if lod surface doesn't start on a page boundary and is non-linear + bool bForceGeneric = ((pDstSurface->tileMode != SWR_TILE_NONE) && (0 != (dstSurfAddress & 0xfff))) || (pDstSurface->bInterleavedSamples); + + pfnStore[sampleNum] = (bForceGeneric || KNOB_USE_GENERIC_STORETILE) ? StoreRasterTile::Store : OptStoreRasterTile::Store; + } + + // Store each raster tile from the hot tile to the destination surface. + for(uint32_t row = 0; row < KNOB_MACROTILE_Y_DIM; row += KNOB_TILE_Y_DIM) + { + for(uint32_t col = 0; col < KNOB_MACROTILE_X_DIM; col += KNOB_TILE_X_DIM) + { + for(uint32_t sampleNum = 0; sampleNum < pDstSurface->numSamples; sampleNum++) + { + pfnStore[sampleNum](pSrcHotTile, pDstSurface, (x + col), (y + row), sampleNum, renderTargetArrayIndex); + pSrcHotTile += KNOB_TILE_X_DIM * KNOB_TILE_Y_DIM * (FormatTraits::bpp / 8); + } + } + } + } +}; + +static void BUCKETS_START(UINT id) +{ +#ifdef KNOB_ENABLE_RDTSC + gBucketMgr.StartBucket(id); +#endif +} + +static void BUCKETS_STOP(UINT id) +{ +#ifdef KNOB_ENABLE_RDTSC + gBucketMgr.StopBucket(id); +#endif +} + +// on demand buckets for store tiles +static std::mutex sBucketMutex; +static std::vector sBuckets(NUM_SWR_FORMATS, -1); + +////////////////////////////////////////////////////////////////////////// +/// @brief Deswizzles and stores a full hottile to a render surface +/// @param hPrivateContext - Handle to private DC +/// @param srcFormat - Format for hot tile. +/// @param renderTargetIndex - Index to destination render target +/// @param x, y - Coordinates to raster tile. +/// @param pSrcHotTile - Pointer to Hot Tile +void StoreHotTile( + SWR_SURFACE_STATE *pDstSurface, + SWR_FORMAT srcFormat, + SWR_RENDERTARGET_ATTACHMENT renderTargetIndex, + uint32_t x, uint32_t y, uint32_t renderTargetArrayIndex, + uint8_t *pSrcHotTile) +{ + if (pDstSurface->type == SURFACE_NULL) + { + return; + } + + // force 0 if requested renderTargetArrayIndex is OOB + if (renderTargetArrayIndex >= pDstSurface->depth) + { + renderTargetArrayIndex = 0; + } + + PFN_STORE_TILES pfnStoreTiles = nullptr; + + if ((renderTargetIndex <= SWR_ATTACHMENT_COLOR7) && (pDstSurface->tileMode != SWR_TILE_MODE_WMAJOR)) + { + pfnStoreTiles = sStoreTilesTableColor[pDstSurface->tileMode][pDstSurface->format]; + } + else if (renderTargetIndex == SWR_ATTACHMENT_DEPTH) + { + pfnStoreTiles = sStoreTilesTableDepth[pDstSurface->tileMode][pDstSurface->format]; + } + else + { + pfnStoreTiles = sStoreTilesTableStencil[pDstSurface->tileMode][pDstSurface->format]; + } + + if(nullptr == pfnStoreTiles) + { + SWR_ASSERT(false, "Invalid pixel format / tile mode for store tiles"); + return; + } + + // Store a macro tile +#ifdef KNOB_ENABLE_RDTSC + if (sBuckets[pDstSurface->format] == -1) + { + // guard sBuckets update since storetiles is called by multiple threads + sBucketMutex.lock(); + if (sBuckets[pDstSurface->format] == -1) + { + const SWR_FORMAT_INFO& info = GetFormatInfo(pDstSurface->format); + BUCKET_DESC desc{info.name, "", false, 0xffffffff}; + sBuckets[pDstSurface->format] = gBucketMgr.RegisterBucket(desc); + } + sBucketMutex.unlock(); + } +#endif + + BUCKETS_START(sBuckets[pDstSurface->format]); + pfnStoreTiles(pSrcHotTile, pDstSurface, x, y, renderTargetArrayIndex); + BUCKETS_STOP(sBuckets[pDstSurface->format]); +} + +////////////////////////////////////////////////////////////////////////// +/// InitStoreTilesTable - Helper for setting up the tables. +template +void InitStoreTilesTableColor( + PFN_STORE_TILES (&table)[NumTileModesT][ArraySizeT]) +{ + table[TileModeT][R32G32B32A32_FLOAT] = StoreMacroTile, R32G32B32A32_FLOAT, R32G32B32A32_FLOAT>::Store; + table[TileModeT][R32G32B32A32_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, R32G32B32A32_SINT>::Store; + table[TileModeT][R32G32B32A32_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, R32G32B32A32_UINT>::Store; + table[TileModeT][R32G32B32X32_FLOAT] = StoreMacroTile, R32G32B32A32_FLOAT, R32G32B32X32_FLOAT>::Store; + table[TileModeT][R32G32B32_FLOAT] = StoreMacroTile, R32G32B32A32_FLOAT, R32G32B32_FLOAT>::Store; + table[TileModeT][R32G32B32_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, R32G32B32_SINT>::Store; + table[TileModeT][R32G32B32_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, R32G32B32_UINT>::Store; + table[TileModeT][R16G16B16A16_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16B16A16_UNORM>::Store; + table[TileModeT][R16G16B16A16_SNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16B16A16_SNORM>::Store; + table[TileModeT][R16G16B16A16_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16B16A16_SINT>::Store; + table[TileModeT][R16G16B16A16_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16B16A16_UINT>::Store; + table[TileModeT][R16G16B16A16_FLOAT] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16B16A16_FLOAT>::Store; + table[TileModeT][R32G32_FLOAT] = StoreMacroTile, R32G32B32A32_FLOAT, R32G32_FLOAT>::Store; + table[TileModeT][R32G32_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, R32G32_SINT>::Store; + table[TileModeT][R32G32_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, R32G32_UINT>::Store; + table[TileModeT][R16G16B16X16_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16B16X16_UNORM>::Store; + table[TileModeT][R16G16B16X16_FLOAT] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16B16X16_FLOAT>::Store; + table[TileModeT][B8G8R8A8_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, B8G8R8A8_UNORM>::Store; + table[TileModeT][B8G8R8A8_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, B8G8R8A8_UNORM_SRGB>::Store; + + // 101010_2, 565, 555_1, and 444_4 formats force generic store tile for now + table[TileModeT][R10G10B10A2_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R10G10B10A2_UNORM>::StoreGeneric; + table[TileModeT][R10G10B10A2_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, R10G10B10A2_UNORM_SRGB>::StoreGeneric; + table[TileModeT][R10G10B10A2_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, R10G10B10A2_UINT>::StoreGeneric; + + table[TileModeT][R8G8B8A8_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8B8A8_UNORM>::Store; + table[TileModeT][R8G8B8A8_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8B8A8_UNORM_SRGB>::Store; + table[TileModeT][R8G8B8A8_SNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8B8A8_SNORM>::Store; + table[TileModeT][R8G8B8A8_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8B8A8_SINT>::Store; + table[TileModeT][R8G8B8A8_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8B8A8_UINT>::Store; + table[TileModeT][R16G16_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16_UNORM>::Store; + table[TileModeT][R16G16_SNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16_SNORM>::Store; + table[TileModeT][R16G16_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16_SINT>::Store; + table[TileModeT][R16G16_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16_UINT>::Store; + table[TileModeT][R16G16_FLOAT] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16_FLOAT>::Store; + + // 101010_2, 565, 555_1, and 444_4 formats force generic store tile for now + table[TileModeT][B10G10R10A2_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, B10G10R10A2_UNORM>::StoreGeneric; + table[TileModeT][B10G10R10A2_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, B10G10R10A2_UNORM_SRGB>::StoreGeneric; + table[TileModeT][R11G11B10_FLOAT] = StoreMacroTile, R32G32B32A32_FLOAT, R11G11B10_FLOAT>::StoreGeneric; + + table[TileModeT][R32_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, R32_SINT>::Store; + table[TileModeT][R32_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, R32_UINT>::Store; + table[TileModeT][R32_FLOAT] = StoreMacroTile, R32G32B32A32_FLOAT, R32_FLOAT>::Store; + table[TileModeT][A32_FLOAT] = StoreMacroTile, R32G32B32A32_FLOAT, A32_FLOAT>::Store; + table[TileModeT][B8G8R8X8_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, B8G8R8X8_UNORM>::Store; + table[TileModeT][B8G8R8X8_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, B8G8R8X8_UNORM_SRGB>::Store; + table[TileModeT][R8G8B8X8_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8B8X8_UNORM>::Store; + table[TileModeT][R8G8B8X8_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8B8X8_UNORM_SRGB>::Store; + + // 101010_2, 565, 555_1, and 444_4 formats force generic store tile for now + table[TileModeT][B10G10R10X2_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, B10G10R10X2_UNORM>::StoreGeneric; + table[TileModeT][B5G6R5_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, B5G6R5_UNORM>::Store; + table[TileModeT][B5G6R5_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, B5G6R5_UNORM_SRGB>::StoreGeneric; + table[TileModeT][B5G5R5A1_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, B5G5R5A1_UNORM>::StoreGeneric; + table[TileModeT][B5G5R5A1_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, B5G5R5A1_UNORM_SRGB>::StoreGeneric; + table[TileModeT][B4G4R4A4_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, B4G4R4A4_UNORM>::StoreGeneric; + table[TileModeT][B4G4R4A4_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, B4G4R4A4_UNORM_SRGB>::StoreGeneric; + + table[TileModeT][R8G8_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8_UNORM>::Store; + table[TileModeT][R8G8_SNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8_SNORM>::Store; + table[TileModeT][R8G8_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8_SINT>::Store; + table[TileModeT][R8G8_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8_UINT>::Store; + table[TileModeT][R16_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R16_UNORM>::Store; + table[TileModeT][R16_SNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R16_SNORM>::Store; + table[TileModeT][R16_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, R16_SINT>::Store; + table[TileModeT][R16_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, R16_UINT>::Store; + table[TileModeT][R16_FLOAT] = StoreMacroTile, R32G32B32A32_FLOAT, R16_FLOAT>::Store; + table[TileModeT][A16_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, A16_UNORM>::Store; + table[TileModeT][A16_FLOAT] = StoreMacroTile, R32G32B32A32_FLOAT, A16_FLOAT>::Store; + + // 101010_2, 565, 555_1, and 444_4 formats force generic store tile for now + table[TileModeT][B5G5R5X1_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, B5G5R5X1_UNORM>::StoreGeneric; + table[TileModeT][B5G5R5X1_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, B5G5R5X1_UNORM_SRGB>::StoreGeneric; + + table[TileModeT][R8_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R8_UNORM>::Store; + table[TileModeT][R8_SNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R8_SNORM>::Store; + table[TileModeT][R8_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, R8_SINT>::Store; + table[TileModeT][R8_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, R8_UINT>::Store; + table[TileModeT][A8_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, A8_UNORM>::Store; + table[TileModeT][BC1_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, BC1_UNORM>::Store; + table[TileModeT][BC2_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, BC2_UNORM>::Store; + table[TileModeT][BC3_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, BC3_UNORM>::Store; + table[TileModeT][BC4_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, BC4_UNORM>::Store; + table[TileModeT][BC5_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, BC5_UNORM>::Store; + table[TileModeT][BC1_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, BC1_UNORM_SRGB>::Store; + table[TileModeT][BC2_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, BC2_UNORM_SRGB>::Store; + table[TileModeT][BC3_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, BC3_UNORM_SRGB>::Store; + table[TileModeT][R8G8B8_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8B8_UNORM>::Store; + table[TileModeT][R8G8B8_SNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8B8_SNORM>::Store; + table[TileModeT][BC4_SNORM] = StoreMacroTile, R32G32B32A32_FLOAT, BC4_SNORM>::Store; + table[TileModeT][BC5_SNORM] = StoreMacroTile, R32G32B32A32_FLOAT, BC5_SNORM>::Store; + table[TileModeT][R16G16B16_FLOAT] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16B16_FLOAT>::Store; + table[TileModeT][R16G16B16_UNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16B16_UNORM>::Store; + table[TileModeT][R16G16B16_SNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16B16_SNORM>::Store; + table[TileModeT][R8G8B8_UNORM_SRGB] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8B8_UNORM_SRGB>::Store; + table[TileModeT][R16G16B16_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16B16_UINT>::Store; + table[TileModeT][R16G16B16_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, R16G16B16_SINT>::Store; + + // 101010_2, 565, 555_1, and 444_4 formats force generic store tile for now + table[TileModeT][R10G10B10A2_SNORM] = StoreMacroTile, R32G32B32A32_FLOAT, R10G10B10A2_SNORM>::StoreGeneric; + table[TileModeT][R10G10B10A2_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, R10G10B10A2_SINT>::StoreGeneric; + table[TileModeT][B10G10R10A2_SNORM] = StoreMacroTile, R32G32B32A32_FLOAT, B10G10R10A2_SNORM>::StoreGeneric; + table[TileModeT][B10G10R10A2_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, B10G10R10A2_UINT>::StoreGeneric; + table[TileModeT][B10G10R10A2_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, B10G10R10A2_SINT>::StoreGeneric; + + table[TileModeT][R8G8B8_UINT] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8B8_UINT>::Store; + table[TileModeT][R8G8B8_SINT] = StoreMacroTile, R32G32B32A32_FLOAT, R8G8B8_SINT>::Store; +} + +////////////////////////////////////////////////////////////////////////// +/// INIT_STORE_TILES_TABLE - Helper macro for setting up the tables. +template +void InitStoreTilesTableDepth( + PFN_STORE_TILES(&table)[NumTileModes][ArraySizeT]) +{ + table[TileModeT][R32_FLOAT] = StoreMacroTile, R32_FLOAT, R32_FLOAT>::Store; + table[TileModeT][R24_UNORM_X8_TYPELESS] = StoreMacroTile, R32_FLOAT, R24_UNORM_X8_TYPELESS>::Store; + table[TileModeT][R16_UNORM] = StoreMacroTile, R32_FLOAT, R16_UNORM>::Store; +} + +template +void InitStoreTilesTableStencil( + PFN_STORE_TILES(&table)[NumTileModes][ArraySizeT]) +{ + table[TileModeT][R32_UINT] = StoreMacroTile, R8_UINT, R32_UINT>::Store; + table[TileModeT][R8_UINT] = StoreMacroTile, R8_UINT, R8_UINT>::Store; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Sets up tables for StoreTile +void InitSimStoreTilesTable() +{ + memset(sStoreTilesTableColor, 0, sizeof(sStoreTilesTableColor)); + memset(sStoreTilesTableDepth, 0, sizeof(sStoreTilesTableDepth)); + + InitStoreTilesTableColor(sStoreTilesTableColor); + InitStoreTilesTableDepth(sStoreTilesTableDepth); + InitStoreTilesTableStencil(sStoreTilesTableStencil); + + InitStoreTilesTableColor(sStoreTilesTableColor); + InitStoreTilesTableColor(sStoreTilesTableColor); + + InitStoreTilesTableDepth(sStoreTilesTableDepth); + InitStoreTilesTableStencil(sStoreTilesTableStencil); +} diff --git a/src/gallium/drivers/swr/rasterizer/memory/TilingFunctions.h b/src/gallium/drivers/swr/rasterizer/memory/TilingFunctions.h new file mode 100644 index 00000000000..a14f3bf3f7c --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/memory/TilingFunctions.h @@ -0,0 +1,581 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file TilingFunctions.h +* +* @brief Tiling functions. +* +******************************************************************************/ +#pragma once + +#include "core/state.h" +#include "core/format_traits.h" +#include "memory/tilingtraits.h" + +#include + +#define MAX_NUM_LOD 15 + +#define GFX_ALIGN(x, a) (((x) + ((a) - 1)) - (((x) + ((a) - 1)) & ((a) - 1))) // Alt implementation with bitwise not (~) has issue with uint32 align used with 64-bit value, since ~'ed value will remain 32-bit. + +////////////////////////////////////////////////////////////////////////// +/// SimdTile SSE(2x2), AVX(4x2), or AVX-512(4x4?) +////////////////////////////////////////////////////////////////////////// +template +struct SimdTile +{ + // SimdTile is SOA (e.g. rrrrrrrr gggggggg bbbbbbbb aaaaaaaa ) + float color[FormatTraits::numComps][KNOB_SIMD_WIDTH]; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Retrieve color from simd. + /// @param index - linear index to color within simd. + /// @param outputColor - output color + INLINE void GetSwizzledColor( + uint32_t index, + float outputColor[4]) + { + // SOA pattern for 2x2 is a subset of 4x2. + // 0 1 4 5 + // 2 3 6 7 + // The offset converts pattern to linear +#if (SIMD_TILE_X_DIM == 4) + static const uint32_t offset[] = { 0, 1, 4, 5, 2, 3, 6, 7 }; +#elif (SIMD_TILE_X_DIM == 2) + static const uint32_t offset[] = { 0, 1, 2, 3 }; +#endif + + for (uint32_t i = 0; i < FormatTraits::numComps; ++i) + { + outputColor[i] = this->color[FormatTraits::swizzle(i)][offset[index]]; + } + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Retrieve color from simd. + /// @param index - linear index to color within simd. + /// @param outputColor - output color + INLINE void SetSwizzledColor( + uint32_t index, + const float src[4]) + { + // SOA pattern for 2x2 is a subset of 4x2. + // 0 1 4 5 + // 2 3 6 7 + // The offset converts pattern to linear +#if (SIMD_TILE_X_DIM == 4) + static const uint32_t offset[] = { 0, 1, 4, 5, 2, 3, 6, 7 }; +#elif (SIMD_TILE_X_DIM == 2) + static const uint32_t offset[] = { 0, 1, 2, 3 }; +#endif + + // Only loop over the components needed for destination. + for (uint32_t i = 0; i < FormatTraits::numComps; ++i) + { + this->color[i][offset[index]] = src[i]; + } + } +}; + +template<> +struct SimdTile +{ + // SimdTile is SOA (e.g. rrrrrrrr gggggggg bbbbbbbb aaaaaaaa ) + uint8_t color[FormatTraits::numComps][KNOB_SIMD_WIDTH]; + + ////////////////////////////////////////////////////////////////////////// + /// @brief Retrieve color from simd. + /// @param index - linear index to color within simd. + /// @param outputColor - output color + INLINE void GetSwizzledColor( + uint32_t index, + float outputColor[4]) + { + // SOA pattern for 2x2 is a subset of 4x2. + // 0 1 4 5 + // 2 3 6 7 + // The offset converts pattern to linear +#if (SIMD_TILE_X_DIM == 4) + static const uint32_t offset[] = { 0, 1, 4, 5, 2, 3, 6, 7 }; +#elif (SIMD_TILE_X_DIM == 2) + static const uint32_t offset[] = { 0, 1, 2, 3 }; +#endif + + for (uint32_t i = 0; i < FormatTraits::numComps; ++i) + { + uint32_t src = this->color[FormatTraits::swizzle(i)][offset[index]]; + outputColor[i] = *(float*)&src; + } + } + + ////////////////////////////////////////////////////////////////////////// + /// @brief Retrieve color from simd. + /// @param index - linear index to color within simd. + /// @param outputColor - output color + INLINE void SetSwizzledColor( + uint32_t index, + const float src[4]) + { + // SOA pattern for 2x2 is a subset of 4x2. + // 0 1 4 5 + // 2 3 6 7 + // The offset converts pattern to linear +#if (SIMD_TILE_X_DIM == 4) + static const uint32_t offset[] = { 0, 1, 4, 5, 2, 3, 6, 7 }; +#elif (SIMD_TILE_X_DIM == 2) + static const uint32_t offset[] = { 0, 1, 2, 3 }; +#endif + + // Only loop over the components needed for destination. + for (uint32_t i = 0; i < FormatTraits::numComps; ++i) + { + this->color[i][offset[index]] = *(uint8_t*)&src[i]; + } + } +}; + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes lod offset for 1D surface at specified lod. +/// @param baseWidth - width of basemip (mip 0). +/// @param hAlign - horizontal alignment per miip, in texels +/// @param lod - lod index +/// @param offset - output offset. +INLINE void ComputeLODOffset1D( + const SWR_FORMAT_INFO& info, + uint32_t baseWidth, + uint32_t hAlign, + uint32_t lod, + uint32_t &offset) +{ + if (lod == 0) + { + offset = 0; + } + else + { + uint32_t curWidth = baseWidth; + // translate mip width from pixels to blocks for block compressed formats + // @note hAlign is already in blocks for compressed formats so no need to convert + if (info.isBC) curWidth /= info.bcWidth; + + offset = GFX_ALIGN(curWidth, hAlign); + for (uint32_t l = 1; l < lod; ++l) + { + curWidth = GFX_ALIGN(std::max(curWidth >> 1, 1U), hAlign); + offset += curWidth; + } + + if (info.isSubsampled) + { + offset /= info.bcWidth; + } + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes x lod offset for 2D surface at specified lod. +/// @param baseWidth - width of basemip (mip 0). +/// @param hAlign - horizontal alignment per mip, in texels +/// @param lod - lod index +/// @param offset - output offset. +INLINE void ComputeLODOffsetX( + const SWR_FORMAT_INFO& info, + uint32_t baseWidth, + uint32_t hAlign, + uint32_t lod, + uint32_t &offset) +{ + if (lod < 2) + { + offset = 0; + } + else + { + uint32_t curWidth = baseWidth; + // convert mip width from pixels to blocks for block compressed formats + // @note hAlign is already in blocks for compressed formats so no need to convert + if (info.isBC) curWidth /= info.bcWidth; + + curWidth = std::max(curWidth >> 1, 1U); + curWidth = GFX_ALIGN(curWidth, hAlign); + + if (info.isSubsampled) + { + curWidth /= info.bcWidth; + } + + offset = curWidth; + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes y lod offset for 2D surface at specified lod. +/// @param baseWidth - width of basemip (mip 0). +/// @param vAlign - vertical alignment per mip, in rows +/// @param lod - lod index +/// @param offset - output offset. +INLINE void ComputeLODOffsetY( + const SWR_FORMAT_INFO& info, + uint32_t baseHeight, + uint32_t vAlign, + uint32_t lod, + uint32_t &offset) +{ + if (lod == 0) + { + offset = 0; + } + else + { + offset = 0; + uint32_t mipHeight = baseHeight; + + // translate mip height from pixels to blocks for block compressed formats + // @note VAlign is already in blocks for compressed formats so no need to convert + if (info.isBC) mipHeight /= info.bcHeight; + + for (uint32_t l = 1; l <= lod; ++l) + { + uint32_t alignedMipHeight = GFX_ALIGN(mipHeight, vAlign); + offset += ((l != 2) ? alignedMipHeight : 0); + mipHeight = std::max(mipHeight >> 1, 1U); + } + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes 1D surface offset +/// @param x - offset from start of array slice at given lod. +/// @param array - array slice index +/// @param lod - lod index +/// @param pState - surface state +/// @param xOffsetBytes - output offset in bytes. +template +INLINE void ComputeSurfaceOffset1D( + uint32_t x, + uint32_t array, + uint32_t lod, + const SWR_SURFACE_STATE *pState, + uint32_t &xOffsetBytes) +{ + const SWR_FORMAT_INFO &info = GetFormatInfo(pState->format); + uint32_t lodOffset; + + if (UseCachedOffsets) + { + lodOffset = pState->lodOffsets[0][lod]; + } + else + { + ComputeLODOffset1D(info, pState->width, pState->halign, lod, lodOffset); + } + + xOffsetBytes = (array * pState->qpitch + lodOffset + x) * info.Bpp; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Adjusts the array slice for legacy TileY MSAA +/// @param pState - surface state +/// @param array - array slice index +/// @param sampleNum - requested sample +INLINE void AdjustCoordsForMSAA(const SWR_SURFACE_STATE *pState, uint32_t& x, uint32_t& y, uint32_t& arrayIndex, uint32_t sampleNum) +{ + /// @todo: might want to templatize adjusting for sample slices when we support tileYS/tileYF. + if((pState->tileMode == SWR_TILE_MODE_YMAJOR || + pState->tileMode == SWR_TILE_MODE_WMAJOR) && + pState->bInterleavedSamples) + { + uint32_t newX, newY, newSampleX, newSampleY; + switch(pState->numSamples) + { + case 1: + newX = x; + newY = y; + newSampleX = newSampleY = 0; + break; + case 2: + { + assert(pState->type == SURFACE_2D); + static const uint32_t xMask = 0xFFFFFFFD; + static const uint32_t sampleMaskX = 0x1; + newX = pdep_u32(x, xMask); + newY = y; + newSampleX = pext_u32(sampleNum, sampleMaskX); + newSampleY = 0; + } + break; + case 4: + { + assert(pState->type == SURFACE_2D); + static const uint32_t mask = 0xFFFFFFFD; + static const uint32_t sampleMaskX = 0x1; + static const uint32_t sampleMaskY = 0x2; + newX = pdep_u32(x, mask); + newY = pdep_u32(y, mask); + newSampleX = pext_u32(sampleNum, sampleMaskX); + newSampleY = pext_u32(sampleNum, sampleMaskY); + } + break; + case 8: + { + assert(pState->type == SURFACE_2D); + static const uint32_t xMask = 0xFFFFFFF9; + static const uint32_t yMask = 0xFFFFFFFD; + static const uint32_t sampleMaskX = 0x5; + static const uint32_t sampleMaskY = 0x2; + newX = pdep_u32(x, xMask); + newY = pdep_u32(y, yMask); + newSampleX = pext_u32(sampleNum, sampleMaskX); + newSampleY = pext_u32(sampleNum, sampleMaskY); + } + break; + case 16: + { + assert(pState->type == SURFACE_2D); + static const uint32_t mask = 0xFFFFFFF9; + static const uint32_t sampleMaskX = 0x5; + static const uint32_t sampleMaskY = 0xA; + newX = pdep_u32(x, mask); + newY = pdep_u32(y, mask); + newSampleX = pext_u32(sampleNum, sampleMaskX); + newSampleY = pext_u32(sampleNum, sampleMaskY); + } + break; + default: + assert(0 && "Unsupported sample count"); + newX = newY = 0; + newSampleX = newSampleY = 0; + break; + } + x = newX | (newSampleX << 1); + y = newY | (newSampleY << 1); + } + else if(pState->tileMode == SWR_TILE_MODE_YMAJOR || + pState->tileMode == SWR_TILE_NONE) + { + uint32_t sampleShift; + switch(pState->numSamples) + { + case 1: + assert(sampleNum == 0); + sampleShift = 0; + break; + case 2: + assert(pState->type == SURFACE_2D); + sampleShift = 1; + break; + case 4: + assert(pState->type == SURFACE_2D); + sampleShift = 2; + break; + case 8: + assert(pState->type == SURFACE_2D); + sampleShift = 3; + break; + case 16: + assert(pState->type == SURFACE_2D); + sampleShift = 4; + break; + default: + assert(0 && "Unsupported sample count"); + sampleShift = 0; + break; + } + arrayIndex = (arrayIndex << sampleShift) | sampleNum; + } +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes 2D surface offset +/// @param x - horizontal offset from start of array slice and lod. +/// @param y - vertical offset from start of array slice and lod. +/// @param array - array slice index +/// @param lod - lod index +/// @param pState - surface state +/// @param xOffsetBytes - output x offset in bytes. +/// @param yOffsetRows - output y offset in bytes. +template +INLINE void ComputeSurfaceOffset2D(uint32_t x, uint32_t y, uint32_t array, uint32_t sampleNum, uint32_t lod, const SWR_SURFACE_STATE *pState, uint32_t &xOffsetBytes, uint32_t &yOffsetRows) +{ + const SWR_FORMAT_INFO &info = GetFormatInfo(pState->format); + uint32_t lodOffsetX, lodOffsetY; + + if (UseCachedOffsets) + { + lodOffsetX = pState->lodOffsets[0][lod]; + lodOffsetY = pState->lodOffsets[1][lod]; + } + else + { + ComputeLODOffsetX(info, pState->width, pState->halign, lod, lodOffsetX); + ComputeLODOffsetY(info, pState->height, pState->valign, lod, lodOffsetY); + } + + AdjustCoordsForMSAA(pState, x, y, array, sampleNum); + xOffsetBytes = (x + lodOffsetX + pState->xOffset) * info.Bpp; + yOffsetRows = (array * pState->qpitch) + lodOffsetY + y + pState->yOffset; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes 3D surface offset +/// @param x - horizontal offset from start of array slice and lod. +/// @param y - vertical offset from start of array slice and lod. +/// @param z - depth offset from start of array slice and lod. +/// @param lod - lod index +/// @param pState - surface state +/// @param xOffsetBytes - output x offset in bytes. +/// @param yOffsetRows - output y offset in rows. +/// @param zOffsetSlices - output y offset in slices. +template +INLINE void ComputeSurfaceOffset3D(uint32_t x, uint32_t y, uint32_t z, uint32_t lod, const SWR_SURFACE_STATE *pState, uint32_t &xOffsetBytes, uint32_t &yOffsetRows, uint32_t &zOffsetSlices) +{ + const SWR_FORMAT_INFO &info = GetFormatInfo(pState->format); + uint32_t lodOffsetX, lodOffsetY; + + if (UseCachedOffsets) + { + lodOffsetX = pState->lodOffsets[0][lod]; + lodOffsetY = pState->lodOffsets[1][lod]; + } + else + { + ComputeLODOffsetX(info, pState->width, pState->halign, lod, lodOffsetX); + ComputeLODOffsetY(info, pState->height, pState->valign, lod, lodOffsetY); + } + + xOffsetBytes = (x + lodOffsetX) * info.Bpp; + yOffsetRows = lodOffsetY + y; + zOffsetSlices = z; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Swizzles the linear x,y offsets depending on surface tiling mode +/// and returns final surface address +/// @param xOffsetBytes - x offset from base of surface in bytes +/// @param yOffsetRows - y offset from base of surface in rows +/// @param pState - pointer to the surface state +template +INLINE uint32_t ComputeTileSwizzle2D(uint32_t xOffsetBytes, uint32_t yOffsetRows, const SWR_SURFACE_STATE *pState) +{ + return ComputeOffset2D(pState->pitch, xOffsetBytes, yOffsetRows); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Swizzles the linear x,y offsets depending on surface tiling mode +/// and returns final surface address +/// @param xOffsetBytes - x offset from base of surface in bytes +/// @param yOffsetRows - y offset from base of surface in rows +/// @param pState - pointer to the surface state +template +INLINE uint32_t ComputeTileSwizzle3D(uint32_t xOffsetBytes, uint32_t yOffsetRows, uint32_t zOffsetSlices, const SWR_SURFACE_STATE *pState) +{ + return ComputeOffset3D(pState->qpitch, pState->pitch, xOffsetBytes, yOffsetRows, zOffsetSlices); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Swizzles the linear x,y offsets depending on surface tiling mode +/// and returns final surface address +/// @param xOffsetBytes - x offset from base of surface in bytes +/// @param yOffsetRows - y offset from base of surface in rows +/// @param pState - pointer to the surface state +INLINE +uint32_t TileSwizzle2D(uint32_t xOffsetBytes, uint32_t yOffsetRows, const SWR_SURFACE_STATE *pState) +{ + switch (pState->tileMode) + { + case SWR_TILE_NONE: return ComputeTileSwizzle2D >(xOffsetBytes, yOffsetRows, pState); + case SWR_TILE_SWRZ: return ComputeTileSwizzle2D >(xOffsetBytes, yOffsetRows, pState); + case SWR_TILE_MODE_XMAJOR: return ComputeTileSwizzle2D >(xOffsetBytes, yOffsetRows, pState); + case SWR_TILE_MODE_YMAJOR: return ComputeTileSwizzle2D >(xOffsetBytes, yOffsetRows, pState); + case SWR_TILE_MODE_WMAJOR: return ComputeTileSwizzle2D >(xOffsetBytes, yOffsetRows, pState); + default: SWR_ASSERT(0, "Unsupported tiling mode"); + } + return (uint32_t) NULL; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Swizzles the linear x,y,z offsets depending on surface tiling mode +/// and returns final surface address +/// @param xOffsetBytes - x offset from base of surface in bytes +/// @param yOffsetRows - y offset from base of surface in rows +/// @param zOffsetSlices - z offset from base of surface in slices +/// @param pState - pointer to the surface state +INLINE +uint32_t TileSwizzle3D(uint32_t xOffsetBytes, uint32_t yOffsetRows, uint32_t zOffsetSlices, const SWR_SURFACE_STATE *pState) +{ + switch (pState->tileMode) + { + case SWR_TILE_NONE: return ComputeTileSwizzle3D >(xOffsetBytes, yOffsetRows, zOffsetSlices, pState); + case SWR_TILE_SWRZ: return ComputeTileSwizzle3D >(xOffsetBytes, yOffsetRows, zOffsetSlices, pState); + case SWR_TILE_MODE_YMAJOR: return ComputeTileSwizzle3D >(xOffsetBytes, yOffsetRows, zOffsetSlices, pState); + default: SWR_ASSERT(0, "Unsupported tiling mode"); + } + return (uint32_t) NULL; +} + +template +INLINE +uint32_t ComputeSurfaceOffset(uint32_t x, uint32_t y, uint32_t z, uint32_t array, uint32_t sampleNum, uint32_t lod, const SWR_SURFACE_STATE *pState) +{ + uint32_t offsetX = 0, offsetY = 0, offsetZ = 0; + switch (pState->type) + { + case SURFACE_BUFFER: + case SURFACE_STRUCTURED_BUFFER: + offsetX = x * pState->pitch; + return offsetX; + break; + case SURFACE_1D: + ComputeSurfaceOffset1D(x, array, lod, pState, offsetX); + return TileSwizzle2D(offsetX, 0, pState); + break; + case SURFACE_2D: + ComputeSurfaceOffset2D(x, y, array, sampleNum, lod, pState, offsetX, offsetY); + return TileSwizzle2D(offsetX, offsetY, pState); + case SURFACE_3D: + ComputeSurfaceOffset3D(x, y, z, lod, pState, offsetX, offsetY, offsetZ); + return TileSwizzle3D(offsetX, offsetY, offsetZ, pState); + break; + case SURFACE_CUBE: + ComputeSurfaceOffset2D(x, y, array, sampleNum, lod, pState, offsetX, offsetY); + return TileSwizzle2D(offsetX, offsetY, pState); + break; + default: SWR_ASSERT(0, "Unsupported format"); + } + + return (uint32_t) NULL; +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes surface address at the given location and lod +/// @param x - x location in pixels +/// @param y - y location in rows +/// @param z - z location for 3D surfaces +/// @param array - array slice for 1D and 2D surfaces +/// @param lod - level of detail +/// @param pState - pointer to the surface state +template +INLINE +void* ComputeSurfaceAddress(uint32_t x, uint32_t y, uint32_t z, uint32_t array, uint32_t sampleNum, uint32_t lod, const SWR_SURFACE_STATE *pState) +{ + return pState->pBaseAddress + ComputeSurfaceOffset(x, y, z, array, sampleNum, lod, pState); +} diff --git a/src/gallium/drivers/swr/rasterizer/memory/tilingtraits.h b/src/gallium/drivers/swr/rasterizer/memory/tilingtraits.h new file mode 100644 index 00000000000..50f8e57c22a --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/memory/tilingtraits.h @@ -0,0 +1,263 @@ +/**************************************************************************** +* Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +* +* Permission is hereby granted, free of charge, to any person obtaining a +* copy of this software and associated documentation files (the "Software"), +* to deal in the Software without restriction, including without limitation +* the rights to use, copy, modify, merge, publish, distribute, sublicense, +* and/or sell copies of the Software, and to permit persons to whom the +* Software is furnished to do so, subject to the following conditions: +* +* The above copyright notice and this permission notice (including the next +* paragraph) shall be included in all copies or substantial portions of the +* Software. +* +* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +* IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +* FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +* THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +* LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +* FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +* IN THE SOFTWARE. +* +* @file tilingtraits.h +* +* @brief Tiling traits. +* +******************************************************************************/ +#pragma once + +#include "core/state.h" + +template +struct TilingTraits +{ + static const SWR_TILE_MODE TileMode{ mode }; + static UINT GetCu() { SWR_ASSERT(0); return 0; } + static UINT GetCv() { SWR_ASSERT(0); return 0; } + static UINT GetCr() { SWR_ASSERT(0); return 0; } + static UINT GetTileIDShift() { SWR_ASSERT(0); return 0; } + + /// @todo correct pdep shifts for all rastertile dims. Unused for now + static UINT GetPdepX() { SWR_ASSERT(0); return 0x37; } + static UINT GetPdepY() { SWR_ASSERT(0); return 0xC8; } +}; + +template struct TilingTraits +{ + static const SWR_TILE_MODE TileMode{ SWR_TILE_NONE }; + static UINT GetCu() { return 0; } + static UINT GetCv() { return 0; } + static UINT GetCr() { return 0; } + static UINT GetTileIDShift() { return 0; } + static UINT GetPdepX() { return 0x00; } + static UINT GetPdepY() { return 0x00; } +}; + +template<> struct TilingTraits +{ + static const SWR_TILE_MODE TileMode{ SWR_TILE_SWRZ }; + static UINT GetCu() { return KNOB_TILE_X_DIM_SHIFT; } + static UINT GetCv() { return KNOB_TILE_Y_DIM_SHIFT; } + static UINT GetCr() { return 0; } + static UINT GetTileIDShift() { return KNOB_TILE_X_DIM_SHIFT + KNOB_TILE_Y_DIM_SHIFT; } + + /// @todo correct pdep shifts for all rastertile dims. Unused for now + static UINT GetPdepX() { SWR_ASSERT(0); return 0x00; } + static UINT GetPdepY() { SWR_ASSERT(0); return 0x00; } +}; + +template<> struct TilingTraits +{ + static const SWR_TILE_MODE TileMode{ SWR_TILE_SWRZ }; + static UINT GetCu() { return KNOB_TILE_X_DIM_SHIFT + 2; } + static UINT GetCv() { return KNOB_TILE_Y_DIM_SHIFT; } + static UINT GetCr() { return 0; } + static UINT GetTileIDShift() { return KNOB_TILE_X_DIM_SHIFT + KNOB_TILE_Y_DIM_SHIFT + 2; } + + static UINT GetPdepX() { return 0x37; } + static UINT GetPdepY() { return 0xC8; } +}; + +template<> struct TilingTraits +{ + static const SWR_TILE_MODE TileMode{ SWR_TILE_SWRZ }; + static UINT GetCu() { return KNOB_TILE_X_DIM_SHIFT + 4; } + static UINT GetCv() { return KNOB_TILE_Y_DIM_SHIFT; } + static UINT GetCr() { return 0; } + static UINT GetTileIDShift() { return KNOB_TILE_X_DIM_SHIFT + KNOB_TILE_Y_DIM_SHIFT + 4; } + + /// @todo correct pdep shifts for all rastertile dims. Unused for now + static UINT GetPdepX() { SWR_ASSERT(0); return 0x37; } + static UINT GetPdepY() { SWR_ASSERT(0); return 0xC8; } +}; + +// y-major tiling layout unaffected by element size +template struct TilingTraits +{ + static const SWR_TILE_MODE TileMode{ SWR_TILE_MODE_YMAJOR }; + static UINT GetCu() { return 7; } + static UINT GetCv() { return 5; } + static UINT GetCr() { return 0; } + static UINT GetTileIDShift() { return 12; } + + static UINT GetPdepX() { return 0xe0f; } + static UINT GetPdepY() { return 0x1f0; } +}; + +// x-major tiling layout unaffected by element size +template struct TilingTraits +{ + static const SWR_TILE_MODE TileMode{ SWR_TILE_MODE_XMAJOR }; + static UINT GetCu() { return 9; } + static UINT GetCv() { return 3; } + static UINT GetCr() { return 0; } + static UINT GetTileIDShift() { return 12; } + + static UINT GetPdepX() { return 0x1ff; } + static UINT GetPdepY() { return 0xe00; } +}; + +template struct TilingTraits +{ + static const SWR_TILE_MODE TileMode{ SWR_TILE_MODE_WMAJOR }; + static UINT GetCu() { return 6; } + static UINT GetCv() { return 6; } + static UINT GetCr() { return 0; } + static UINT GetTileIDShift() { return 12; } + + static UINT GetPdepX() { return 0xe15; } + static UINT GetPdepY() { return 0x1ea; } +}; + +INLINE +UINT pdep_u32(UINT a, UINT mask) +{ +#if KNOB_ARCH==KNOB_ARCH_AVX2 + return _pdep_u32(a, mask); +#else + UINT result = 0; + + // copied from http://wm.ite.pl/articles/pdep-soft-emu.html + // using bsf instead of funky loop + DWORD maskIndex; + while (_BitScanForward(&maskIndex, mask)) + { + // 1. isolate lowest set bit of mask + const UINT lowest = 1 << maskIndex; + + // 2. populate LSB from src + const UINT LSB = (UINT)((int)(a << 31) >> 31); + + // 3. copy bit from mask + result |= LSB & lowest; + + // 4. clear lowest bit + mask &= ~lowest; + + // 5. prepare for next iteration + a >>= 1; + } + + return result; +#endif +} + +INLINE +UINT pext_u32(UINT a, UINT mask) +{ +#if KNOB_ARCH==KNOB_ARCH_AVX2 + return _pext_u32(a, mask); +#else + UINT result = 0; + DWORD maskIndex; + uint32_t currentBit = 0; + while (_BitScanForward(&maskIndex, mask)) + { + // 1. isolate lowest set bit of mask + const UINT lowest = 1 << maskIndex; + + // 2. copy bit from mask + result |= ((a & lowest) > 0) << currentBit++; + + // 3. clear lowest bit + mask &= ~lowest; + } + return result; +#endif +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes the tileID for 2D tiled surfaces +/// @param pitch - surface pitch in bytes +/// @param tileX - x offset in tiles +/// @param tileY - y offset in tiles +template +INLINE UINT ComputeTileOffset2D(UINT pitch, UINT tileX, UINT tileY) +{ + UINT tileID = tileY * (pitch >> TTraits::GetCu()) + tileX; + return tileID << TTraits::GetTileIDShift(); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes the tileID for 3D tiled surfaces +/// @param qpitch - surface qpitch in rows +/// @param pitch - surface pitch in bytes +/// @param tileX - x offset in tiles +/// @param tileY - y offset in tiles +/// @param tileZ - y offset in tiles +template +INLINE UINT ComputeTileOffset3D(UINT qpitch, UINT pitch, UINT tileX, UINT tileY, UINT tileZ) +{ + UINT tileID = (tileZ * (qpitch >> TTraits::GetCv()) + tileY) * (pitch >> TTraits::GetCu()) + tileX; + return tileID << TTraits::GetTileIDShift(); +} + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes the byte offset for 2D tiled surfaces +/// @param pitch - surface pitch in bytes +/// @param x - x offset in bytes +/// @param y - y offset in rows +template +INLINE UINT ComputeOffset2D(UINT pitch, UINT x, UINT y) +{ + UINT tileID = ComputeTileOffset2D(pitch, x >> TTraits::GetCu(), y >> TTraits::GetCv()); + UINT xSwizzle = pdep_u32(x, TTraits::GetPdepX()); + UINT ySwizzle = pdep_u32(y, TTraits::GetPdepY()); + return (tileID | xSwizzle | ySwizzle); +} + +#if KNOB_ARCH <= KNOB_ARCH_AVX +////////////////////////////////////////////////////////////////////////// +/// @brief Computes the byte offset for 2D tiled surfaces. Specialization +/// for tile-y surfaces that uses bit twiddling instead of pdep emulation. +/// @param pitch - surface pitch in bytes +/// @param x - x offset in bytes +/// @param y - y offset in rows +template<> +INLINE UINT ComputeOffset2D >(UINT pitch, UINT x, UINT y) +{ + typedef TilingTraits TTraits; + + UINT tileID = ComputeTileOffset2D(pitch, x >> TTraits::GetCu(), y >> TTraits::GetCv()); + UINT xSwizzle = ((x << 5) & 0xe00) | (x & 0xf); + UINT ySwizzle = (y << 4) & 0x1f0; + return (tileID | xSwizzle | ySwizzle); +} +#endif + +////////////////////////////////////////////////////////////////////////// +/// @brief Computes the byte offset for 3D tiled surfaces +/// @param qpitch - depth pitch in rows +/// @param pitch - surface pitch in bytes +/// @param x - x offset in bytes +/// @param y - y offset in rows +/// @param z - y offset in slices +template +INLINE UINT ComputeOffset3D(UINT qpitch, UINT pitch, UINT x, UINT y, UINT z) +{ + UINT tileID = ComputeTileOffset3D(qpitch, pitch, x >> TTraits::GetCu(), y >> TTraits::GetCv(), z >> TTraits::GetCr()); + UINT xSwizzle = pdep_u32(x, TTraits::GetPdepX()); + UINT ySwizzle = pdep_u32(y, TTraits::GetPdepY()); + return (tileID | xSwizzle | ySwizzle); +} diff --git a/src/gallium/drivers/swr/rasterizer/scripts/gen_knobs.py b/src/gallium/drivers/swr/rasterizer/scripts/gen_knobs.py new file mode 100644 index 00000000000..44ab69815b1 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/gen_knobs.py @@ -0,0 +1,79 @@ +# Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +# +# Permission is hereby granted, free of charge, to any person obtaining a +# copy of this software and associated documentation files (the "Software"), +# to deal in the Software without restriction, including without limitation +# the rights to use, copy, modify, merge, publish, distribute, sublicense, +# and/or sell copies of the Software, and to permit persons to whom the +# Software is furnished to do so, subject to the following conditions: +# +# The above copyright notice and this permission notice (including the next +# paragraph) shall be included in all copies or substantial portions of the +# Software. +# +# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +# IN THE SOFTWARE. + +# Python source +from __future__ import print_function +import os +import sys +import knob_defs +from mako.template import Template +from mako.exceptions import RichTraceback + +def write_template_to_string(template_filename, **kwargs): + try: + template = Template(filename=template_filename) + # Split + Join fixes line-endings for whatever platform you are using + return '\n'.join(template.render(**kwargs).splitlines()) + except: + traceback = RichTraceback() + for (filename, lineno, function, line) in traceback.traceback: + print("File %s, line %s, in %s" % (filename, lineno, function)) + print(line, "\n") + print("%s: %s" % (str(traceback.error.__class__.__name__), traceback.error)) + +def write_template_to_file(template_filename, output_filename, **kwargs): + with open(output_filename, "w") as outfile: + print(write_template_to_string(template_filename, **kwargs), file=outfile) + +def main(args=sys.argv[1:]): + if len(args) != 1: + print('Usage:', sys.argv[0], '', file=sys.stderr) + return 1 + + output_dir = args[0] + if not os.path.isdir(output_dir): + if os.path.exists(output_dir): + print('ERROR: Invalid output directory:', output_dir, file=sys.stderr) + return 1 + + try: + os.makedirs(output_dir) + except: + print('ERROR: Could not create output directory:', output_dir, file=sys.stderr) + return 1 + + # Output path exists, now just run the template + template_file = os.sep.join([sys.path[0], 'templates', 'knobs.template']) + output_file = os.sep.join([output_dir, 'gen_knobs.cpp']) + output_header = os.sep.join([output_dir, 'gen_knobs.h']) + + for f in [output_header, output_file]: + write_template_to_file(template_file, f, + filename='gen_knobs', + knobs=knob_defs.KNOBS, + includes=['core/knobs_init.h', 'common/os.h', 'sstream', 'iomanip'], + gen_header=True if f == output_header else False) + + return 0 + +if __name__ == '__main__': + sys.exit(main()) + diff --git a/src/gallium/drivers/swr/rasterizer/scripts/knob_defs.py b/src/gallium/drivers/swr/rasterizer/scripts/knob_defs.py new file mode 100644 index 00000000000..8c51e1e8e73 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/knob_defs.py @@ -0,0 +1,226 @@ +# Copyright (C) 2014-2015 Intel Corporation. All Rights Reserved. +# +# Permission is hereby granted, free of charge, to any person obtaining a +# copy of this software and associated documentation files (the "Software"), +# to deal in the Software without restriction, including without limitation +# the rights to use, copy, modify, merge, publish, distribute, sublicense, +# and/or sell copies of the Software, and to permit persons to whom the +# Software is furnished to do so, subject to the following conditions: +# +# The above copyright notice and this permission notice (including the next +# paragraph) shall be included in all copies or substantial portions of the +# Software. +# +# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR +# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, +# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL +# THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER +# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING +# FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS +# IN THE SOFTWARE. + +# Python source +KNOBS = [ + ['ENABLE_ASSERT_DIALOGS', { + 'type' : 'bool', + 'default' : 'true', + 'desc' : ['Use dialogs when asserts fire.', + 'Asserts are only enabled in debug builds'], + }], + + ['SINGLE_THREADED', { + 'type' : 'bool', + 'default' : 'false', + 'desc' : ['If enabled will perform all rendering on the API thread.', + 'This is useful mainly for debugging purposes.'], + }], + + ['DUMP_SHADER_IR', { + 'type' : 'bool', + 'default' : 'false', + 'desc' : ['Dumps shader LLVM IR at various stages of jit compilation.'], + }], + + ['USE_GENERIC_STORETILE', { + 'type' : 'bool', + 'default' : 'false', + 'desc' : ['Always use generic function for performing StoreTile.', + 'Will be slightly slower than using optimized (jitted) path'], + }], + + ['FAST_CLEAR', { + 'type' : 'bool', + 'default' : 'true', + 'desc' : ['Replace 3D primitive execute with a SWRClearRT operation and', + 'defer clear execution to first backend op on hottile, or hottile store'], + }], + + ['MAX_NUMA_NODES', { + 'type' : 'uint32_t', + 'default' : '0', + 'desc' : ['Maximum # of NUMA-nodes per system used for worker threads', + ' 0 == ALL NUMA-nodes in the system', + ' N == Use at most N NUMA-nodes for rendering'], + }], + + ['MAX_CORES_PER_NUMA_NODE', { + 'type' : 'uint32_t', + 'default' : '0', + 'desc' : ['Maximum # of cores per NUMA-node used for worker threads.', + ' 0 == ALL non-API thread cores per NUMA-node', + ' N == Use at most N cores per NUMA-node'], + }], + + ['MAX_THREADS_PER_CORE', { + 'type' : 'uint32_t', + 'default' : '1', + 'desc' : ['Maximum # of (hyper)threads per physical core used for worker threads.', + ' 0 == ALL hyper-threads per core', + ' N == Use at most N hyper-threads per physical core'], + }], + + ['MAX_WORKER_THREADS', { + 'type' : 'uint32_t', + 'default' : '0', + 'desc' : ['Maximum worker threads to spawn.', + '', + 'IMPORTANT: If this is non-zero, no worker threads will be bound to', + 'specific HW threads. They will all be "floating" SW threads.', + 'In this case, the above 3 KNOBS will be ignored.'], + }], + + ['BUCKETS_START_FRAME', { + 'type' : 'uint32_t', + 'default' : '1200', + 'desc' : ['Frame from when to start saving buckets data.', + '', + 'NOTE: KNOB_ENABLE_RDTSC must be enabled in core/knobs.h', + 'for this to have an effect.'], + }], + + ['BUCKETS_END_FRAME', { + 'type' : 'uint32_t', + 'default' : '1400', + 'desc' : ['Frame at which to stop saving buckets data.', + '', + 'NOTE: KNOB_ENABLE_RDTSC must be enabled in core/knobs.h', + 'for this to have an effect.'], + }], + + ['WORKER_SPIN_LOOP_COUNT', { + 'type' : 'uint32_t', + 'default' : '5000', + 'desc' : ['Number of spin-loop iterations worker threads will perform', + 'before going to sleep when waiting for work'], + }], + + ['MAX_DRAWS_IN_FLIGHT', { + 'type' : 'uint32_t', + 'default' : '160', + 'desc' : ['Maximum number of draws outstanding before API thread blocks.'], + }], + + ['MAX_PRIMS_PER_DRAW', { + 'type' : 'uint32_t', + 'default' : '2040', + 'desc' : ['Maximum primitives in a single Draw().', + 'Larger primitives are split into smaller Draw calls.', + 'Should be a multiple of (3 * vectorWidth).'], + }], + + ['MAX_TESS_PRIMS_PER_DRAW', { + 'type' : 'uint32_t', + 'default' : '16', + 'desc' : ['Maximum primitives in a single Draw() with tessellation enabled.', + 'Larger primitives are split into smaller Draw calls.', + 'Should be a multiple of (vectorWidth).'], + }], + + ['MAX_FRAC_ODD_TESS_FACTOR', { + 'type' : 'float', + 'default' : '63.0f', + 'desc' : ['(DEBUG) Maximum tessellation factor for fractional-odd partitioning.'], + }], + + ['MAX_FRAC_EVEN_TESS_FACTOR', { + 'type' : 'float', + 'default' : '64.0f', + 'desc' : ['(DEBUG) Maximum tessellation factor for fractional-even partitioning.'], + }], + + ['MAX_INTEGER_TESS_FACTOR', { + 'type' : 'uint32_t', + 'default' : '64', + 'desc' : ['(DEBUG) Maximum tessellation factor for integer partitioning.'], + }], + + + ['BUCKETS_ENABLE_THREADVIZ', { + 'type' : 'bool', + 'default' : 'false', + 'desc' : ['Enable threadviz output.'], + }], + + ['TOSS_DRAW', { + 'type' : 'bool', + 'default' : 'false', + 'desc' : ['Disable per-draw/dispatch execution'], + }], + + ['TOSS_QUEUE_FE', { + 'type' : 'bool', + 'default' : 'false', + 'desc' : ['Stop per-draw execution at worker FE', + '', + 'NOTE: Requires KNOB_ENABLE_TOSS_POINTS to be enabled in core/knobs.h'], + }], + + ['TOSS_FETCH', { + 'type' : 'bool', + 'default' : 'false', + 'desc' : ['Stop per-draw execution at vertex fetch', + '', + 'NOTE: Requires KNOB_ENABLE_TOSS_POINTS to be enabled in core/knobs.h'], + }], + + ['TOSS_IA', { + 'type' : 'bool', + 'default' : 'false', + 'desc' : ['Stop per-draw execution at input assembler', + '', + 'NOTE: Requires KNOB_ENABLE_TOSS_POINTS to be enabled in core/knobs.h'], + }], + + ['TOSS_VS', { + 'type' : 'bool', + 'default' : 'false', + 'desc' : ['Stop per-draw execution at vertex shader', + '', + 'NOTE: Requires KNOB_ENABLE_TOSS_POINTS to be enabled in core/knobs.h'], + }], + + ['TOSS_SETUP_TRIS', { + 'type' : 'bool', + 'default' : 'false', + 'desc' : ['Stop per-draw execution at primitive setup', + '', + 'NOTE: Requires KNOB_ENABLE_TOSS_POINTS to be enabled in core/knobs.h'], + }], + + ['TOSS_BIN_TRIS', { + 'type' : 'bool', + 'default' : 'false', + 'desc' : ['Stop per-draw execution at primitive binning', + '', + 'NOTE: Requires KNOB_ENABLE_TOSS_POINTS to be enabled in core/knobs.h'], + }], + + ['TOSS_RS', { + 'type' : 'bool', + 'default' : 'false', + 'desc' : ['Stop per-draw execution at rasterizer', + '', + 'NOTE: Requires KNOB_ENABLE_TOSS_POINTS to be enabled in core/knobs.h'], + }], + +] diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/__init__.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/__init__.py new file mode 100644 index 00000000000..d9638481889 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/__init__.py @@ -0,0 +1,8 @@ +# mako/__init__.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + + +__version__ = '1.0.1' diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/_ast_util.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/_ast_util.py new file mode 100644 index 00000000000..efbc4fc245d --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/_ast_util.py @@ -0,0 +1,845 @@ +# mako/_ast_util.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + +""" + ast + ~~~ + + The `ast` module helps Python applications to process trees of the Python + abstract syntax grammar. The abstract syntax itself might change with + each Python release; this module helps to find out programmatically what + the current grammar looks like and allows modifications of it. + + An abstract syntax tree can be generated by passing `ast.PyCF_ONLY_AST` as + a flag to the `compile()` builtin function or by using the `parse()` + function from this module. The result will be a tree of objects whose + classes all inherit from `ast.AST`. + + A modified abstract syntax tree can be compiled into a Python code object + using the built-in `compile()` function. + + Additionally various helper functions are provided that make working with + the trees simpler. The main intention of the helper functions and this + module in general is to provide an easy to use interface for libraries + that work tightly with the python syntax (template engines for example). + + + :copyright: Copyright 2008 by Armin Ronacher. + :license: Python License. +""" +from _ast import * +from mako.compat import arg_stringname + +BOOLOP_SYMBOLS = { + And: 'and', + Or: 'or' +} + +BINOP_SYMBOLS = { + Add: '+', + Sub: '-', + Mult: '*', + Div: '/', + FloorDiv: '//', + Mod: '%', + LShift: '<<', + RShift: '>>', + BitOr: '|', + BitAnd: '&', + BitXor: '^' +} + +CMPOP_SYMBOLS = { + Eq: '==', + Gt: '>', + GtE: '>=', + In: 'in', + Is: 'is', + IsNot: 'is not', + Lt: '<', + LtE: '<=', + NotEq: '!=', + NotIn: 'not in' +} + +UNARYOP_SYMBOLS = { + Invert: '~', + Not: 'not', + UAdd: '+', + USub: '-' +} + +ALL_SYMBOLS = {} +ALL_SYMBOLS.update(BOOLOP_SYMBOLS) +ALL_SYMBOLS.update(BINOP_SYMBOLS) +ALL_SYMBOLS.update(CMPOP_SYMBOLS) +ALL_SYMBOLS.update(UNARYOP_SYMBOLS) + + +def parse(expr, filename='', mode='exec'): + """Parse an expression into an AST node.""" + return compile(expr, filename, mode, PyCF_ONLY_AST) + + +def to_source(node, indent_with=' ' * 4): + """ + This function can convert a node tree back into python sourcecode. This + is useful for debugging purposes, especially if you're dealing with custom + asts not generated by python itself. + + It could be that the sourcecode is evaluable when the AST itself is not + compilable / evaluable. The reason for this is that the AST contains some + more data than regular sourcecode does, which is dropped during + conversion. + + Each level of indentation is replaced with `indent_with`. Per default this + parameter is equal to four spaces as suggested by PEP 8, but it might be + adjusted to match the application's styleguide. + """ + generator = SourceGenerator(indent_with) + generator.visit(node) + return ''.join(generator.result) + + +def dump(node): + """ + A very verbose representation of the node passed. This is useful for + debugging purposes. + """ + def _format(node): + if isinstance(node, AST): + return '%s(%s)' % (node.__class__.__name__, + ', '.join('%s=%s' % (a, _format(b)) + for a, b in iter_fields(node))) + elif isinstance(node, list): + return '[%s]' % ', '.join(_format(x) for x in node) + return repr(node) + if not isinstance(node, AST): + raise TypeError('expected AST, got %r' % node.__class__.__name__) + return _format(node) + + +def copy_location(new_node, old_node): + """ + Copy the source location hint (`lineno` and `col_offset`) from the + old to the new node if possible and return the new one. + """ + for attr in 'lineno', 'col_offset': + if attr in old_node._attributes and attr in new_node._attributes \ + and hasattr(old_node, attr): + setattr(new_node, attr, getattr(old_node, attr)) + return new_node + + +def fix_missing_locations(node): + """ + Some nodes require a line number and the column offset. Without that + information the compiler will abort the compilation. Because it can be + a dull task to add appropriate line numbers and column offsets when + adding new nodes this function can help. It copies the line number and + column offset of the parent node to the child nodes without this + information. + + Unlike `copy_location` this works recursive and won't touch nodes that + already have a location information. + """ + def _fix(node, lineno, col_offset): + if 'lineno' in node._attributes: + if not hasattr(node, 'lineno'): + node.lineno = lineno + else: + lineno = node.lineno + if 'col_offset' in node._attributes: + if not hasattr(node, 'col_offset'): + node.col_offset = col_offset + else: + col_offset = node.col_offset + for child in iter_child_nodes(node): + _fix(child, lineno, col_offset) + _fix(node, 1, 0) + return node + + +def increment_lineno(node, n=1): + """ + Increment the line numbers of all nodes by `n` if they have line number + attributes. This is useful to "move code" to a different location in a + file. + """ + for node in zip((node,), walk(node)): + if 'lineno' in node._attributes: + node.lineno = getattr(node, 'lineno', 0) + n + + +def iter_fields(node): + """Iterate over all fields of a node, only yielding existing fields.""" + # CPython 2.5 compat + if not hasattr(node, '_fields') or not node._fields: + return + for field in node._fields: + try: + yield field, getattr(node, field) + except AttributeError: + pass + + +def get_fields(node): + """Like `iter_fiels` but returns a dict.""" + return dict(iter_fields(node)) + + +def iter_child_nodes(node): + """Iterate over all child nodes or a node.""" + for name, field in iter_fields(node): + if isinstance(field, AST): + yield field + elif isinstance(field, list): + for item in field: + if isinstance(item, AST): + yield item + + +def get_child_nodes(node): + """Like `iter_child_nodes` but returns a list.""" + return list(iter_child_nodes(node)) + + +def get_compile_mode(node): + """ + Get the mode for `compile` of a given node. If the node is not a `mod` + node (`Expression`, `Module` etc.) a `TypeError` is thrown. + """ + if not isinstance(node, mod): + raise TypeError('expected mod node, got %r' % node.__class__.__name__) + return { + Expression: 'eval', + Interactive: 'single' + }.get(node.__class__, 'expr') + + +def get_docstring(node): + """ + Return the docstring for the given node or `None` if no docstring can be + found. If the node provided does not accept docstrings a `TypeError` + will be raised. + """ + if not isinstance(node, (FunctionDef, ClassDef, Module)): + raise TypeError("%r can't have docstrings" % node.__class__.__name__) + if node.body and isinstance(node.body[0], Str): + return node.body[0].s + + +def walk(node): + """ + Iterate over all nodes. This is useful if you only want to modify nodes in + place and don't care about the context or the order the nodes are returned. + """ + from collections import deque + todo = deque([node]) + while todo: + node = todo.popleft() + todo.extend(iter_child_nodes(node)) + yield node + + +class NodeVisitor(object): + """ + Walks the abstract syntax tree and call visitor functions for every node + found. The visitor functions may return values which will be forwarded + by the `visit` method. + + Per default the visitor functions for the nodes are ``'visit_'`` + + class name of the node. So a `TryFinally` node visit function would + be `visit_TryFinally`. This behavior can be changed by overriding + the `get_visitor` function. If no visitor function exists for a node + (return value `None`) the `generic_visit` visitor is used instead. + + Don't use the `NodeVisitor` if you want to apply changes to nodes during + traversing. For this a special visitor exists (`NodeTransformer`) that + allows modifications. + """ + + def get_visitor(self, node): + """ + Return the visitor function for this node or `None` if no visitor + exists for this node. In that case the generic visit function is + used instead. + """ + method = 'visit_' + node.__class__.__name__ + return getattr(self, method, None) + + def visit(self, node): + """Visit a node.""" + f = self.get_visitor(node) + if f is not None: + return f(node) + return self.generic_visit(node) + + def generic_visit(self, node): + """Called if no explicit visitor function exists for a node.""" + for field, value in iter_fields(node): + if isinstance(value, list): + for item in value: + if isinstance(item, AST): + self.visit(item) + elif isinstance(value, AST): + self.visit(value) + + +class NodeTransformer(NodeVisitor): + """ + Walks the abstract syntax tree and allows modifications of nodes. + + The `NodeTransformer` will walk the AST and use the return value of the + visitor functions to replace or remove the old node. If the return + value of the visitor function is `None` the node will be removed + from the previous location otherwise it's replaced with the return + value. The return value may be the original node in which case no + replacement takes place. + + Here an example transformer that rewrites all `foo` to `data['foo']`:: + + class RewriteName(NodeTransformer): + + def visit_Name(self, node): + return copy_location(Subscript( + value=Name(id='data', ctx=Load()), + slice=Index(value=Str(s=node.id)), + ctx=node.ctx + ), node) + + Keep in mind that if the node you're operating on has child nodes + you must either transform the child nodes yourself or call the generic + visit function for the node first. + + Nodes that were part of a collection of statements (that applies to + all statement nodes) may also return a list of nodes rather than just + a single node. + + Usually you use the transformer like this:: + + node = YourTransformer().visit(node) + """ + + def generic_visit(self, node): + for field, old_value in iter_fields(node): + old_value = getattr(node, field, None) + if isinstance(old_value, list): + new_values = [] + for value in old_value: + if isinstance(value, AST): + value = self.visit(value) + if value is None: + continue + elif not isinstance(value, AST): + new_values.extend(value) + continue + new_values.append(value) + old_value[:] = new_values + elif isinstance(old_value, AST): + new_node = self.visit(old_value) + if new_node is None: + delattr(node, field) + else: + setattr(node, field, new_node) + return node + + +class SourceGenerator(NodeVisitor): + """ + This visitor is able to transform a well formed syntax tree into python + sourcecode. For more details have a look at the docstring of the + `node_to_source` function. + """ + + def __init__(self, indent_with): + self.result = [] + self.indent_with = indent_with + self.indentation = 0 + self.new_lines = 0 + + def write(self, x): + if self.new_lines: + if self.result: + self.result.append('\n' * self.new_lines) + self.result.append(self.indent_with * self.indentation) + self.new_lines = 0 + self.result.append(x) + + def newline(self, n=1): + self.new_lines = max(self.new_lines, n) + + def body(self, statements): + self.new_line = True + self.indentation += 1 + for stmt in statements: + self.visit(stmt) + self.indentation -= 1 + + def body_or_else(self, node): + self.body(node.body) + if node.orelse: + self.newline() + self.write('else:') + self.body(node.orelse) + + def signature(self, node): + want_comma = [] + def write_comma(): + if want_comma: + self.write(', ') + else: + want_comma.append(True) + + padding = [None] * (len(node.args) - len(node.defaults)) + for arg, default in zip(node.args, padding + node.defaults): + write_comma() + self.visit(arg) + if default is not None: + self.write('=') + self.visit(default) + if node.vararg is not None: + write_comma() + self.write('*' + arg_stringname(node.vararg)) + if node.kwarg is not None: + write_comma() + self.write('**' + arg_stringname(node.kwarg)) + + def decorators(self, node): + for decorator in node.decorator_list: + self.newline() + self.write('@') + self.visit(decorator) + + # Statements + + def visit_Assign(self, node): + self.newline() + for idx, target in enumerate(node.targets): + if idx: + self.write(', ') + self.visit(target) + self.write(' = ') + self.visit(node.value) + + def visit_AugAssign(self, node): + self.newline() + self.visit(node.target) + self.write(BINOP_SYMBOLS[type(node.op)] + '=') + self.visit(node.value) + + def visit_ImportFrom(self, node): + self.newline() + self.write('from %s%s import ' % ('.' * node.level, node.module)) + for idx, item in enumerate(node.names): + if idx: + self.write(', ') + self.write(item) + + def visit_Import(self, node): + self.newline() + for item in node.names: + self.write('import ') + self.visit(item) + + def visit_Expr(self, node): + self.newline() + self.generic_visit(node) + + def visit_FunctionDef(self, node): + self.newline(n=2) + self.decorators(node) + self.newline() + self.write('def %s(' % node.name) + self.signature(node.args) + self.write('):') + self.body(node.body) + + def visit_ClassDef(self, node): + have_args = [] + def paren_or_comma(): + if have_args: + self.write(', ') + else: + have_args.append(True) + self.write('(') + + self.newline(n=3) + self.decorators(node) + self.newline() + self.write('class %s' % node.name) + for base in node.bases: + paren_or_comma() + self.visit(base) + # XXX: the if here is used to keep this module compatible + # with python 2.6. + if hasattr(node, 'keywords'): + for keyword in node.keywords: + paren_or_comma() + self.write(keyword.arg + '=') + self.visit(keyword.value) + if node.starargs is not None: + paren_or_comma() + self.write('*') + self.visit(node.starargs) + if node.kwargs is not None: + paren_or_comma() + self.write('**') + self.visit(node.kwargs) + self.write(have_args and '):' or ':') + self.body(node.body) + + def visit_If(self, node): + self.newline() + self.write('if ') + self.visit(node.test) + self.write(':') + self.body(node.body) + while True: + else_ = node.orelse + if len(else_) == 1 and isinstance(else_[0], If): + node = else_[0] + self.newline() + self.write('elif ') + self.visit(node.test) + self.write(':') + self.body(node.body) + else: + self.newline() + self.write('else:') + self.body(else_) + break + + def visit_For(self, node): + self.newline() + self.write('for ') + self.visit(node.target) + self.write(' in ') + self.visit(node.iter) + self.write(':') + self.body_or_else(node) + + def visit_While(self, node): + self.newline() + self.write('while ') + self.visit(node.test) + self.write(':') + self.body_or_else(node) + + def visit_With(self, node): + self.newline() + self.write('with ') + self.visit(node.context_expr) + if node.optional_vars is not None: + self.write(' as ') + self.visit(node.optional_vars) + self.write(':') + self.body(node.body) + + def visit_Pass(self, node): + self.newline() + self.write('pass') + + def visit_Print(self, node): + # XXX: python 2.6 only + self.newline() + self.write('print ') + want_comma = False + if node.dest is not None: + self.write(' >> ') + self.visit(node.dest) + want_comma = True + for value in node.values: + if want_comma: + self.write(', ') + self.visit(value) + want_comma = True + if not node.nl: + self.write(',') + + def visit_Delete(self, node): + self.newline() + self.write('del ') + for idx, target in enumerate(node): + if idx: + self.write(', ') + self.visit(target) + + def visit_TryExcept(self, node): + self.newline() + self.write('try:') + self.body(node.body) + for handler in node.handlers: + self.visit(handler) + + def visit_TryFinally(self, node): + self.newline() + self.write('try:') + self.body(node.body) + self.newline() + self.write('finally:') + self.body(node.finalbody) + + def visit_Global(self, node): + self.newline() + self.write('global ' + ', '.join(node.names)) + + def visit_Nonlocal(self, node): + self.newline() + self.write('nonlocal ' + ', '.join(node.names)) + + def visit_Return(self, node): + self.newline() + self.write('return ') + self.visit(node.value) + + def visit_Break(self, node): + self.newline() + self.write('break') + + def visit_Continue(self, node): + self.newline() + self.write('continue') + + def visit_Raise(self, node): + # XXX: Python 2.6 / 3.0 compatibility + self.newline() + self.write('raise') + if hasattr(node, 'exc') and node.exc is not None: + self.write(' ') + self.visit(node.exc) + if node.cause is not None: + self.write(' from ') + self.visit(node.cause) + elif hasattr(node, 'type') and node.type is not None: + self.visit(node.type) + if node.inst is not None: + self.write(', ') + self.visit(node.inst) + if node.tback is not None: + self.write(', ') + self.visit(node.tback) + + # Expressions + + def visit_Attribute(self, node): + self.visit(node.value) + self.write('.' + node.attr) + + def visit_Call(self, node): + want_comma = [] + def write_comma(): + if want_comma: + self.write(', ') + else: + want_comma.append(True) + + self.visit(node.func) + self.write('(') + for arg in node.args: + write_comma() + self.visit(arg) + for keyword in node.keywords: + write_comma() + self.write(keyword.arg + '=') + self.visit(keyword.value) + if node.starargs is not None: + write_comma() + self.write('*') + self.visit(node.starargs) + if node.kwargs is not None: + write_comma() + self.write('**') + self.visit(node.kwargs) + self.write(')') + + def visit_Name(self, node): + self.write(node.id) + + def visit_NameConstant(self, node): + self.write(str(node.value)) + + def visit_arg(self, node): + self.write(node.arg) + + def visit_Str(self, node): + self.write(repr(node.s)) + + def visit_Bytes(self, node): + self.write(repr(node.s)) + + def visit_Num(self, node): + self.write(repr(node.n)) + + def visit_Tuple(self, node): + self.write('(') + idx = -1 + for idx, item in enumerate(node.elts): + if idx: + self.write(', ') + self.visit(item) + self.write(idx and ')' or ',)') + + def sequence_visit(left, right): + def visit(self, node): + self.write(left) + for idx, item in enumerate(node.elts): + if idx: + self.write(', ') + self.visit(item) + self.write(right) + return visit + + visit_List = sequence_visit('[', ']') + visit_Set = sequence_visit('{', '}') + del sequence_visit + + def visit_Dict(self, node): + self.write('{') + for idx, (key, value) in enumerate(zip(node.keys, node.values)): + if idx: + self.write(', ') + self.visit(key) + self.write(': ') + self.visit(value) + self.write('}') + + def visit_BinOp(self, node): + self.write('(') + self.visit(node.left) + self.write(' %s ' % BINOP_SYMBOLS[type(node.op)]) + self.visit(node.right) + self.write(')') + + def visit_BoolOp(self, node): + self.write('(') + for idx, value in enumerate(node.values): + if idx: + self.write(' %s ' % BOOLOP_SYMBOLS[type(node.op)]) + self.visit(value) + self.write(')') + + def visit_Compare(self, node): + self.write('(') + self.visit(node.left) + for op, right in zip(node.ops, node.comparators): + self.write(' %s ' % CMPOP_SYMBOLS[type(op)]) + self.visit(right) + self.write(')') + + def visit_UnaryOp(self, node): + self.write('(') + op = UNARYOP_SYMBOLS[type(node.op)] + self.write(op) + if op == 'not': + self.write(' ') + self.visit(node.operand) + self.write(')') + + def visit_Subscript(self, node): + self.visit(node.value) + self.write('[') + self.visit(node.slice) + self.write(']') + + def visit_Slice(self, node): + if node.lower is not None: + self.visit(node.lower) + self.write(':') + if node.upper is not None: + self.visit(node.upper) + if node.step is not None: + self.write(':') + if not (isinstance(node.step, Name) and node.step.id == 'None'): + self.visit(node.step) + + def visit_ExtSlice(self, node): + for idx, item in node.dims: + if idx: + self.write(', ') + self.visit(item) + + def visit_Yield(self, node): + self.write('yield ') + self.visit(node.value) + + def visit_Lambda(self, node): + self.write('lambda ') + self.signature(node.args) + self.write(': ') + self.visit(node.body) + + def visit_Ellipsis(self, node): + self.write('Ellipsis') + + def generator_visit(left, right): + def visit(self, node): + self.write(left) + self.visit(node.elt) + for comprehension in node.generators: + self.visit(comprehension) + self.write(right) + return visit + + visit_ListComp = generator_visit('[', ']') + visit_GeneratorExp = generator_visit('(', ')') + visit_SetComp = generator_visit('{', '}') + del generator_visit + + def visit_DictComp(self, node): + self.write('{') + self.visit(node.key) + self.write(': ') + self.visit(node.value) + for comprehension in node.generators: + self.visit(comprehension) + self.write('}') + + def visit_IfExp(self, node): + self.visit(node.body) + self.write(' if ') + self.visit(node.test) + self.write(' else ') + self.visit(node.orelse) + + def visit_Starred(self, node): + self.write('*') + self.visit(node.value) + + def visit_Repr(self, node): + # XXX: python 2.6 only + self.write('`') + self.visit(node.value) + self.write('`') + + # Helper Nodes + + def visit_alias(self, node): + self.write(node.name) + if node.asname is not None: + self.write(' as ' + node.asname) + + def visit_comprehension(self, node): + self.write(' for ') + self.visit(node.target) + self.write(' in ') + self.visit(node.iter) + if node.ifs: + for if_ in node.ifs: + self.write(' if ') + self.visit(if_) + + def visit_excepthandler(self, node): + self.newline() + self.write('except') + if node.type is not None: + self.write(' ') + self.visit(node.type) + if node.name is not None: + self.write(' as ') + self.visit(node.name) + self.write(':') + self.body(node.body) diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/ast.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/ast.py new file mode 100644 index 00000000000..65fd84dfe15 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/ast.py @@ -0,0 +1,178 @@ +# mako/ast.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + +"""utilities for analyzing expressions and blocks of Python +code, as well as generating Python from AST nodes""" + +from mako import exceptions, pyparser, compat +import re + +class PythonCode(object): + """represents information about a string containing Python code""" + def __init__(self, code, **exception_kwargs): + self.code = code + + # represents all identifiers which are assigned to at some point in + # the code + self.declared_identifiers = set() + + # represents all identifiers which are referenced before their + # assignment, if any + self.undeclared_identifiers = set() + + # note that an identifier can be in both the undeclared and declared + # lists. + + # using AST to parse instead of using code.co_varnames, + # code.co_names has several advantages: + # - we can locate an identifier as "undeclared" even if + # its declared later in the same block of code + # - AST is less likely to break with version changes + # (for example, the behavior of co_names changed a little bit + # in python version 2.5) + if isinstance(code, compat.string_types): + expr = pyparser.parse(code.lstrip(), "exec", **exception_kwargs) + else: + expr = code + + f = pyparser.FindIdentifiers(self, **exception_kwargs) + f.visit(expr) + +class ArgumentList(object): + """parses a fragment of code as a comma-separated list of expressions""" + def __init__(self, code, **exception_kwargs): + self.codeargs = [] + self.args = [] + self.declared_identifiers = set() + self.undeclared_identifiers = set() + if isinstance(code, compat.string_types): + if re.match(r"\S", code) and not re.match(r",\s*$", code): + # if theres text and no trailing comma, insure its parsed + # as a tuple by adding a trailing comma + code += "," + expr = pyparser.parse(code, "exec", **exception_kwargs) + else: + expr = code + + f = pyparser.FindTuple(self, PythonCode, **exception_kwargs) + f.visit(expr) + +class PythonFragment(PythonCode): + """extends PythonCode to provide identifier lookups in partial control + statements + + e.g. + for x in 5: + elif y==9: + except (MyException, e): + etc. + """ + def __init__(self, code, **exception_kwargs): + m = re.match(r'^(\w+)(?:\s+(.*?))?:\s*(#|$)', code.strip(), re.S) + if not m: + raise exceptions.CompileException( + "Fragment '%s' is not a partial control statement" % + code, **exception_kwargs) + if m.group(3): + code = code[:m.start(3)] + (keyword, expr) = m.group(1,2) + if keyword in ['for','if', 'while']: + code = code + "pass" + elif keyword == 'try': + code = code + "pass\nexcept:pass" + elif keyword == 'elif' or keyword == 'else': + code = "if False:pass\n" + code + "pass" + elif keyword == 'except': + code = "try:pass\n" + code + "pass" + elif keyword == 'with': + code = code + "pass" + else: + raise exceptions.CompileException( + "Unsupported control keyword: '%s'" % + keyword, **exception_kwargs) + super(PythonFragment, self).__init__(code, **exception_kwargs) + + +class FunctionDecl(object): + """function declaration""" + def __init__(self, code, allow_kwargs=True, **exception_kwargs): + self.code = code + expr = pyparser.parse(code, "exec", **exception_kwargs) + + f = pyparser.ParseFunc(self, **exception_kwargs) + f.visit(expr) + if not hasattr(self, 'funcname'): + raise exceptions.CompileException( + "Code '%s' is not a function declaration" % code, + **exception_kwargs) + if not allow_kwargs and self.kwargs: + raise exceptions.CompileException( + "'**%s' keyword argument not allowed here" % + self.kwargnames[-1], **exception_kwargs) + + def get_argument_expressions(self, as_call=False): + """Return the argument declarations of this FunctionDecl as a printable + list. + + By default the return value is appropriate for writing in a ``def``; + set `as_call` to true to build arguments to be passed to the function + instead (assuming locals with the same names as the arguments exist). + """ + + namedecls = [] + + # Build in reverse order, since defaults and slurpy args come last + argnames = self.argnames[::-1] + kwargnames = self.kwargnames[::-1] + defaults = self.defaults[::-1] + kwdefaults = self.kwdefaults[::-1] + + # Named arguments + if self.kwargs: + namedecls.append("**" + kwargnames.pop(0)) + + for name in kwargnames: + # Keyword-only arguments must always be used by name, so even if + # this is a call, print out `foo=foo` + if as_call: + namedecls.append("%s=%s" % (name, name)) + elif kwdefaults: + default = kwdefaults.pop(0) + if default is None: + # The AST always gives kwargs a default, since you can do + # `def foo(*, a=1, b, c=3)` + namedecls.append(name) + else: + namedecls.append("%s=%s" % ( + name, pyparser.ExpressionGenerator(default).value())) + else: + namedecls.append(name) + + # Positional arguments + if self.varargs: + namedecls.append("*" + argnames.pop(0)) + + for name in argnames: + if as_call or not defaults: + namedecls.append(name) + else: + default = defaults.pop(0) + namedecls.append("%s=%s" % ( + name, pyparser.ExpressionGenerator(default).value())) + + namedecls.reverse() + return namedecls + + @property + def allargnames(self): + return tuple(self.argnames) + tuple(self.kwargnames) + +class FunctionArgs(FunctionDecl): + """the argument portion of a function declaration""" + + def __init__(self, code, **kwargs): + super(FunctionArgs, self).__init__("def ANON(%s):pass" % code, + **kwargs) diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/cache.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/cache.py new file mode 100644 index 00000000000..c405c5171d7 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/cache.py @@ -0,0 +1,238 @@ +# mako/cache.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + +from mako import compat, util + +_cache_plugins = util.PluginLoader("mako.cache") + +register_plugin = _cache_plugins.register +register_plugin("beaker", "mako.ext.beaker_cache", "BeakerCacheImpl") + + +class Cache(object): + """Represents a data content cache made available to the module + space of a specific :class:`.Template` object. + + .. versionadded:: 0.6 + :class:`.Cache` by itself is mostly a + container for a :class:`.CacheImpl` object, which implements + a fixed API to provide caching services; specific subclasses exist to + implement different + caching strategies. Mako includes a backend that works with + the Beaker caching system. Beaker itself then supports + a number of backends (i.e. file, memory, memcached, etc.) + + The construction of a :class:`.Cache` is part of the mechanics + of a :class:`.Template`, and programmatic access to this + cache is typically via the :attr:`.Template.cache` attribute. + + """ + + impl = None + """Provide the :class:`.CacheImpl` in use by this :class:`.Cache`. + + This accessor allows a :class:`.CacheImpl` with additional + methods beyond that of :class:`.Cache` to be used programmatically. + + """ + + id = None + """Return the 'id' that identifies this cache. + + This is a value that should be globally unique to the + :class:`.Template` associated with this cache, and can + be used by a caching system to name a local container + for data specific to this template. + + """ + + starttime = None + """Epochal time value for when the owning :class:`.Template` was + first compiled. + + A cache implementation may wish to invalidate data earlier than + this timestamp; this has the effect of the cache for a specific + :class:`.Template` starting clean any time the :class:`.Template` + is recompiled, such as when the original template file changed on + the filesystem. + + """ + + def __init__(self, template, *args): + # check for a stale template calling the + # constructor + if isinstance(template, compat.string_types) and args: + return + self.template = template + self.id = template.module.__name__ + self.starttime = template.module._modified_time + self._def_regions = {} + self.impl = self._load_impl(self.template.cache_impl) + + def _load_impl(self, name): + return _cache_plugins.load(name)(self) + + def get_or_create(self, key, creation_function, **kw): + """Retrieve a value from the cache, using the given creation function + to generate a new value.""" + + return self._ctx_get_or_create(key, creation_function, None, **kw) + + def _ctx_get_or_create(self, key, creation_function, context, **kw): + """Retrieve a value from the cache, using the given creation function + to generate a new value.""" + + if not self.template.cache_enabled: + return creation_function() + + return self.impl.get_or_create( + key, + creation_function, + **self._get_cache_kw(kw, context)) + + def set(self, key, value, **kw): + """Place a value in the cache. + + :param key: the value's key. + :param value: the value. + :param \**kw: cache configuration arguments. + + """ + + self.impl.set(key, value, **self._get_cache_kw(kw, None)) + + put = set + """A synonym for :meth:`.Cache.set`. + + This is here for backwards compatibility. + + """ + + def get(self, key, **kw): + """Retrieve a value from the cache. + + :param key: the value's key. + :param \**kw: cache configuration arguments. The + backend is configured using these arguments upon first request. + Subsequent requests that use the same series of configuration + values will use that same backend. + + """ + return self.impl.get(key, **self._get_cache_kw(kw, None)) + + def invalidate(self, key, **kw): + """Invalidate a value in the cache. + + :param key: the value's key. + :param \**kw: cache configuration arguments. The + backend is configured using these arguments upon first request. + Subsequent requests that use the same series of configuration + values will use that same backend. + + """ + self.impl.invalidate(key, **self._get_cache_kw(kw, None)) + + def invalidate_body(self): + """Invalidate the cached content of the "body" method for this + template. + + """ + self.invalidate('render_body', __M_defname='render_body') + + def invalidate_def(self, name): + """Invalidate the cached content of a particular ``<%def>`` within this + template. + + """ + + self.invalidate('render_%s' % name, __M_defname='render_%s' % name) + + def invalidate_closure(self, name): + """Invalidate a nested ``<%def>`` within this template. + + Caching of nested defs is a blunt tool as there is no + management of scope -- nested defs that use cache tags + need to have names unique of all other nested defs in the + template, else their content will be overwritten by + each other. + + """ + + self.invalidate(name, __M_defname=name) + + def _get_cache_kw(self, kw, context): + defname = kw.pop('__M_defname', None) + if not defname: + tmpl_kw = self.template.cache_args.copy() + tmpl_kw.update(kw) + elif defname in self._def_regions: + tmpl_kw = self._def_regions[defname] + else: + tmpl_kw = self.template.cache_args.copy() + tmpl_kw.update(kw) + self._def_regions[defname] = tmpl_kw + if context and self.impl.pass_context: + tmpl_kw = tmpl_kw.copy() + tmpl_kw.setdefault('context', context) + return tmpl_kw + + +class CacheImpl(object): + """Provide a cache implementation for use by :class:`.Cache`.""" + + def __init__(self, cache): + self.cache = cache + + pass_context = False + """If ``True``, the :class:`.Context` will be passed to + :meth:`get_or_create <.CacheImpl.get_or_create>` as the name ``'context'``. + """ + + def get_or_create(self, key, creation_function, **kw): + """Retrieve a value from the cache, using the given creation function + to generate a new value. + + This function *must* return a value, either from + the cache, or via the given creation function. + If the creation function is called, the newly + created value should be populated into the cache + under the given key before being returned. + + :param key: the value's key. + :param creation_function: function that when called generates + a new value. + :param \**kw: cache configuration arguments. + + """ + raise NotImplementedError() + + def set(self, key, value, **kw): + """Place a value in the cache. + + :param key: the value's key. + :param value: the value. + :param \**kw: cache configuration arguments. + + """ + raise NotImplementedError() + + def get(self, key, **kw): + """Retrieve a value from the cache. + + :param key: the value's key. + :param \**kw: cache configuration arguments. + + """ + raise NotImplementedError() + + def invalidate(self, key, **kw): + """Invalidate a value in the cache. + + :param key: the value's key. + :param \**kw: cache configuration arguments. + + """ + raise NotImplementedError() diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/cmd.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/cmd.py new file mode 100644 index 00000000000..1a9ca56637c --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/cmd.py @@ -0,0 +1,62 @@ +# mako/cmd.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php +from argparse import ArgumentParser +from os.path import isfile, dirname +import sys +from mako.template import Template +from mako.lookup import TemplateLookup +from mako import exceptions + +def varsplit(var): + if "=" not in var: + return (var, "") + return var.split("=", 1) + +def _exit(): + sys.stderr.write(exceptions.text_error_template().render()) + sys.exit(1) + +def cmdline(argv=None): + + parser = ArgumentParser("usage: %prog [FILENAME]") + parser.add_argument("--var", default=[], action="append", + help="variable (can be used multiple times, use name=value)") + parser.add_argument("--template-dir", default=[], action="append", + help="Directory to use for template lookup (multiple " + "directories may be provided). If not given then if the " + "template is read from stdin, the value defaults to be " + "the current directory, otherwise it defaults to be the " + "parent directory of the file provided.") + parser.add_argument('input', nargs='?', default='-') + + options = parser.parse_args(argv) + if options.input == '-': + lookup_dirs = options.template_dir or ["."] + lookup = TemplateLookup(lookup_dirs) + try: + template = Template(sys.stdin.read(), lookup=lookup) + except: + _exit() + else: + filename = options.input + if not isfile(filename): + raise SystemExit("error: can't find %s" % filename) + lookup_dirs = options.template_dir or [dirname(filename)] + lookup = TemplateLookup(lookup_dirs) + try: + template = Template(filename=filename, lookup=lookup) + except: + _exit() + + kw = dict([varsplit(var) for var in options.var]) + try: + print(template.render(**kw)) + except: + _exit() + + +if __name__ == "__main__": + cmdline() diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/codegen.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/codegen.py new file mode 100644 index 00000000000..4b0bda86731 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/codegen.py @@ -0,0 +1,1237 @@ +# mako/codegen.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + +"""provides functionality for rendering a parsetree constructing into module +source code.""" + +import time +import re +from mako.pygen import PythonPrinter +from mako import util, ast, parsetree, filters, exceptions +from mako import compat + + +MAGIC_NUMBER = 10 + +# names which are hardwired into the +# template and are not accessed via the +# context itself +RESERVED_NAMES = set(['context', 'loop', 'UNDEFINED']) + +def compile(node, + uri, + filename=None, + default_filters=None, + buffer_filters=None, + imports=None, + future_imports=None, + source_encoding=None, + generate_magic_comment=True, + disable_unicode=False, + strict_undefined=False, + enable_loop=True, + reserved_names=frozenset()): + + """Generate module source code given a parsetree node, + uri, and optional source filename""" + + # if on Py2K, push the "source_encoding" string to be + # a bytestring itself, as we will be embedding it into + # the generated source and we don't want to coerce the + # result into a unicode object, in "disable_unicode" mode + if not compat.py3k and isinstance(source_encoding, compat.text_type): + source_encoding = source_encoding.encode(source_encoding) + + + buf = util.FastEncodingBuffer() + + printer = PythonPrinter(buf) + _GenerateRenderMethod(printer, + _CompileContext(uri, + filename, + default_filters, + buffer_filters, + imports, + future_imports, + source_encoding, + generate_magic_comment, + disable_unicode, + strict_undefined, + enable_loop, + reserved_names), + node) + return buf.getvalue() + +class _CompileContext(object): + def __init__(self, + uri, + filename, + default_filters, + buffer_filters, + imports, + future_imports, + source_encoding, + generate_magic_comment, + disable_unicode, + strict_undefined, + enable_loop, + reserved_names): + self.uri = uri + self.filename = filename + self.default_filters = default_filters + self.buffer_filters = buffer_filters + self.imports = imports + self.future_imports = future_imports + self.source_encoding = source_encoding + self.generate_magic_comment = generate_magic_comment + self.disable_unicode = disable_unicode + self.strict_undefined = strict_undefined + self.enable_loop = enable_loop + self.reserved_names = reserved_names + +class _GenerateRenderMethod(object): + """A template visitor object which generates the + full module source for a template. + + """ + def __init__(self, printer, compiler, node): + self.printer = printer + self.compiler = compiler + self.node = node + self.identifier_stack = [None] + self.in_def = isinstance(node, (parsetree.DefTag, parsetree.BlockTag)) + + if self.in_def: + name = "render_%s" % node.funcname + args = node.get_argument_expressions() + filtered = len(node.filter_args.args) > 0 + buffered = eval(node.attributes.get('buffered', 'False')) + cached = eval(node.attributes.get('cached', 'False')) + defs = None + pagetag = None + if node.is_block and not node.is_anonymous: + args += ['**pageargs'] + else: + defs = self.write_toplevel() + pagetag = self.compiler.pagetag + name = "render_body" + if pagetag is not None: + args = pagetag.body_decl.get_argument_expressions() + if not pagetag.body_decl.kwargs: + args += ['**pageargs'] + cached = eval(pagetag.attributes.get('cached', 'False')) + self.compiler.enable_loop = self.compiler.enable_loop or eval( + pagetag.attributes.get( + 'enable_loop', 'False') + ) + else: + args = ['**pageargs'] + cached = False + buffered = filtered = False + if args is None: + args = ['context'] + else: + args = [a for a in ['context'] + args] + + self.write_render_callable( + pagetag or node, + name, args, + buffered, filtered, cached) + + if defs is not None: + for node in defs: + _GenerateRenderMethod(printer, compiler, node) + + if not self.in_def: + self.write_metadata_struct() + + def write_metadata_struct(self): + self.printer.source_map[self.printer.lineno] = \ + max(self.printer.source_map) + struct = { + "filename": self.compiler.filename, + "uri": self.compiler.uri, + "source_encoding": self.compiler.source_encoding, + "line_map": self.printer.source_map, + } + self.printer.writelines( + '"""', + '__M_BEGIN_METADATA', + compat.json.dumps(struct), + '__M_END_METADATA\n' + '"""' + ) + + @property + def identifiers(self): + return self.identifier_stack[-1] + + def write_toplevel(self): + """Traverse a template structure for module-level directives and + generate the start of module-level code. + + """ + inherit = [] + namespaces = {} + module_code = [] + + self.compiler.pagetag = None + + class FindTopLevel(object): + def visitInheritTag(s, node): + inherit.append(node) + def visitNamespaceTag(s, node): + namespaces[node.name] = node + def visitPageTag(s, node): + self.compiler.pagetag = node + def visitCode(s, node): + if node.ismodule: + module_code.append(node) + + f = FindTopLevel() + for n in self.node.nodes: + n.accept_visitor(f) + + self.compiler.namespaces = namespaces + + module_ident = set() + for n in module_code: + module_ident = module_ident.union(n.declared_identifiers()) + + module_identifiers = _Identifiers(self.compiler) + module_identifiers.declared = module_ident + + # module-level names, python code + if self.compiler.generate_magic_comment and \ + self.compiler.source_encoding: + self.printer.writeline("# -*- coding:%s -*-" % + self.compiler.source_encoding) + + if self.compiler.future_imports: + self.printer.writeline("from __future__ import %s" % + (", ".join(self.compiler.future_imports),)) + self.printer.writeline("from mako import runtime, filters, cache") + self.printer.writeline("UNDEFINED = runtime.UNDEFINED") + self.printer.writeline("__M_dict_builtin = dict") + self.printer.writeline("__M_locals_builtin = locals") + self.printer.writeline("_magic_number = %r" % MAGIC_NUMBER) + self.printer.writeline("_modified_time = %r" % time.time()) + self.printer.writeline("_enable_loop = %r" % self.compiler.enable_loop) + self.printer.writeline( + "_template_filename = %r" % self.compiler.filename) + self.printer.writeline("_template_uri = %r" % self.compiler.uri) + self.printer.writeline( + "_source_encoding = %r" % self.compiler.source_encoding) + if self.compiler.imports: + buf = '' + for imp in self.compiler.imports: + buf += imp + "\n" + self.printer.writeline(imp) + impcode = ast.PythonCode( + buf, + source='', lineno=0, + pos=0, + filename='template defined imports') + else: + impcode = None + + main_identifiers = module_identifiers.branch(self.node) + module_identifiers.topleveldefs = \ + module_identifiers.topleveldefs.\ + union(main_identifiers.topleveldefs) + module_identifiers.declared.add("UNDEFINED") + if impcode: + module_identifiers.declared.update(impcode.declared_identifiers) + + self.compiler.identifiers = module_identifiers + self.printer.writeline("_exports = %r" % + [n.name for n in + main_identifiers.topleveldefs.values()] + ) + self.printer.write_blanks(2) + + if len(module_code): + self.write_module_code(module_code) + + if len(inherit): + self.write_namespaces(namespaces) + self.write_inherit(inherit[-1]) + elif len(namespaces): + self.write_namespaces(namespaces) + + return list(main_identifiers.topleveldefs.values()) + + def write_render_callable(self, node, name, args, buffered, filtered, + cached): + """write a top-level render callable. + + this could be the main render() method or that of a top-level def.""" + + if self.in_def: + decorator = node.decorator + if decorator: + self.printer.writeline( + "@runtime._decorate_toplevel(%s)" % decorator) + + self.printer.start_source(node.lineno) + self.printer.writelines( + "def %s(%s):" % (name, ','.join(args)), + # push new frame, assign current frame to __M_caller + "__M_caller = context.caller_stack._push_frame()", + "try:" + ) + if buffered or filtered or cached: + self.printer.writeline("context._push_buffer()") + + self.identifier_stack.append( + self.compiler.identifiers.branch(self.node)) + if (not self.in_def or self.node.is_block) and '**pageargs' in args: + self.identifier_stack[-1].argument_declared.add('pageargs') + + if not self.in_def and ( + len(self.identifiers.locally_assigned) > 0 or + len(self.identifiers.argument_declared) > 0 + ): + self.printer.writeline("__M_locals = __M_dict_builtin(%s)" % + ','.join([ + "%s=%s" % (x, x) for x in + self.identifiers.argument_declared + ])) + + self.write_variable_declares(self.identifiers, toplevel=True) + + for n in self.node.nodes: + n.accept_visitor(self) + + self.write_def_finish(self.node, buffered, filtered, cached) + self.printer.writeline(None) + self.printer.write_blanks(2) + if cached: + self.write_cache_decorator( + node, name, + args, buffered, + self.identifiers, toplevel=True) + + def write_module_code(self, module_code): + """write module-level template code, i.e. that which + is enclosed in <%! %> tags in the template.""" + for n in module_code: + self.printer.start_source(n.lineno) + self.printer.write_indented_block(n.text) + + def write_inherit(self, node): + """write the module-level inheritance-determination callable.""" + + self.printer.writelines( + "def _mako_inherit(template, context):", + "_mako_generate_namespaces(context)", + "return runtime._inherit_from(context, %s, _template_uri)" % + (node.parsed_attributes['file']), + None + ) + + def write_namespaces(self, namespaces): + """write the module-level namespace-generating callable.""" + self.printer.writelines( + "def _mako_get_namespace(context, name):", + "try:", + "return context.namespaces[(__name__, name)]", + "except KeyError:", + "_mako_generate_namespaces(context)", + "return context.namespaces[(__name__, name)]", + None, None + ) + self.printer.writeline("def _mako_generate_namespaces(context):") + + + for node in namespaces.values(): + if 'import' in node.attributes: + self.compiler.has_ns_imports = True + self.printer.start_source(node.lineno) + if len(node.nodes): + self.printer.writeline("def make_namespace():") + export = [] + identifiers = self.compiler.identifiers.branch(node) + self.in_def = True + class NSDefVisitor(object): + def visitDefTag(s, node): + s.visitDefOrBase(node) + + def visitBlockTag(s, node): + s.visitDefOrBase(node) + + def visitDefOrBase(s, node): + if node.is_anonymous: + raise exceptions.CompileException( + "Can't put anonymous blocks inside " + "<%namespace>", + **node.exception_kwargs + ) + self.write_inline_def(node, identifiers, nested=False) + export.append(node.funcname) + vis = NSDefVisitor() + for n in node.nodes: + n.accept_visitor(vis) + self.printer.writeline("return [%s]" % (','.join(export))) + self.printer.writeline(None) + self.in_def = False + callable_name = "make_namespace()" + else: + callable_name = "None" + + if 'file' in node.parsed_attributes: + self.printer.writeline( + "ns = runtime.TemplateNamespace(%r," + " context._clean_inheritance_tokens()," + " templateuri=%s, callables=%s, " + " calling_uri=_template_uri)" % + ( + node.name, + node.parsed_attributes.get('file', 'None'), + callable_name, + ) + ) + elif 'module' in node.parsed_attributes: + self.printer.writeline( + "ns = runtime.ModuleNamespace(%r," + " context._clean_inheritance_tokens()," + " callables=%s, calling_uri=_template_uri," + " module=%s)" % + ( + node.name, + callable_name, + node.parsed_attributes.get( + 'module', 'None') + ) + ) + else: + self.printer.writeline( + "ns = runtime.Namespace(%r," + " context._clean_inheritance_tokens()," + " callables=%s, calling_uri=_template_uri)" % + ( + node.name, + callable_name, + ) + ) + if eval(node.attributes.get('inheritable', "False")): + self.printer.writeline("context['self'].%s = ns" % (node.name)) + + self.printer.writeline( + "context.namespaces[(__name__, %s)] = ns" % repr(node.name)) + self.printer.write_blanks(1) + if not len(namespaces): + self.printer.writeline("pass") + self.printer.writeline(None) + + def write_variable_declares(self, identifiers, toplevel=False, limit=None): + """write variable declarations at the top of a function. + + the variable declarations are in the form of callable + definitions for defs and/or name lookup within the + function's context argument. the names declared are based + on the names that are referenced in the function body, + which don't otherwise have any explicit assignment + operation. names that are assigned within the body are + assumed to be locally-scoped variables and are not + separately declared. + + for def callable definitions, if the def is a top-level + callable then a 'stub' callable is generated which wraps + the current Context into a closure. if the def is not + top-level, it is fully rendered as a local closure. + + """ + + # collection of all defs available to us in this scope + comp_idents = dict([(c.funcname, c) for c in identifiers.defs]) + to_write = set() + + # write "context.get()" for all variables we are going to + # need that arent in the namespace yet + to_write = to_write.union(identifiers.undeclared) + + # write closure functions for closures that we define + # right here + to_write = to_write.union( + [c.funcname for c in identifiers.closuredefs.values()]) + + # remove identifiers that are declared in the argument + # signature of the callable + to_write = to_write.difference(identifiers.argument_declared) + + # remove identifiers that we are going to assign to. + # in this way we mimic Python's behavior, + # i.e. assignment to a variable within a block + # means that variable is now a "locally declared" var, + # which cannot be referenced beforehand. + to_write = to_write.difference(identifiers.locally_declared) + + if self.compiler.enable_loop: + has_loop = "loop" in to_write + to_write.discard("loop") + else: + has_loop = False + + # if a limiting set was sent, constraint to those items in that list + # (this is used for the caching decorator) + if limit is not None: + to_write = to_write.intersection(limit) + + if toplevel and getattr(self.compiler, 'has_ns_imports', False): + self.printer.writeline("_import_ns = {}") + self.compiler.has_imports = True + for ident, ns in self.compiler.namespaces.items(): + if 'import' in ns.attributes: + self.printer.writeline( + "_mako_get_namespace(context, %r)." + "_populate(_import_ns, %r)" % + ( + ident, + re.split(r'\s*,\s*', ns.attributes['import']) + )) + + if has_loop: + self.printer.writeline( + 'loop = __M_loop = runtime.LoopStack()' + ) + + for ident in to_write: + if ident in comp_idents: + comp = comp_idents[ident] + if comp.is_block: + if not comp.is_anonymous: + self.write_def_decl(comp, identifiers) + else: + self.write_inline_def(comp, identifiers, nested=True) + else: + if comp.is_root(): + self.write_def_decl(comp, identifiers) + else: + self.write_inline_def(comp, identifiers, nested=True) + + elif ident in self.compiler.namespaces: + self.printer.writeline( + "%s = _mako_get_namespace(context, %r)" % + (ident, ident) + ) + else: + if getattr(self.compiler, 'has_ns_imports', False): + if self.compiler.strict_undefined: + self.printer.writelines( + "%s = _import_ns.get(%r, UNDEFINED)" % + (ident, ident), + "if %s is UNDEFINED:" % ident, + "try:", + "%s = context[%r]" % (ident, ident), + "except KeyError:", + "raise NameError(\"'%s' is not defined\")" % + ident, + None, None + ) + else: + self.printer.writeline( + "%s = _import_ns.get(%r, context.get(%r, UNDEFINED))" % + (ident, ident, ident)) + else: + if self.compiler.strict_undefined: + self.printer.writelines( + "try:", + "%s = context[%r]" % (ident, ident), + "except KeyError:", + "raise NameError(\"'%s' is not defined\")" % + ident, + None + ) + else: + self.printer.writeline( + "%s = context.get(%r, UNDEFINED)" % (ident, ident) + ) + + self.printer.writeline("__M_writer = context.writer()") + + def write_def_decl(self, node, identifiers): + """write a locally-available callable referencing a top-level def""" + funcname = node.funcname + namedecls = node.get_argument_expressions() + nameargs = node.get_argument_expressions(as_call=True) + + if not self.in_def and ( + len(self.identifiers.locally_assigned) > 0 or + len(self.identifiers.argument_declared) > 0): + nameargs.insert(0, 'context._locals(__M_locals)') + else: + nameargs.insert(0, 'context') + self.printer.writeline("def %s(%s):" % (funcname, ",".join(namedecls))) + self.printer.writeline( + "return render_%s(%s)" % (funcname, ",".join(nameargs))) + self.printer.writeline(None) + + def write_inline_def(self, node, identifiers, nested): + """write a locally-available def callable inside an enclosing def.""" + + namedecls = node.get_argument_expressions() + + decorator = node.decorator + if decorator: + self.printer.writeline( + "@runtime._decorate_inline(context, %s)" % decorator) + self.printer.writeline( + "def %s(%s):" % (node.funcname, ",".join(namedecls))) + filtered = len(node.filter_args.args) > 0 + buffered = eval(node.attributes.get('buffered', 'False')) + cached = eval(node.attributes.get('cached', 'False')) + self.printer.writelines( + # push new frame, assign current frame to __M_caller + "__M_caller = context.caller_stack._push_frame()", + "try:" + ) + if buffered or filtered or cached: + self.printer.writelines( + "context._push_buffer()", + ) + + identifiers = identifiers.branch(node, nested=nested) + + self.write_variable_declares(identifiers) + + self.identifier_stack.append(identifiers) + for n in node.nodes: + n.accept_visitor(self) + self.identifier_stack.pop() + + self.write_def_finish(node, buffered, filtered, cached) + self.printer.writeline(None) + if cached: + self.write_cache_decorator(node, node.funcname, + namedecls, False, identifiers, + inline=True, toplevel=False) + + def write_def_finish(self, node, buffered, filtered, cached, + callstack=True): + """write the end section of a rendering function, either outermost or + inline. + + this takes into account if the rendering function was filtered, + buffered, etc. and closes the corresponding try: block if any, and + writes code to retrieve captured content, apply filters, send proper + return value.""" + + if not buffered and not cached and not filtered: + self.printer.writeline("return ''") + if callstack: + self.printer.writelines( + "finally:", + "context.caller_stack._pop_frame()", + None + ) + + if buffered or filtered or cached: + if buffered or cached: + # in a caching scenario, don't try to get a writer + # from the context after popping; assume the caching + # implemenation might be using a context with no + # extra buffers + self.printer.writelines( + "finally:", + "__M_buf = context._pop_buffer()" + ) + else: + self.printer.writelines( + "finally:", + "__M_buf, __M_writer = context._pop_buffer_and_writer()" + ) + + if callstack: + self.printer.writeline("context.caller_stack._pop_frame()") + + s = "__M_buf.getvalue()" + if filtered: + s = self.create_filter_callable(node.filter_args.args, s, + False) + self.printer.writeline(None) + if buffered and not cached: + s = self.create_filter_callable(self.compiler.buffer_filters, + s, False) + if buffered or cached: + self.printer.writeline("return %s" % s) + else: + self.printer.writelines( + "__M_writer(%s)" % s, + "return ''" + ) + + def write_cache_decorator(self, node_or_pagetag, name, + args, buffered, identifiers, + inline=False, toplevel=False): + """write a post-function decorator to replace a rendering + callable with a cached version of itself.""" + + self.printer.writeline("__M_%s = %s" % (name, name)) + cachekey = node_or_pagetag.parsed_attributes.get('cache_key', + repr(name)) + + cache_args = {} + if self.compiler.pagetag is not None: + cache_args.update( + ( + pa[6:], + self.compiler.pagetag.parsed_attributes[pa] + ) + for pa in self.compiler.pagetag.parsed_attributes + if pa.startswith('cache_') and pa != 'cache_key' + ) + cache_args.update( + ( + pa[6:], + node_or_pagetag.parsed_attributes[pa] + ) for pa in node_or_pagetag.parsed_attributes + if pa.startswith('cache_') and pa != 'cache_key' + ) + if 'timeout' in cache_args: + cache_args['timeout'] = int(eval(cache_args['timeout'])) + + self.printer.writeline("def %s(%s):" % (name, ','.join(args))) + + # form "arg1, arg2, arg3=arg3, arg4=arg4", etc. + pass_args = [ + "%s=%s" % ((a.split('=')[0],) * 2) if '=' in a else a + for a in args + ] + + self.write_variable_declares( + identifiers, + toplevel=toplevel, + limit=node_or_pagetag.undeclared_identifiers() + ) + if buffered: + s = "context.get('local')."\ + "cache._ctx_get_or_create("\ + "%s, lambda:__M_%s(%s), context, %s__M_defname=%r)" % ( + cachekey, name, ','.join(pass_args), + ''.join(["%s=%s, " % (k, v) + for k, v in cache_args.items()]), + name + ) + # apply buffer_filters + s = self.create_filter_callable(self.compiler.buffer_filters, s, + False) + self.printer.writelines("return " + s, None) + else: + self.printer.writelines( + "__M_writer(context.get('local')." + "cache._ctx_get_or_create(" + "%s, lambda:__M_%s(%s), context, %s__M_defname=%r))" % + ( + cachekey, name, ','.join(pass_args), + ''.join(["%s=%s, " % (k, v) + for k, v in cache_args.items()]), + name, + ), + "return ''", + None + ) + + def create_filter_callable(self, args, target, is_expression): + """write a filter-applying expression based on the filters + present in the given filter names, adjusting for the global + 'default' filter aliases as needed.""" + + def locate_encode(name): + if re.match(r'decode\..+', name): + return "filters." + name + elif self.compiler.disable_unicode: + return filters.NON_UNICODE_ESCAPES.get(name, name) + else: + return filters.DEFAULT_ESCAPES.get(name, name) + + if 'n' not in args: + if is_expression: + if self.compiler.pagetag: + args = self.compiler.pagetag.filter_args.args + args + if self.compiler.default_filters: + args = self.compiler.default_filters + args + for e in args: + # if filter given as a function, get just the identifier portion + if e == 'n': + continue + m = re.match(r'(.+?)(\(.*\))', e) + if m: + ident, fargs = m.group(1, 2) + f = locate_encode(ident) + e = f + fargs + else: + e = locate_encode(e) + assert e is not None + target = "%s(%s)" % (e, target) + return target + + def visitExpression(self, node): + self.printer.start_source(node.lineno) + if len(node.escapes) or \ + ( + self.compiler.pagetag is not None and + len(self.compiler.pagetag.filter_args.args) + ) or \ + len(self.compiler.default_filters): + + s = self.create_filter_callable(node.escapes_code.args, + "%s" % node.text, True) + self.printer.writeline("__M_writer(%s)" % s) + else: + self.printer.writeline("__M_writer(%s)" % node.text) + + def visitControlLine(self, node): + if node.isend: + self.printer.writeline(None) + if node.has_loop_context: + self.printer.writeline('finally:') + self.printer.writeline("loop = __M_loop._exit()") + self.printer.writeline(None) + else: + self.printer.start_source(node.lineno) + if self.compiler.enable_loop and node.keyword == 'for': + text = mangle_mako_loop(node, self.printer) + else: + text = node.text + self.printer.writeline(text) + children = node.get_children() + # this covers the three situations where we want to insert a pass: + # 1) a ternary control line with no children, + # 2) a primary control line with nothing but its own ternary + # and end control lines, and + # 3) any control line with no content other than comments + if not children or ( + compat.all(isinstance(c, (parsetree.Comment, + parsetree.ControlLine)) + for c in children) and + compat.all((node.is_ternary(c.keyword) or c.isend) + for c in children + if isinstance(c, parsetree.ControlLine))): + self.printer.writeline("pass") + + def visitText(self, node): + self.printer.start_source(node.lineno) + self.printer.writeline("__M_writer(%s)" % repr(node.content)) + + def visitTextTag(self, node): + filtered = len(node.filter_args.args) > 0 + if filtered: + self.printer.writelines( + "__M_writer = context._push_writer()", + "try:", + ) + for n in node.nodes: + n.accept_visitor(self) + if filtered: + self.printer.writelines( + "finally:", + "__M_buf, __M_writer = context._pop_buffer_and_writer()", + "__M_writer(%s)" % + self.create_filter_callable( + node.filter_args.args, + "__M_buf.getvalue()", + False), + None + ) + + def visitCode(self, node): + if not node.ismodule: + self.printer.start_source(node.lineno) + self.printer.write_indented_block(node.text) + + if not self.in_def and len(self.identifiers.locally_assigned) > 0: + # if we are the "template" def, fudge locally + # declared/modified variables into the "__M_locals" dictionary, + # which is used for def calls within the same template, + # to simulate "enclosing scope" + self.printer.writeline( + '__M_locals_builtin_stored = __M_locals_builtin()') + self.printer.writeline( + '__M_locals.update(__M_dict_builtin([(__M_key,' + ' __M_locals_builtin_stored[__M_key]) for __M_key in' + ' [%s] if __M_key in __M_locals_builtin_stored]))' % + ','.join([repr(x) for x in node.declared_identifiers()])) + + def visitIncludeTag(self, node): + self.printer.start_source(node.lineno) + args = node.attributes.get('args') + if args: + self.printer.writeline( + "runtime._include_file(context, %s, _template_uri, %s)" % + (node.parsed_attributes['file'], args)) + else: + self.printer.writeline( + "runtime._include_file(context, %s, _template_uri)" % + (node.parsed_attributes['file'])) + + def visitNamespaceTag(self, node): + pass + + def visitDefTag(self, node): + pass + + def visitBlockTag(self, node): + if node.is_anonymous: + self.printer.writeline("%s()" % node.funcname) + else: + nameargs = node.get_argument_expressions(as_call=True) + nameargs += ['**pageargs'] + self.printer.writeline("if 'parent' not in context._data or " + "not hasattr(context._data['parent'], '%s'):" + % node.funcname) + self.printer.writeline( + "context['self'].%s(%s)" % (node.funcname, ",".join(nameargs))) + self.printer.writeline("\n") + + def visitCallNamespaceTag(self, node): + # TODO: we can put namespace-specific checks here, such + # as ensure the given namespace will be imported, + # pre-import the namespace, etc. + self.visitCallTag(node) + + def visitCallTag(self, node): + self.printer.writeline("def ccall(caller):") + export = ['body'] + callable_identifiers = self.identifiers.branch(node, nested=True) + body_identifiers = callable_identifiers.branch(node, nested=False) + # we want the 'caller' passed to ccall to be used + # for the body() function, but for other non-body() + # <%def>s within <%call> we want the current caller + # off the call stack (if any) + body_identifiers.add_declared('caller') + + self.identifier_stack.append(body_identifiers) + class DefVisitor(object): + def visitDefTag(s, node): + s.visitDefOrBase(node) + + def visitBlockTag(s, node): + s.visitDefOrBase(node) + + def visitDefOrBase(s, node): + self.write_inline_def(node, callable_identifiers, nested=False) + if not node.is_anonymous: + export.append(node.funcname) + # remove defs that are within the <%call> from the + # "closuredefs" defined in the body, so they dont render twice + if node.funcname in body_identifiers.closuredefs: + del body_identifiers.closuredefs[node.funcname] + + vis = DefVisitor() + for n in node.nodes: + n.accept_visitor(vis) + self.identifier_stack.pop() + + bodyargs = node.body_decl.get_argument_expressions() + self.printer.writeline("def body(%s):" % ','.join(bodyargs)) + + # TODO: figure out best way to specify + # buffering/nonbuffering (at call time would be better) + buffered = False + if buffered: + self.printer.writelines( + "context._push_buffer()", + "try:" + ) + self.write_variable_declares(body_identifiers) + self.identifier_stack.append(body_identifiers) + + for n in node.nodes: + n.accept_visitor(self) + self.identifier_stack.pop() + + self.write_def_finish(node, buffered, False, False, callstack=False) + self.printer.writelines( + None, + "return [%s]" % (','.join(export)), + None + ) + + self.printer.writelines( + # push on caller for nested call + "context.caller_stack.nextcaller = " + "runtime.Namespace('caller', context, " + "callables=ccall(__M_caller))", + "try:") + self.printer.start_source(node.lineno) + self.printer.writelines( + "__M_writer(%s)" % self.create_filter_callable( + [], node.expression, True), + "finally:", + "context.caller_stack.nextcaller = None", + None + ) + +class _Identifiers(object): + """tracks the status of identifier names as template code is rendered.""" + + def __init__(self, compiler, node=None, parent=None, nested=False): + if parent is not None: + # if we are the branch created in write_namespaces(), + # we don't share any context from the main body(). + if isinstance(node, parsetree.NamespaceTag): + self.declared = set() + self.topleveldefs = util.SetLikeDict() + else: + # things that have already been declared + # in an enclosing namespace (i.e. names we can just use) + self.declared = set(parent.declared).\ + union([c.name for c in parent.closuredefs.values()]).\ + union(parent.locally_declared).\ + union(parent.argument_declared) + + # if these identifiers correspond to a "nested" + # scope, it means whatever the parent identifiers + # had as undeclared will have been declared by that parent, + # and therefore we have them in our scope. + if nested: + self.declared = self.declared.union(parent.undeclared) + + # top level defs that are available + self.topleveldefs = util.SetLikeDict(**parent.topleveldefs) + else: + self.declared = set() + self.topleveldefs = util.SetLikeDict() + + self.compiler = compiler + + # things within this level that are referenced before they + # are declared (e.g. assigned to) + self.undeclared = set() + + # things that are declared locally. some of these things + # could be in the "undeclared" list as well if they are + # referenced before declared + self.locally_declared = set() + + # assignments made in explicit python blocks. + # these will be propagated to + # the context of local def calls. + self.locally_assigned = set() + + # things that are declared in the argument + # signature of the def callable + self.argument_declared = set() + + # closure defs that are defined in this level + self.closuredefs = util.SetLikeDict() + + self.node = node + + if node is not None: + node.accept_visitor(self) + + illegal_names = self.compiler.reserved_names.intersection( + self.locally_declared) + if illegal_names: + raise exceptions.NameConflictError( + "Reserved words declared in template: %s" % + ", ".join(illegal_names)) + + + def branch(self, node, **kwargs): + """create a new Identifiers for a new Node, with + this Identifiers as the parent.""" + + return _Identifiers(self.compiler, node, self, **kwargs) + + @property + def defs(self): + return set(self.topleveldefs.union(self.closuredefs).values()) + + def __repr__(self): + return "Identifiers(declared=%r, locally_declared=%r, "\ + "undeclared=%r, topleveldefs=%r, closuredefs=%r, "\ + "argumentdeclared=%r)" %\ + ( + list(self.declared), + list(self.locally_declared), + list(self.undeclared), + [c.name for c in self.topleveldefs.values()], + [c.name for c in self.closuredefs.values()], + self.argument_declared) + + def check_declared(self, node): + """update the state of this Identifiers with the undeclared + and declared identifiers of the given node.""" + + for ident in node.undeclared_identifiers(): + if ident != 'context' and\ + ident not in self.declared.union(self.locally_declared): + self.undeclared.add(ident) + for ident in node.declared_identifiers(): + self.locally_declared.add(ident) + + def add_declared(self, ident): + self.declared.add(ident) + if ident in self.undeclared: + self.undeclared.remove(ident) + + def visitExpression(self, node): + self.check_declared(node) + + def visitControlLine(self, node): + self.check_declared(node) + + def visitCode(self, node): + if not node.ismodule: + self.check_declared(node) + self.locally_assigned = self.locally_assigned.union( + node.declared_identifiers()) + + def visitNamespaceTag(self, node): + # only traverse into the sub-elements of a + # <%namespace> tag if we are the branch created in + # write_namespaces() + if self.node is node: + for n in node.nodes: + n.accept_visitor(self) + + def _check_name_exists(self, collection, node): + existing = collection.get(node.funcname) + collection[node.funcname] = node + if existing is not None and \ + existing is not node and \ + (node.is_block or existing.is_block): + raise exceptions.CompileException( + "%%def or %%block named '%s' already " + "exists in this template." % + node.funcname, **node.exception_kwargs) + + def visitDefTag(self, node): + if node.is_root() and not node.is_anonymous: + self._check_name_exists(self.topleveldefs, node) + elif node is not self.node: + self._check_name_exists(self.closuredefs, node) + + for ident in node.undeclared_identifiers(): + if ident != 'context' and \ + ident not in self.declared.union(self.locally_declared): + self.undeclared.add(ident) + + # visit defs only one level deep + if node is self.node: + for ident in node.declared_identifiers(): + self.argument_declared.add(ident) + + for n in node.nodes: + n.accept_visitor(self) + + def visitBlockTag(self, node): + if node is not self.node and not node.is_anonymous: + + if isinstance(self.node, parsetree.DefTag): + raise exceptions.CompileException( + "Named block '%s' not allowed inside of def '%s'" + % (node.name, self.node.name), **node.exception_kwargs) + elif isinstance(self.node, + (parsetree.CallTag, parsetree.CallNamespaceTag)): + raise exceptions.CompileException( + "Named block '%s' not allowed inside of <%%call> tag" + % (node.name, ), **node.exception_kwargs) + + for ident in node.undeclared_identifiers(): + if ident != 'context' and \ + ident not in self.declared.union(self.locally_declared): + self.undeclared.add(ident) + + if not node.is_anonymous: + self._check_name_exists(self.topleveldefs, node) + self.undeclared.add(node.funcname) + elif node is not self.node: + self._check_name_exists(self.closuredefs, node) + for ident in node.declared_identifiers(): + self.argument_declared.add(ident) + for n in node.nodes: + n.accept_visitor(self) + + def visitTextTag(self, node): + for ident in node.undeclared_identifiers(): + if ident != 'context' and \ + ident not in self.declared.union(self.locally_declared): + self.undeclared.add(ident) + + def visitIncludeTag(self, node): + self.check_declared(node) + + def visitPageTag(self, node): + for ident in node.declared_identifiers(): + self.argument_declared.add(ident) + self.check_declared(node) + + def visitCallNamespaceTag(self, node): + self.visitCallTag(node) + + def visitCallTag(self, node): + if node is self.node: + for ident in node.undeclared_identifiers(): + if ident != 'context' and \ + ident not in self.declared.union( + self.locally_declared): + self.undeclared.add(ident) + for ident in node.declared_identifiers(): + self.argument_declared.add(ident) + for n in node.nodes: + n.accept_visitor(self) + else: + for ident in node.undeclared_identifiers(): + if ident != 'context' and \ + ident not in self.declared.union( + self.locally_declared): + self.undeclared.add(ident) + + +_FOR_LOOP = re.compile( + r'^for\s+((?:\(?)\s*[A-Za-z_][A-Za-z_0-9]*' + r'(?:\s*,\s*(?:[A-Za-z_][A-Za-z0-9_]*),??)*\s*(?:\)?))\s+in\s+(.*):' +) + +def mangle_mako_loop(node, printer): + """converts a for loop into a context manager wrapped around a for loop + when access to the `loop` variable has been detected in the for loop body + """ + loop_variable = LoopVariable() + node.accept_visitor(loop_variable) + if loop_variable.detected: + node.nodes[-1].has_loop_context = True + match = _FOR_LOOP.match(node.text) + if match: + printer.writelines( + 'loop = __M_loop._enter(%s)' % match.group(2), + 'try:' + #'with __M_loop(%s) as loop:' % match.group(2) + ) + text = 'for %s in loop:' % match.group(1) + else: + raise SyntaxError("Couldn't apply loop context: %s" % node.text) + else: + text = node.text + return text + + +class LoopVariable(object): + """A node visitor which looks for the name 'loop' within undeclared + identifiers.""" + + def __init__(self): + self.detected = False + + def _loop_reference_detected(self, node): + if 'loop' in node.undeclared_identifiers(): + self.detected = True + else: + for n in node.get_children(): + n.accept_visitor(self) + + def visitControlLine(self, node): + self._loop_reference_detected(node) + + def visitCode(self, node): + self._loop_reference_detected(node) + + def visitExpression(self, node): + self._loop_reference_detected(node) diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/compat.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/compat.py new file mode 100644 index 00000000000..fe277bbf05a --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/compat.py @@ -0,0 +1,174 @@ +import sys +import time + +py3k = sys.version_info >= (3, 0) +py33 = sys.version_info >= (3, 3) +py2k = sys.version_info < (3,) +py26 = sys.version_info >= (2, 6) +jython = sys.platform.startswith('java') +win32 = sys.platform.startswith('win') +pypy = hasattr(sys, 'pypy_version_info') + +if py3k: + from io import StringIO + import builtins as compat_builtins + from urllib.parse import quote_plus, unquote_plus + from html.entities import codepoint2name, name2codepoint + string_types = str, + binary_type = bytes + text_type = str + + from io import BytesIO as byte_buffer + + def u(s): + return s + + def b(s): + return s.encode("latin-1") + + def octal(lit): + return eval("0o" + lit) + +else: + import __builtin__ as compat_builtins + try: + from cStringIO import StringIO + except: + from StringIO import StringIO + + byte_buffer = StringIO + + from urllib import quote_plus, unquote_plus + from htmlentitydefs import codepoint2name, name2codepoint + string_types = basestring, + binary_type = str + text_type = unicode + + def u(s): + return unicode(s, "utf-8") + + def b(s): + return s + + def octal(lit): + return eval("0" + lit) + + +if py33: + from importlib import machinery + def load_module(module_id, path): + return machinery.SourceFileLoader(module_id, path).load_module() +else: + import imp + def load_module(module_id, path): + fp = open(path, 'rb') + try: + return imp.load_source(module_id, path, fp) + finally: + fp.close() + + +if py3k: + def reraise(tp, value, tb=None, cause=None): + if cause is not None: + value.__cause__ = cause + if value.__traceback__ is not tb: + raise value.with_traceback(tb) + raise value +else: + exec("def reraise(tp, value, tb=None, cause=None):\n" + " raise tp, value, tb\n") + + +def exception_as(): + return sys.exc_info()[1] + +try: + import threading + if py3k: + import _thread as thread + else: + import thread +except ImportError: + import dummy_threading as threading + if py3k: + import _dummy_thread as thread + else: + import dummy_thread as thread + +if win32 or jython: + time_func = time.clock +else: + time_func = time.time + +try: + from functools import partial +except: + def partial(func, *args, **keywords): + def newfunc(*fargs, **fkeywords): + newkeywords = keywords.copy() + newkeywords.update(fkeywords) + return func(*(args + fargs), **newkeywords) + return newfunc + + +all = all +import json + +def exception_name(exc): + return exc.__class__.__name__ + +try: + from inspect import CO_VARKEYWORDS, CO_VARARGS + def inspect_func_args(fn): + if py3k: + co = fn.__code__ + else: + co = fn.func_code + + nargs = co.co_argcount + names = co.co_varnames + args = list(names[:nargs]) + + varargs = None + if co.co_flags & CO_VARARGS: + varargs = co.co_varnames[nargs] + nargs = nargs + 1 + varkw = None + if co.co_flags & CO_VARKEYWORDS: + varkw = co.co_varnames[nargs] + + if py3k: + return args, varargs, varkw, fn.__defaults__ + else: + return args, varargs, varkw, fn.func_defaults +except ImportError: + import inspect + def inspect_func_args(fn): + return inspect.getargspec(fn) + +if py3k: + def callable(fn): + return hasattr(fn, '__call__') +else: + callable = callable + + +################################################ +# cross-compatible metaclass implementation +# Copyright (c) 2010-2012 Benjamin Peterson +def with_metaclass(meta, base=object): + """Create a base class with a metaclass.""" + return meta("%sBase" % meta.__name__, (base,), {}) +################################################ + + +def arg_stringname(func_arg): + """Gets the string name of a kwarg or vararg + In Python3.4 a function's args are + of _ast.arg type not _ast.name + """ + if hasattr(func_arg, 'arg'): + return func_arg.arg + else: + return str(func_arg) diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/exceptions.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/exceptions.py new file mode 100644 index 00000000000..c531f2118d0 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/exceptions.py @@ -0,0 +1,373 @@ +# mako/exceptions.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + +"""exception classes""" + +import traceback +import sys +from mako import util, compat + +class MakoException(Exception): + pass + +class RuntimeException(MakoException): + pass + +def _format_filepos(lineno, pos, filename): + if filename is None: + return " at line: %d char: %d" % (lineno, pos) + else: + return " in file '%s' at line: %d char: %d" % (filename, lineno, pos) + + +class CompileException(MakoException): + def __init__(self, message, source, lineno, pos, filename): + MakoException.__init__(self, + message + _format_filepos(lineno, pos, filename)) + self.lineno = lineno + self.pos = pos + self.filename = filename + self.source = source + +class SyntaxException(MakoException): + def __init__(self, message, source, lineno, pos, filename): + MakoException.__init__(self, + message + _format_filepos(lineno, pos, filename)) + self.lineno = lineno + self.pos = pos + self.filename = filename + self.source = source + +class UnsupportedError(MakoException): + """raised when a retired feature is used.""" + +class NameConflictError(MakoException): + """raised when a reserved word is used inappropriately""" + +class TemplateLookupException(MakoException): + pass + +class TopLevelLookupException(TemplateLookupException): + pass + +class RichTraceback(object): + """Pull the current exception from the ``sys`` traceback and extracts + Mako-specific template information. + + See the usage examples in :ref:`handling_exceptions`. + + """ + def __init__(self, error=None, traceback=None): + self.source, self.lineno = "", 0 + + if error is None or traceback is None: + t, value, tback = sys.exc_info() + + if error is None: + error = value or t + + if traceback is None: + traceback = tback + + self.error = error + self.records = self._init(traceback) + + if isinstance(self.error, (CompileException, SyntaxException)): + self.source = self.error.source + self.lineno = self.error.lineno + self._has_source = True + + self._init_message() + + @property + def errorname(self): + return compat.exception_name(self.error) + + def _init_message(self): + """Find a unicode representation of self.error""" + try: + self.message = compat.text_type(self.error) + except UnicodeError: + try: + self.message = str(self.error) + except UnicodeEncodeError: + # Fallback to args as neither unicode nor + # str(Exception(u'\xe6')) work in Python < 2.6 + self.message = self.error.args[0] + if not isinstance(self.message, compat.text_type): + self.message = compat.text_type(self.message, 'ascii', 'replace') + + def _get_reformatted_records(self, records): + for rec in records: + if rec[6] is not None: + yield (rec[4], rec[5], rec[2], rec[6]) + else: + yield tuple(rec[0:4]) + + @property + def traceback(self): + """Return a list of 4-tuple traceback records (i.e. normal python + format) with template-corresponding lines remapped to the originating + template. + + """ + return list(self._get_reformatted_records(self.records)) + + @property + def reverse_records(self): + return reversed(self.records) + + @property + def reverse_traceback(self): + """Return the same data as traceback, except in reverse order. + """ + + return list(self._get_reformatted_records(self.reverse_records)) + + def _init(self, trcback): + """format a traceback from sys.exc_info() into 7-item tuples, + containing the regular four traceback tuple items, plus the original + template filename, the line number adjusted relative to the template + source, and code line from that line number of the template.""" + + import mako.template + mods = {} + rawrecords = traceback.extract_tb(trcback) + new_trcback = [] + for filename, lineno, function, line in rawrecords: + if not line: + line = '' + try: + (line_map, template_lines) = mods[filename] + except KeyError: + try: + info = mako.template._get_module_info(filename) + module_source = info.code + template_source = info.source + template_filename = info.template_filename or filename + except KeyError: + # A normal .py file (not a Template) + if not compat.py3k: + try: + fp = open(filename, 'rb') + encoding = util.parse_encoding(fp) + fp.close() + except IOError: + encoding = None + if encoding: + line = line.decode(encoding) + else: + line = line.decode('ascii', 'replace') + new_trcback.append((filename, lineno, function, line, + None, None, None, None)) + continue + + template_ln = 1 + + source_map = mako.template.ModuleInfo.\ + get_module_source_metadata( + module_source, full_line_map=True) + line_map = source_map['full_line_map'] + + template_lines = [line for line in + template_source.split("\n")] + mods[filename] = (line_map, template_lines) + + template_ln = line_map[lineno - 1] + + if template_ln <= len(template_lines): + template_line = template_lines[template_ln - 1] + else: + template_line = None + new_trcback.append((filename, lineno, function, + line, template_filename, template_ln, + template_line, template_source)) + if not self.source: + for l in range(len(new_trcback) - 1, 0, -1): + if new_trcback[l][5]: + self.source = new_trcback[l][7] + self.lineno = new_trcback[l][5] + break + else: + if new_trcback: + try: + # A normal .py file (not a Template) + fp = open(new_trcback[-1][0], 'rb') + encoding = util.parse_encoding(fp) + fp.seek(0) + self.source = fp.read() + fp.close() + if encoding: + self.source = self.source.decode(encoding) + except IOError: + self.source = '' + self.lineno = new_trcback[-1][1] + return new_trcback + + +def text_error_template(lookup=None): + """Provides a template that renders a stack trace in a similar format to + the Python interpreter, substituting source template filenames, line + numbers and code for that of the originating source template, as + applicable. + + """ + import mako.template + return mako.template.Template(r""" +<%page args="error=None, traceback=None"/> +<%! + from mako.exceptions import RichTraceback +%>\ +<% + tback = RichTraceback(error=error, traceback=traceback) +%>\ +Traceback (most recent call last): +% for (filename, lineno, function, line) in tback.traceback: + File "${filename}", line ${lineno}, in ${function or '?'} + ${line | trim} +% endfor +${tback.errorname}: ${tback.message} +""") + + +def _install_pygments(): + global syntax_highlight, pygments_html_formatter + from mako.ext.pygmentplugin import syntax_highlight,\ + pygments_html_formatter + +def _install_fallback(): + global syntax_highlight, pygments_html_formatter + from mako.filters import html_escape + pygments_html_formatter = None + def syntax_highlight(filename='', language=None): + return html_escape + +def _install_highlighting(): + try: + _install_pygments() + except ImportError: + _install_fallback() +_install_highlighting() + +def html_error_template(): + """Provides a template that renders a stack trace in an HTML format, + providing an excerpt of code as well as substituting source template + filenames, line numbers and code for that of the originating source + template, as applicable. + + The template's default ``encoding_errors`` value is + ``'htmlentityreplace'``. The template has two options. With the + ``full`` option disabled, only a section of an HTML document is + returned. With the ``css`` option disabled, the default stylesheet + won't be included. + + """ + import mako.template + return mako.template.Template(r""" +<%! + from mako.exceptions import RichTraceback, syntax_highlight,\ + pygments_html_formatter +%> +<%page args="full=True, css=True, error=None, traceback=None"/> +% if full: + + + Mako Runtime Error +% endif +% if css: + +% endif +% if full: + + +% endif + +

Error !

+<% + tback = RichTraceback(error=error, traceback=traceback) + src = tback.source + line = tback.lineno + if src: + lines = src.split('\n') + else: + lines = None +%> +

${tback.errorname}: ${tback.message|h}

+ +% if lines: +
+
+% for index in range(max(0, line-4),min(len(lines), line+5)): + <% + if pygments_html_formatter: + pygments_html_formatter.linenostart = index + 1 + %> + % if index + 1 == line: + <% + if pygments_html_formatter: + old_cssclass = pygments_html_formatter.cssclass + pygments_html_formatter.cssclass = 'error ' + old_cssclass + %> + ${lines[index] | syntax_highlight(language='mako')} + <% + if pygments_html_formatter: + pygments_html_formatter.cssclass = old_cssclass + %> + % else: + ${lines[index] | syntax_highlight(language='mako')} + % endif +% endfor +
+
+% endif + +
+% for (filename, lineno, function, line) in tback.reverse_traceback: +
${filename}, line ${lineno}:
+
+ <% + if pygments_html_formatter: + pygments_html_formatter.linenostart = lineno + %> +
${line | syntax_highlight(filename)}
+
+% endfor +
+ +% if full: + + +% endif +""", output_encoding=sys.getdefaultencoding(), + encoding_errors='htmlentityreplace') diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/filters.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/filters.py new file mode 100644 index 00000000000..d79ce2388f6 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/filters.py @@ -0,0 +1,201 @@ +# mako/filters.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + + +import re +import codecs + +from mako.compat import quote_plus, unquote_plus, codepoint2name, \ + name2codepoint + +from mako import compat + +xml_escapes = { + '&': '&', + '>': '>', + '<': '<', + '"': '"', # also " in html-only + "'": ''' # also ' in html-only +} + +# XXX: " is valid in HTML and XML +# ' is not valid HTML, but is valid XML + +def legacy_html_escape(s): + """legacy HTML escape for non-unicode mode.""" + s = s.replace("&", "&") + s = s.replace(">", ">") + s = s.replace("<", "<") + s = s.replace('"', """) + s = s.replace("'", "'") + return s + + +try: + import markupsafe + html_escape = markupsafe.escape +except ImportError: + html_escape = legacy_html_escape + +def xml_escape(string): + return re.sub(r'([&<"\'>])', lambda m: xml_escapes[m.group()], string) + +def url_escape(string): + # convert into a list of octets + string = string.encode("utf8") + return quote_plus(string) + +def legacy_url_escape(string): + # convert into a list of octets + return quote_plus(string) + +def url_unescape(string): + text = unquote_plus(string) + if not is_ascii_str(text): + text = text.decode("utf8") + return text + +def trim(string): + return string.strip() + + +class Decode(object): + def __getattr__(self, key): + def decode(x): + if isinstance(x, compat.text_type): + return x + elif not isinstance(x, compat.binary_type): + return decode(str(x)) + else: + return compat.text_type(x, encoding=key) + return decode +decode = Decode() + + +_ASCII_re = re.compile(r'\A[\x00-\x7f]*\Z') + +def is_ascii_str(text): + return isinstance(text, str) and _ASCII_re.match(text) + +################################################################ + +class XMLEntityEscaper(object): + def __init__(self, codepoint2name, name2codepoint): + self.codepoint2entity = dict([(c, compat.text_type('&%s;' % n)) + for c, n in codepoint2name.items()]) + self.name2codepoint = name2codepoint + + def escape_entities(self, text): + """Replace characters with their character entity references. + + Only characters corresponding to a named entity are replaced. + """ + return compat.text_type(text).translate(self.codepoint2entity) + + def __escape(self, m): + codepoint = ord(m.group()) + try: + return self.codepoint2entity[codepoint] + except (KeyError, IndexError): + return '&#x%X;' % codepoint + + + __escapable = re.compile(r'["&<>]|[^\x00-\x7f]') + + def escape(self, text): + """Replace characters with their character references. + + Replace characters by their named entity references. + Non-ASCII characters, if they do not have a named entity reference, + are replaced by numerical character references. + + The return value is guaranteed to be ASCII. + """ + return self.__escapable.sub(self.__escape, compat.text_type(text) + ).encode('ascii') + + # XXX: This regexp will not match all valid XML entity names__. + # (It punts on details involving involving CombiningChars and Extenders.) + # + # .. __: http://www.w3.org/TR/2000/REC-xml-20001006#NT-EntityRef + __characterrefs = re.compile(r'''& (?: + \#(\d+) + | \#x([\da-f]+) + | ( (?!\d) [:\w] [-.:\w]+ ) + ) ;''', + re.X | re.UNICODE) + + def __unescape(self, m): + dval, hval, name = m.groups() + if dval: + codepoint = int(dval) + elif hval: + codepoint = int(hval, 16) + else: + codepoint = self.name2codepoint.get(name, 0xfffd) + # U+FFFD = "REPLACEMENT CHARACTER" + if codepoint < 128: + return chr(codepoint) + return chr(codepoint) + + def unescape(self, text): + """Unescape character references. + + All character references (both entity references and numerical + character references) are unescaped. + """ + return self.__characterrefs.sub(self.__unescape, text) + + +_html_entities_escaper = XMLEntityEscaper(codepoint2name, name2codepoint) + +html_entities_escape = _html_entities_escaper.escape_entities +html_entities_unescape = _html_entities_escaper.unescape + + +def htmlentityreplace_errors(ex): + """An encoding error handler. + + This python `codecs`_ error handler replaces unencodable + characters with HTML entities, or, if no HTML entity exists for + the character, XML character references. + + >>> u'The cost was \u20ac12.'.encode('latin1', 'htmlentityreplace') + 'The cost was €12.' + """ + if isinstance(ex, UnicodeEncodeError): + # Handle encoding errors + bad_text = ex.object[ex.start:ex.end] + text = _html_entities_escaper.escape(bad_text) + return (compat.text_type(text), ex.end) + raise ex + +codecs.register_error('htmlentityreplace', htmlentityreplace_errors) + + +# TODO: options to make this dynamic per-compilation will be added in a later +# release +DEFAULT_ESCAPES = { + 'x': 'filters.xml_escape', + 'h': 'filters.html_escape', + 'u': 'filters.url_escape', + 'trim': 'filters.trim', + 'entity': 'filters.html_entities_escape', + 'unicode': 'unicode', + 'decode': 'decode', + 'str': 'str', + 'n': 'n' +} + +if compat.py3k: + DEFAULT_ESCAPES.update({ + 'unicode': 'str' + }) + +NON_UNICODE_ESCAPES = DEFAULT_ESCAPES.copy() +NON_UNICODE_ESCAPES['h'] = 'filters.legacy_html_escape' +NON_UNICODE_ESCAPES['u'] = 'filters.legacy_url_escape' + diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/lexer.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/lexer.py new file mode 100644 index 00000000000..1dda398215d --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/lexer.py @@ -0,0 +1,441 @@ +# mako/lexer.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + +"""provides the Lexer class for parsing template strings into parse trees.""" + +import re +import codecs +from mako import parsetree, exceptions, compat +from mako.pygen import adjust_whitespace + +_regexp_cache = {} + +class Lexer(object): + def __init__(self, text, filename=None, + disable_unicode=False, + input_encoding=None, preprocessor=None): + self.text = text + self.filename = filename + self.template = parsetree.TemplateNode(self.filename) + self.matched_lineno = 1 + self.matched_charpos = 0 + self.lineno = 1 + self.match_position = 0 + self.tag = [] + self.control_line = [] + self.ternary_stack = [] + self.disable_unicode = disable_unicode + self.encoding = input_encoding + + if compat.py3k and disable_unicode: + raise exceptions.UnsupportedError( + "Mako for Python 3 does not " + "support disabling Unicode") + + if preprocessor is None: + self.preprocessor = [] + elif not hasattr(preprocessor, '__iter__'): + self.preprocessor = [preprocessor] + else: + self.preprocessor = preprocessor + + @property + def exception_kwargs(self): + return {'source': self.text, + 'lineno': self.matched_lineno, + 'pos': self.matched_charpos, + 'filename': self.filename} + + def match(self, regexp, flags=None): + """compile the given regexp, cache the reg, and call match_reg().""" + + try: + reg = _regexp_cache[(regexp, flags)] + except KeyError: + if flags: + reg = re.compile(regexp, flags) + else: + reg = re.compile(regexp) + _regexp_cache[(regexp, flags)] = reg + + return self.match_reg(reg) + + def match_reg(self, reg): + """match the given regular expression object to the current text + position. + + if a match occurs, update the current text and line position. + + """ + + mp = self.match_position + + match = reg.match(self.text, self.match_position) + if match: + (start, end) = match.span() + if end == start: + self.match_position = end + 1 + else: + self.match_position = end + self.matched_lineno = self.lineno + lines = re.findall(r"\n", self.text[mp:self.match_position]) + cp = mp - 1 + while (cp >= 0 and cp < self.textlength and self.text[cp] != '\n'): + cp -= 1 + self.matched_charpos = mp - cp + self.lineno += len(lines) + #print "MATCHED:", match.group(0), "LINE START:", + # self.matched_lineno, "LINE END:", self.lineno + #print "MATCH:", regexp, "\n", self.text[mp : mp + 15], \ + # (match and "TRUE" or "FALSE") + return match + + def parse_until_text(self, *text): + startpos = self.match_position + text_re = r'|'.join(text) + brace_level = 0 + while True: + match = self.match(r'#.*\n') + if match: + continue + match = self.match(r'(\"\"\"|\'\'\'|\"|\')((? 0: + brace_level -= 1 + continue + return \ + self.text[startpos: + self.match_position - len(match.group(1))],\ + match.group(1) + match = self.match(r"(.*?)(?=\"|\'|#|%s)" % text_re, re.S) + if match: + brace_level += match.group(1).count('{') + brace_level -= match.group(1).count('}') + continue + raise exceptions.SyntaxException( + "Expected: %s" % + ','.join(text), + **self.exception_kwargs) + + def append_node(self, nodecls, *args, **kwargs): + kwargs.setdefault('source', self.text) + kwargs.setdefault('lineno', self.matched_lineno) + kwargs.setdefault('pos', self.matched_charpos) + kwargs['filename'] = self.filename + node = nodecls(*args, **kwargs) + if len(self.tag): + self.tag[-1].nodes.append(node) + else: + self.template.nodes.append(node) + # build a set of child nodes for the control line + # (used for loop variable detection) + # also build a set of child nodes on ternary control lines + # (used for determining if a pass needs to be auto-inserted + if self.control_line: + control_frame = self.control_line[-1] + control_frame.nodes.append(node) + if not (isinstance(node, parsetree.ControlLine) and + control_frame.is_ternary(node.keyword)): + if self.ternary_stack and self.ternary_stack[-1]: + self.ternary_stack[-1][-1].nodes.append(node) + if isinstance(node, parsetree.Tag): + if len(self.tag): + node.parent = self.tag[-1] + self.tag.append(node) + elif isinstance(node, parsetree.ControlLine): + if node.isend: + self.control_line.pop() + self.ternary_stack.pop() + elif node.is_primary: + self.control_line.append(node) + self.ternary_stack.append([]) + elif self.control_line and \ + self.control_line[-1].is_ternary(node.keyword): + self.ternary_stack[-1].append(node) + elif self.control_line and \ + not self.control_line[-1].is_ternary(node.keyword): + raise exceptions.SyntaxException( + "Keyword '%s' not a legal ternary for keyword '%s'" % + (node.keyword, self.control_line[-1].keyword), + **self.exception_kwargs) + + _coding_re = re.compile(r'#.*coding[:=]\s*([-\w.]+).*\r?\n') + + def decode_raw_stream(self, text, decode_raw, known_encoding, filename): + """given string/unicode or bytes/string, determine encoding + from magic encoding comment, return body as unicode + or raw if decode_raw=False + + """ + if isinstance(text, compat.text_type): + m = self._coding_re.match(text) + encoding = m and m.group(1) or known_encoding or 'ascii' + return encoding, text + + if text.startswith(codecs.BOM_UTF8): + text = text[len(codecs.BOM_UTF8):] + parsed_encoding = 'utf-8' + m = self._coding_re.match(text.decode('utf-8', 'ignore')) + if m is not None and m.group(1) != 'utf-8': + raise exceptions.CompileException( + "Found utf-8 BOM in file, with conflicting " + "magic encoding comment of '%s'" % m.group(1), + text.decode('utf-8', 'ignore'), + 0, 0, filename) + else: + m = self._coding_re.match(text.decode('utf-8', 'ignore')) + if m: + parsed_encoding = m.group(1) + else: + parsed_encoding = known_encoding or 'ascii' + + if decode_raw: + try: + text = text.decode(parsed_encoding) + except UnicodeDecodeError: + raise exceptions.CompileException( + "Unicode decode operation of encoding '%s' failed" % + parsed_encoding, + text.decode('utf-8', 'ignore'), + 0, 0, filename) + + return parsed_encoding, text + + def parse(self): + self.encoding, self.text = self.decode_raw_stream(self.text, + not self.disable_unicode, + self.encoding, + self.filename,) + + for preproc in self.preprocessor: + self.text = preproc(self.text) + + # push the match marker past the + # encoding comment. + self.match_reg(self._coding_re) + + self.textlength = len(self.text) + + while (True): + if self.match_position > self.textlength: + break + + if self.match_end(): + break + if self.match_expression(): + continue + if self.match_control_line(): + continue + if self.match_comment(): + continue + if self.match_tag_start(): + continue + if self.match_tag_end(): + continue + if self.match_python_block(): + continue + if self.match_text(): + continue + + if self.match_position > self.textlength: + break + raise exceptions.CompileException("assertion failed") + + if len(self.tag): + raise exceptions.SyntaxException("Unclosed tag: <%%%s>" % + self.tag[-1].keyword, + **self.exception_kwargs) + if len(self.control_line): + raise exceptions.SyntaxException( + "Unterminated control keyword: '%s'" % + self.control_line[-1].keyword, + self.text, + self.control_line[-1].lineno, + self.control_line[-1].pos, self.filename) + return self.template + + def match_tag_start(self): + match = self.match(r''' + \<% # opening tag + + ([\w\.\:]+) # keyword + + ((?:\s+\w+|\s*=\s*|".*?"|'.*?')*) # attrname, = \ + # sign, string expression + + \s* # more whitespace + + (/)?> # closing + + ''', + + re.I | re.S | re.X) + + if match: + keyword, attr, isend = match.groups() + self.keyword = keyword + attributes = {} + if attr: + for att in re.findall( + r"\s*(\w+)\s*=\s*(?:'([^']*)'|\"([^\"]*)\")", attr): + key, val1, val2 = att + text = val1 or val2 + text = text.replace('\r\n', '\n') + attributes[key] = text + self.append_node(parsetree.Tag, keyword, attributes) + if isend: + self.tag.pop() + else: + if keyword == 'text': + match = self.match(r'(.*?)(?=\)', re.S) + if not match: + raise exceptions.SyntaxException( + "Unclosed tag: <%%%s>" % + self.tag[-1].keyword, + **self.exception_kwargs) + self.append_node(parsetree.Text, match.group(1)) + return self.match_tag_end() + return True + else: + return False + + def match_tag_end(self): + match = self.match(r'\') + if match: + if not len(self.tag): + raise exceptions.SyntaxException( + "Closing tag without opening tag: " % + match.group(1), + **self.exception_kwargs) + elif self.tag[-1].keyword != match.group(1): + raise exceptions.SyntaxException( + "Closing tag does not match tag: <%%%s>" % + (match.group(1), self.tag[-1].keyword), + **self.exception_kwargs) + self.tag.pop() + return True + else: + return False + + def match_end(self): + match = self.match(r'\Z', re.S) + if match: + string = match.group() + if string: + return string + else: + return True + else: + return False + + def match_text(self): + match = self.match(r""" + (.*?) # anything, followed by: + ( + (?<=\n)(?=[ \t]*(?=%|\#\#)) # an eval or line-based + # comment preceded by a + # consumed newline and whitespace + | + (?=\${) # an expression + | + (?=') + # the trailing newline helps + # compiler.parse() not complain about indentation + text = adjust_whitespace(text) + "\n" + self.append_node( + parsetree.Code, + text, + match.group(1) == '!', lineno=line, pos=pos) + return True + else: + return False + + def match_expression(self): + match = self.match(r"\${") + if match: + line, pos = self.matched_lineno, self.matched_charpos + text, end = self.parse_until_text(r'\|', r'}') + if end == '|': + escapes, end = self.parse_until_text(r'}') + else: + escapes = "" + text = text.replace('\r\n', '\n') + self.append_node( + parsetree.Expression, + text, escapes.strip(), + lineno=line, pos=pos) + return True + else: + return False + + def match_control_line(self): + match = self.match( + r"(?<=^)[\t ]*(%(?!%)|##)[\t ]*((?:(?:\\r?\n)|[^\r\n])*)" + r"(?:\r?\n|\Z)", re.M) + if match: + operator = match.group(1) + text = match.group(2) + if operator == '%': + m2 = re.match(r'(end)?(\w+)\s*(.*)', text) + if not m2: + raise exceptions.SyntaxException( + "Invalid control line: '%s'" % + text, + **self.exception_kwargs) + isend, keyword = m2.group(1, 2) + isend = (isend is not None) + + if isend: + if not len(self.control_line): + raise exceptions.SyntaxException( + "No starting keyword '%s' for '%s'" % + (keyword, text), + **self.exception_kwargs) + elif self.control_line[-1].keyword != keyword: + raise exceptions.SyntaxException( + "Keyword '%s' doesn't match keyword '%s'" % + (text, self.control_line[-1].keyword), + **self.exception_kwargs) + self.append_node(parsetree.ControlLine, keyword, isend, text) + else: + self.append_node(parsetree.Comment, text) + return True + else: + return False + + def match_comment(self): + """matches the multiline version of a comment""" + match = self.match(r"<%doc>(.*?)", re.S) + if match: + self.append_node(parsetree.Comment, match.group(1)) + return True + else: + return False + diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/lookup.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/lookup.py new file mode 100644 index 00000000000..2af5411907a --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/lookup.py @@ -0,0 +1,359 @@ +# mako/lookup.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + +import os, stat, posixpath, re +from mako import exceptions, util +from mako.template import Template + +try: + import threading +except: + import dummy_threading as threading + +class TemplateCollection(object): + """Represent a collection of :class:`.Template` objects, + identifiable via URI. + + A :class:`.TemplateCollection` is linked to the usage of + all template tags that address other templates, such + as ``<%include>``, ``<%namespace>``, and ``<%inherit>``. + The ``file`` attribute of each of those tags refers + to a string URI that is passed to that :class:`.Template` + object's :class:`.TemplateCollection` for resolution. + + :class:`.TemplateCollection` is an abstract class, + with the usual default implementation being :class:`.TemplateLookup`. + + """ + + def has_template(self, uri): + """Return ``True`` if this :class:`.TemplateLookup` is + capable of returning a :class:`.Template` object for the + given ``uri``. + + :param uri: String URI of the template to be resolved. + + """ + try: + self.get_template(uri) + return True + except exceptions.TemplateLookupException: + return False + + def get_template(self, uri, relativeto=None): + """Return a :class:`.Template` object corresponding to the given + ``uri``. + + The default implementation raises + :class:`.NotImplementedError`. Implementations should + raise :class:`.TemplateLookupException` if the given ``uri`` + cannot be resolved. + + :param uri: String URI of the template to be resolved. + :param relativeto: if present, the given ``uri`` is assumed to + be relative to this URI. + + """ + raise NotImplementedError() + + def filename_to_uri(self, uri, filename): + """Convert the given ``filename`` to a URI relative to + this :class:`.TemplateCollection`.""" + + return uri + + def adjust_uri(self, uri, filename): + """Adjust the given ``uri`` based on the calling ``filename``. + + When this method is called from the runtime, the + ``filename`` parameter is taken directly to the ``filename`` + attribute of the calling template. Therefore a custom + :class:`.TemplateCollection` subclass can place any string + identifier desired in the ``filename`` parameter of the + :class:`.Template` objects it constructs and have them come back + here. + + """ + return uri + +class TemplateLookup(TemplateCollection): + """Represent a collection of templates that locates template source files + from the local filesystem. + + The primary argument is the ``directories`` argument, the list of + directories to search: + + .. sourcecode:: python + + lookup = TemplateLookup(["/path/to/templates"]) + some_template = lookup.get_template("/index.html") + + The :class:`.TemplateLookup` can also be given :class:`.Template` objects + programatically using :meth:`.put_string` or :meth:`.put_template`: + + .. sourcecode:: python + + lookup = TemplateLookup() + lookup.put_string("base.html", ''' + ${self.next()} + ''') + lookup.put_string("hello.html", ''' + <%include file='base.html'/> + + Hello, world ! + ''') + + + :param directories: A list of directory names which will be + searched for a particular template URI. The URI is appended + to each directory and the filesystem checked. + + :param collection_size: Approximate size of the collection used + to store templates. If left at its default of ``-1``, the size + is unbounded, and a plain Python dictionary is used to + relate URI strings to :class:`.Template` instances. + Otherwise, a least-recently-used cache object is used which + will maintain the size of the collection approximately to + the number given. + + :param filesystem_checks: When at its default value of ``True``, + each call to :meth:`.TemplateLookup.get_template()` will + compare the filesystem last modified time to the time in + which an existing :class:`.Template` object was created. + This allows the :class:`.TemplateLookup` to regenerate a + new :class:`.Template` whenever the original source has + been updated. Set this to ``False`` for a very minor + performance increase. + + :param modulename_callable: A callable which, when present, + is passed the path of the source file as well as the + requested URI, and then returns the full path of the + generated Python module file. This is used to inject + alternate schemes for Python module location. If left at + its default of ``None``, the built in system of generation + based on ``module_directory`` plus ``uri`` is used. + + All other keyword parameters available for + :class:`.Template` are mirrored here. When new + :class:`.Template` objects are created, the keywords + established with this :class:`.TemplateLookup` are passed on + to each new :class:`.Template`. + + """ + + def __init__(self, + directories=None, + module_directory=None, + filesystem_checks=True, + collection_size=-1, + format_exceptions=False, + error_handler=None, + disable_unicode=False, + bytestring_passthrough=False, + output_encoding=None, + encoding_errors='strict', + + cache_args=None, + cache_impl='beaker', + cache_enabled=True, + cache_type=None, + cache_dir=None, + cache_url=None, + + modulename_callable=None, + module_writer=None, + default_filters=None, + buffer_filters=(), + strict_undefined=False, + imports=None, + future_imports=None, + enable_loop=True, + input_encoding=None, + preprocessor=None, + lexer_cls=None): + + self.directories = [posixpath.normpath(d) for d in + util.to_list(directories, ()) + ] + self.module_directory = module_directory + self.modulename_callable = modulename_callable + self.filesystem_checks = filesystem_checks + self.collection_size = collection_size + + if cache_args is None: + cache_args = {} + # transfer deprecated cache_* args + if cache_dir: + cache_args.setdefault('dir', cache_dir) + if cache_url: + cache_args.setdefault('url', cache_url) + if cache_type: + cache_args.setdefault('type', cache_type) + + self.template_args = { + 'format_exceptions':format_exceptions, + 'error_handler':error_handler, + 'disable_unicode':disable_unicode, + 'bytestring_passthrough':bytestring_passthrough, + 'output_encoding':output_encoding, + 'cache_impl':cache_impl, + 'encoding_errors':encoding_errors, + 'input_encoding':input_encoding, + 'module_directory':module_directory, + 'module_writer':module_writer, + 'cache_args':cache_args, + 'cache_enabled':cache_enabled, + 'default_filters':default_filters, + 'buffer_filters':buffer_filters, + 'strict_undefined':strict_undefined, + 'imports':imports, + 'future_imports':future_imports, + 'enable_loop':enable_loop, + 'preprocessor':preprocessor, + 'lexer_cls':lexer_cls + } + + if collection_size == -1: + self._collection = {} + self._uri_cache = {} + else: + self._collection = util.LRUCache(collection_size) + self._uri_cache = util.LRUCache(collection_size) + self._mutex = threading.Lock() + + def get_template(self, uri): + """Return a :class:`.Template` object corresponding to the given + ``uri``. + + .. note:: The ``relativeto`` argument is not supported here at the moment. + + """ + + try: + if self.filesystem_checks: + return self._check(uri, self._collection[uri]) + else: + return self._collection[uri] + except KeyError: + u = re.sub(r'^\/+', '', uri) + for dir in self.directories: + srcfile = posixpath.normpath(posixpath.join(dir, u)) + if os.path.isfile(srcfile): + return self._load(srcfile, uri) + else: + raise exceptions.TopLevelLookupException( + "Cant locate template for uri %r" % uri) + + def adjust_uri(self, uri, relativeto): + """Adjust the given ``uri`` based on the given relative URI.""" + + key = (uri, relativeto) + if key in self._uri_cache: + return self._uri_cache[key] + + if uri[0] != '/': + if relativeto is not None: + v = self._uri_cache[key] = posixpath.join( + posixpath.dirname(relativeto), uri) + else: + v = self._uri_cache[key] = '/' + uri + else: + v = self._uri_cache[key] = uri + return v + + + def filename_to_uri(self, filename): + """Convert the given ``filename`` to a URI relative to + this :class:`.TemplateCollection`.""" + + try: + return self._uri_cache[filename] + except KeyError: + value = self._relativeize(filename) + self._uri_cache[filename] = value + return value + + def _relativeize(self, filename): + """Return the portion of a filename that is 'relative' + to the directories in this lookup. + + """ + + filename = posixpath.normpath(filename) + for dir in self.directories: + if filename[0:len(dir)] == dir: + return filename[len(dir):] + else: + return None + + def _load(self, filename, uri): + self._mutex.acquire() + try: + try: + # try returning from collection one + # more time in case concurrent thread already loaded + return self._collection[uri] + except KeyError: + pass + try: + if self.modulename_callable is not None: + module_filename = self.modulename_callable(filename, uri) + else: + module_filename = None + self._collection[uri] = template = Template( + uri=uri, + filename=posixpath.normpath(filename), + lookup=self, + module_filename=module_filename, + **self.template_args) + return template + except: + # if compilation fails etc, ensure + # template is removed from collection, + # re-raise + self._collection.pop(uri, None) + raise + finally: + self._mutex.release() + + def _check(self, uri, template): + if template.filename is None: + return template + + try: + template_stat = os.stat(template.filename) + if template.module._modified_time < \ + template_stat[stat.ST_MTIME]: + self._collection.pop(uri, None) + return self._load(template.filename, uri) + else: + return template + except OSError: + self._collection.pop(uri, None) + raise exceptions.TemplateLookupException( + "Cant locate template for uri %r" % uri) + + + def put_string(self, uri, text): + """Place a new :class:`.Template` object into this + :class:`.TemplateLookup`, based on the given string of + ``text``. + + """ + self._collection[uri] = Template( + text, + lookup=self, + uri=uri, + **self.template_args) + + def put_template(self, uri, template): + """Place a new :class:`.Template` object into this + :class:`.TemplateLookup`, based on the given + :class:`.Template` object. + + """ + self._collection[uri] = template + diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/parsetree.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/parsetree.py new file mode 100644 index 00000000000..49ec4e0696c --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/parsetree.py @@ -0,0 +1,594 @@ +# mako/parsetree.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + +"""defines the parse tree components for Mako templates.""" + +from mako import exceptions, ast, util, filters, compat +import re + +class Node(object): + """base class for a Node in the parse tree.""" + + def __init__(self, source, lineno, pos, filename): + self.source = source + self.lineno = lineno + self.pos = pos + self.filename = filename + + @property + def exception_kwargs(self): + return {'source': self.source, 'lineno': self.lineno, + 'pos': self.pos, 'filename': self.filename} + + def get_children(self): + return [] + + def accept_visitor(self, visitor): + def traverse(node): + for n in node.get_children(): + n.accept_visitor(visitor) + + method = getattr(visitor, "visit" + self.__class__.__name__, traverse) + method(self) + +class TemplateNode(Node): + """a 'container' node that stores the overall collection of nodes.""" + + def __init__(self, filename): + super(TemplateNode, self).__init__('', 0, 0, filename) + self.nodes = [] + self.page_attributes = {} + + def get_children(self): + return self.nodes + + def __repr__(self): + return "TemplateNode(%s, %r)" % ( + util.sorted_dict_repr(self.page_attributes), + self.nodes) + +class ControlLine(Node): + """defines a control line, a line-oriented python line or end tag. + + e.g.:: + + % if foo: + (markup) + % endif + + """ + + has_loop_context = False + + def __init__(self, keyword, isend, text, **kwargs): + super(ControlLine, self).__init__(**kwargs) + self.text = text + self.keyword = keyword + self.isend = isend + self.is_primary = keyword in ['for', 'if', 'while', 'try', 'with'] + self.nodes = [] + if self.isend: + self._declared_identifiers = [] + self._undeclared_identifiers = [] + else: + code = ast.PythonFragment(text, **self.exception_kwargs) + self._declared_identifiers = code.declared_identifiers + self._undeclared_identifiers = code.undeclared_identifiers + + def get_children(self): + return self.nodes + + def declared_identifiers(self): + return self._declared_identifiers + + def undeclared_identifiers(self): + return self._undeclared_identifiers + + def is_ternary(self, keyword): + """return true if the given keyword is a ternary keyword + for this ControlLine""" + + return keyword in { + 'if':set(['else', 'elif']), + 'try':set(['except', 'finally']), + 'for':set(['else']) + }.get(self.keyword, []) + + def __repr__(self): + return "ControlLine(%r, %r, %r, %r)" % ( + self.keyword, + self.text, + self.isend, + (self.lineno, self.pos) + ) + +class Text(Node): + """defines plain text in the template.""" + + def __init__(self, content, **kwargs): + super(Text, self).__init__(**kwargs) + self.content = content + + def __repr__(self): + return "Text(%r, %r)" % (self.content, (self.lineno, self.pos)) + +class Code(Node): + """defines a Python code block, either inline or module level. + + e.g.:: + + inline: + <% + x = 12 + %> + + module level: + <%! + import logger + %> + + """ + + def __init__(self, text, ismodule, **kwargs): + super(Code, self).__init__(**kwargs) + self.text = text + self.ismodule = ismodule + self.code = ast.PythonCode(text, **self.exception_kwargs) + + def declared_identifiers(self): + return self.code.declared_identifiers + + def undeclared_identifiers(self): + return self.code.undeclared_identifiers + + def __repr__(self): + return "Code(%r, %r, %r)" % ( + self.text, + self.ismodule, + (self.lineno, self.pos) + ) + +class Comment(Node): + """defines a comment line. + + # this is a comment + + """ + + def __init__(self, text, **kwargs): + super(Comment, self).__init__(**kwargs) + self.text = text + + def __repr__(self): + return "Comment(%r, %r)" % (self.text, (self.lineno, self.pos)) + +class Expression(Node): + """defines an inline expression. + + ${x+y} + + """ + + def __init__(self, text, escapes, **kwargs): + super(Expression, self).__init__(**kwargs) + self.text = text + self.escapes = escapes + self.escapes_code = ast.ArgumentList(escapes, **self.exception_kwargs) + self.code = ast.PythonCode(text, **self.exception_kwargs) + + def declared_identifiers(self): + return [] + + def undeclared_identifiers(self): + # TODO: make the "filter" shortcut list configurable at parse/gen time + return self.code.undeclared_identifiers.union( + self.escapes_code.undeclared_identifiers.difference( + set(filters.DEFAULT_ESCAPES.keys()) + ) + ).difference(self.code.declared_identifiers) + + def __repr__(self): + return "Expression(%r, %r, %r)" % ( + self.text, + self.escapes_code.args, + (self.lineno, self.pos) + ) + +class _TagMeta(type): + """metaclass to allow Tag to produce a subclass according to + its keyword""" + + _classmap = {} + + def __init__(cls, clsname, bases, dict): + if getattr(cls, '__keyword__', None) is not None: + cls._classmap[cls.__keyword__] = cls + super(_TagMeta, cls).__init__(clsname, bases, dict) + + def __call__(cls, keyword, attributes, **kwargs): + if ":" in keyword: + ns, defname = keyword.split(':') + return type.__call__(CallNamespaceTag, ns, defname, + attributes, **kwargs) + + try: + cls = _TagMeta._classmap[keyword] + except KeyError: + raise exceptions.CompileException( + "No such tag: '%s'" % keyword, + source=kwargs['source'], + lineno=kwargs['lineno'], + pos=kwargs['pos'], + filename=kwargs['filename'] + ) + return type.__call__(cls, keyword, attributes, **kwargs) + +class Tag(compat.with_metaclass(_TagMeta, Node)): + """abstract base class for tags. + + <%sometag/> + + <%someothertag> + stuff + + + """ + __keyword__ = None + + def __init__(self, keyword, attributes, expressions, + nonexpressions, required, **kwargs): + """construct a new Tag instance. + + this constructor not called directly, and is only called + by subclasses. + + :param keyword: the tag keyword + + :param attributes: raw dictionary of attribute key/value pairs + + :param expressions: a set of identifiers that are legal attributes, + which can also contain embedded expressions + + :param nonexpressions: a set of identifiers that are legal + attributes, which cannot contain embedded expressions + + :param \**kwargs: + other arguments passed to the Node superclass (lineno, pos) + + """ + super(Tag, self).__init__(**kwargs) + self.keyword = keyword + self.attributes = attributes + self._parse_attributes(expressions, nonexpressions) + missing = [r for r in required if r not in self.parsed_attributes] + if len(missing): + raise exceptions.CompileException( + "Missing attribute(s): %s" % + ",".join([repr(m) for m in missing]), + **self.exception_kwargs) + self.parent = None + self.nodes = [] + + def is_root(self): + return self.parent is None + + def get_children(self): + return self.nodes + + def _parse_attributes(self, expressions, nonexpressions): + undeclared_identifiers = set() + self.parsed_attributes = {} + for key in self.attributes: + if key in expressions: + expr = [] + for x in re.compile(r'(\${.+?})', + re.S).split(self.attributes[key]): + m = re.compile(r'^\${(.+?)}$', re.S).match(x) + if m: + code = ast.PythonCode(m.group(1).rstrip(), + **self.exception_kwargs) + # we aren't discarding "declared_identifiers" here, + # which we do so that list comprehension-declared + # variables aren't counted. As yet can't find a + # condition that requires it here. + undeclared_identifiers = \ + undeclared_identifiers.union( + code.undeclared_identifiers) + expr.append('(%s)' % m.group(1)) + else: + if x: + expr.append(repr(x)) + self.parsed_attributes[key] = " + ".join(expr) or repr('') + elif key in nonexpressions: + if re.search(r'\${.+?}', self.attributes[key]): + raise exceptions.CompileException( + "Attibute '%s' in tag '%s' does not allow embedded " + "expressions" % (key, self.keyword), + **self.exception_kwargs) + self.parsed_attributes[key] = repr(self.attributes[key]) + else: + raise exceptions.CompileException( + "Invalid attribute for tag '%s': '%s'" % + (self.keyword, key), + **self.exception_kwargs) + self.expression_undeclared_identifiers = undeclared_identifiers + + def declared_identifiers(self): + return [] + + def undeclared_identifiers(self): + return self.expression_undeclared_identifiers + + def __repr__(self): + return "%s(%r, %s, %r, %r)" % (self.__class__.__name__, + self.keyword, + util.sorted_dict_repr(self.attributes), + (self.lineno, self.pos), + self.nodes + ) + +class IncludeTag(Tag): + __keyword__ = 'include' + + def __init__(self, keyword, attributes, **kwargs): + super(IncludeTag, self).__init__( + keyword, + attributes, + ('file', 'import', 'args'), + (), ('file',), **kwargs) + self.page_args = ast.PythonCode( + "__DUMMY(%s)" % attributes.get('args', ''), + **self.exception_kwargs) + + def declared_identifiers(self): + return [] + + def undeclared_identifiers(self): + identifiers = self.page_args.undeclared_identifiers.\ + difference(set(["__DUMMY"])).\ + difference(self.page_args.declared_identifiers) + return identifiers.union(super(IncludeTag, self). + undeclared_identifiers()) + +class NamespaceTag(Tag): + __keyword__ = 'namespace' + + def __init__(self, keyword, attributes, **kwargs): + super(NamespaceTag, self).__init__( + keyword, attributes, + ('file',), + ('name','inheritable', + 'import','module'), + (), **kwargs) + + self.name = attributes.get('name', '__anon_%s' % hex(abs(id(self)))) + if not 'name' in attributes and not 'import' in attributes: + raise exceptions.CompileException( + "'name' and/or 'import' attributes are required " + "for <%namespace>", + **self.exception_kwargs) + if 'file' in attributes and 'module' in attributes: + raise exceptions.CompileException( + "<%namespace> may only have one of 'file' or 'module'", + **self.exception_kwargs + ) + + def declared_identifiers(self): + return [] + +class TextTag(Tag): + __keyword__ = 'text' + + def __init__(self, keyword, attributes, **kwargs): + super(TextTag, self).__init__( + keyword, + attributes, (), + ('filter'), (), **kwargs) + self.filter_args = ast.ArgumentList( + attributes.get('filter', ''), + **self.exception_kwargs) + + def undeclared_identifiers(self): + return self.filter_args.\ + undeclared_identifiers.\ + difference(filters.DEFAULT_ESCAPES.keys()).union( + self.expression_undeclared_identifiers + ) + +class DefTag(Tag): + __keyword__ = 'def' + + def __init__(self, keyword, attributes, **kwargs): + expressions = ['buffered', 'cached'] + [ + c for c in attributes if c.startswith('cache_')] + + + super(DefTag, self).__init__( + keyword, + attributes, + expressions, + ('name', 'filter', 'decorator'), + ('name',), + **kwargs) + name = attributes['name'] + if re.match(r'^[\w_]+$', name): + raise exceptions.CompileException( + "Missing parenthesis in %def", + **self.exception_kwargs) + self.function_decl = ast.FunctionDecl("def " + name + ":pass", + **self.exception_kwargs) + self.name = self.function_decl.funcname + self.decorator = attributes.get('decorator', '') + self.filter_args = ast.ArgumentList( + attributes.get('filter', ''), + **self.exception_kwargs) + + is_anonymous = False + is_block = False + + @property + def funcname(self): + return self.function_decl.funcname + + def get_argument_expressions(self, **kw): + return self.function_decl.get_argument_expressions(**kw) + + def declared_identifiers(self): + return self.function_decl.allargnames + + def undeclared_identifiers(self): + res = [] + for c in self.function_decl.defaults: + res += list(ast.PythonCode(c, **self.exception_kwargs). + undeclared_identifiers) + return set(res).union( + self.filter_args.\ + undeclared_identifiers.\ + difference(filters.DEFAULT_ESCAPES.keys()) + ).union( + self.expression_undeclared_identifiers + ).difference( + self.function_decl.allargnames + ) + +class BlockTag(Tag): + __keyword__ = 'block' + + def __init__(self, keyword, attributes, **kwargs): + expressions = ['buffered', 'cached', 'args'] + [ + c for c in attributes if c.startswith('cache_')] + + super(BlockTag, self).__init__( + keyword, + attributes, + expressions, + ('name','filter', 'decorator'), + (), + **kwargs) + name = attributes.get('name') + if name and not re.match(r'^[\w_]+$',name): + raise exceptions.CompileException( + "%block may not specify an argument signature", + **self.exception_kwargs) + if not name and attributes.get('args', None): + raise exceptions.CompileException( + "Only named %blocks may specify args", + **self.exception_kwargs + ) + self.body_decl = ast.FunctionArgs(attributes.get('args', ''), + **self.exception_kwargs) + + self.name = name + self.decorator = attributes.get('decorator', '') + self.filter_args = ast.ArgumentList( + attributes.get('filter', ''), + **self.exception_kwargs) + + + is_block = True + + @property + def is_anonymous(self): + return self.name is None + + @property + def funcname(self): + return self.name or "__M_anon_%d" % (self.lineno, ) + + def get_argument_expressions(self, **kw): + return self.body_decl.get_argument_expressions(**kw) + + def declared_identifiers(self): + return self.body_decl.allargnames + + def undeclared_identifiers(self): + return (self.filter_args.\ + undeclared_identifiers.\ + difference(filters.DEFAULT_ESCAPES.keys()) + ).union(self.expression_undeclared_identifiers) + + + +class CallTag(Tag): + __keyword__ = 'call' + + def __init__(self, keyword, attributes, **kwargs): + super(CallTag, self).__init__(keyword, attributes, + ('args'), ('expr',), ('expr',), **kwargs) + self.expression = attributes['expr'] + self.code = ast.PythonCode(self.expression, **self.exception_kwargs) + self.body_decl = ast.FunctionArgs(attributes.get('args', ''), + **self.exception_kwargs) + + def declared_identifiers(self): + return self.code.declared_identifiers.union(self.body_decl.allargnames) + + def undeclared_identifiers(self): + return self.code.undeclared_identifiers.\ + difference(self.code.declared_identifiers) + +class CallNamespaceTag(Tag): + + def __init__(self, namespace, defname, attributes, **kwargs): + super(CallNamespaceTag, self).__init__( + namespace + ":" + defname, + attributes, + tuple(attributes.keys()) + ('args', ), + (), + (), + **kwargs) + + self.expression = "%s.%s(%s)" % ( + namespace, + defname, + ",".join(["%s=%s" % (k, v) for k, v in + self.parsed_attributes.items() + if k != 'args']) + ) + self.code = ast.PythonCode(self.expression, **self.exception_kwargs) + self.body_decl = ast.FunctionArgs( + attributes.get('args', ''), + **self.exception_kwargs) + + def declared_identifiers(self): + return self.code.declared_identifiers.union(self.body_decl.allargnames) + + def undeclared_identifiers(self): + return self.code.undeclared_identifiers.\ + difference(self.code.declared_identifiers) + +class InheritTag(Tag): + __keyword__ = 'inherit' + + def __init__(self, keyword, attributes, **kwargs): + super(InheritTag, self).__init__( + keyword, attributes, + ('file',), (), ('file',), **kwargs) + +class PageTag(Tag): + __keyword__ = 'page' + + def __init__(self, keyword, attributes, **kwargs): + expressions = ['cached', 'args', 'expression_filter', 'enable_loop'] + [ + c for c in attributes if c.startswith('cache_')] + + super(PageTag, self).__init__( + keyword, + attributes, + expressions, + (), + (), + **kwargs) + self.body_decl = ast.FunctionArgs(attributes.get('args', ''), + **self.exception_kwargs) + self.filter_args = ast.ArgumentList( + attributes.get('expression_filter', ''), + **self.exception_kwargs) + + def declared_identifiers(self): + return self.body_decl.allargnames + + diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/pygen.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/pygen.py new file mode 100644 index 00000000000..5ba5125a4c7 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/pygen.py @@ -0,0 +1,299 @@ +# mako/pygen.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + +"""utilities for generating and formatting literal Python code.""" + +import re +from mako import exceptions + +class PythonPrinter(object): + def __init__(self, stream): + # indentation counter + self.indent = 0 + + # a stack storing information about why we incremented + # the indentation counter, to help us determine if we + # should decrement it + self.indent_detail = [] + + # the string of whitespace multiplied by the indent + # counter to produce a line + self.indentstring = " " + + # the stream we are writing to + self.stream = stream + + # current line number + self.lineno = 1 + + # a list of lines that represents a buffered "block" of code, + # which can be later printed relative to an indent level + self.line_buffer = [] + + self.in_indent_lines = False + + self._reset_multi_line_flags() + + # mapping of generated python lines to template + # source lines + self.source_map = {} + + def _update_lineno(self, num): + self.lineno += num + + def start_source(self, lineno): + if self.lineno not in self.source_map: + self.source_map[self.lineno] = lineno + + def write_blanks(self, num): + self.stream.write("\n" * num) + self._update_lineno(num) + + def write_indented_block(self, block): + """print a line or lines of python which already contain indentation. + + The indentation of the total block of lines will be adjusted to that of + the current indent level.""" + self.in_indent_lines = False + for l in re.split(r'\r?\n', block): + self.line_buffer.append(l) + self._update_lineno(1) + + def writelines(self, *lines): + """print a series of lines of python.""" + for line in lines: + self.writeline(line) + + def writeline(self, line): + """print a line of python, indenting it according to the current + indent level. + + this also adjusts the indentation counter according to the + content of the line. + + """ + + if not self.in_indent_lines: + self._flush_adjusted_lines() + self.in_indent_lines = True + + if (line is None or + re.match(r"^\s*#",line) or + re.match(r"^\s*$", line) + ): + hastext = False + else: + hastext = True + + is_comment = line and len(line) and line[0] == '#' + + # see if this line should decrease the indentation level + if (not is_comment and + (not hastext or self._is_unindentor(line)) + ): + + if self.indent > 0: + self.indent -= 1 + # if the indent_detail stack is empty, the user + # probably put extra closures - the resulting + # module wont compile. + if len(self.indent_detail) == 0: + raise exceptions.SyntaxException( + "Too many whitespace closures") + self.indent_detail.pop() + + if line is None: + return + + # write the line + self.stream.write(self._indent_line(line) + "\n") + self._update_lineno(len(line.split("\n"))) + + # see if this line should increase the indentation level. + # note that a line can both decrase (before printing) and + # then increase (after printing) the indentation level. + + if re.search(r":[ \t]*(?:#.*)?$", line): + # increment indentation count, and also + # keep track of what the keyword was that indented us, + # if it is a python compound statement keyword + # where we might have to look for an "unindent" keyword + match = re.match(r"^\s*(if|try|elif|while|for|with)", line) + if match: + # its a "compound" keyword, so we will check for "unindentors" + indentor = match.group(1) + self.indent += 1 + self.indent_detail.append(indentor) + else: + indentor = None + # its not a "compound" keyword. but lets also + # test for valid Python keywords that might be indenting us, + # else assume its a non-indenting line + m2 = re.match(r"^\s*(def|class|else|elif|except|finally)", + line) + if m2: + self.indent += 1 + self.indent_detail.append(indentor) + + def close(self): + """close this printer, flushing any remaining lines.""" + self._flush_adjusted_lines() + + def _is_unindentor(self, line): + """return true if the given line is an 'unindentor', + relative to the last 'indent' event received. + + """ + + # no indentation detail has been pushed on; return False + if len(self.indent_detail) == 0: + return False + + indentor = self.indent_detail[-1] + + # the last indent keyword we grabbed is not a + # compound statement keyword; return False + if indentor is None: + return False + + # if the current line doesnt have one of the "unindentor" keywords, + # return False + match = re.match(r"^\s*(else|elif|except|finally).*\:", line) + if not match: + return False + + # whitespace matches up, we have a compound indentor, + # and this line has an unindentor, this + # is probably good enough + return True + + # should we decide that its not good enough, heres + # more stuff to check. + #keyword = match.group(1) + + # match the original indent keyword + #for crit in [ + # (r'if|elif', r'else|elif'), + # (r'try', r'except|finally|else'), + # (r'while|for', r'else'), + #]: + # if re.match(crit[0], indentor) and re.match(crit[1], keyword): + # return True + + #return False + + def _indent_line(self, line, stripspace=''): + """indent the given line according to the current indent level. + + stripspace is a string of space that will be truncated from the + start of the line before indenting.""" + + return re.sub(r"^%s" % stripspace, self.indentstring + * self.indent, line) + + def _reset_multi_line_flags(self): + """reset the flags which would indicate we are in a backslashed + or triple-quoted section.""" + + self.backslashed, self.triplequoted = False, False + + def _in_multi_line(self, line): + """return true if the given line is part of a multi-line block, + via backslash or triple-quote.""" + + # we are only looking for explicitly joined lines here, not + # implicit ones (i.e. brackets, braces etc.). this is just to + # guard against the possibility of modifying the space inside of + # a literal multiline string with unfortunately placed + # whitespace + + current_state = (self.backslashed or self.triplequoted) + + if re.search(r"\\$", line): + self.backslashed = True + else: + self.backslashed = False + + triples = len(re.findall(r"\"\"\"|\'\'\'", line)) + if triples == 1 or triples % 2 != 0: + self.triplequoted = not self.triplequoted + + return current_state + + def _flush_adjusted_lines(self): + stripspace = None + self._reset_multi_line_flags() + + for entry in self.line_buffer: + if self._in_multi_line(entry): + self.stream.write(entry + "\n") + else: + entry = entry.expandtabs() + if stripspace is None and re.search(r"^[ \t]*[^# \t]", entry): + stripspace = re.match(r"^([ \t]*)", entry).group(1) + self.stream.write(self._indent_line(entry, stripspace) + "\n") + + self.line_buffer = [] + self._reset_multi_line_flags() + + +def adjust_whitespace(text): + """remove the left-whitespace margin of a block of Python code.""" + + state = [False, False] + (backslashed, triplequoted) = (0, 1) + + def in_multi_line(line): + start_state = (state[backslashed] or state[triplequoted]) + + if re.search(r"\\$", line): + state[backslashed] = True + else: + state[backslashed] = False + + def match(reg, t): + m = re.match(reg, t) + if m: + return m, t[len(m.group(0)):] + else: + return None, t + + while line: + if state[triplequoted]: + m, line = match(r"%s" % state[triplequoted], line) + if m: + state[triplequoted] = False + else: + m, line = match(r".*?(?=%s|$)" % state[triplequoted], line) + else: + m, line = match(r'#', line) + if m: + return start_state + + m, line = match(r"\"\"\"|\'\'\'", line) + if m: + state[triplequoted] = m.group(0) + continue + + m, line = match(r".*?(?=\"\"\"|\'\'\'|#|$)", line) + + return start_state + + def _indent_line(line, stripspace=''): + return re.sub(r"^%s" % stripspace, '', line) + + lines = [] + stripspace = None + + for line in re.split(r'\r?\n', text): + if in_multi_line(line): + lines.append(line) + else: + line = line.expandtabs() + if stripspace is None and re.search(r"^[ \t]*[^# \t]", line): + stripspace = re.match(r"^([ \t]*)", line).group(1) + lines.append(_indent_line(line, stripspace)) + return "\n".join(lines) diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/pyparser.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/pyparser.py new file mode 100644 index 00000000000..bfa46a9fafd --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/pyparser.py @@ -0,0 +1,232 @@ +# mako/pyparser.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + +"""Handles parsing of Python code. + +Parsing to AST is done via _ast on Python > 2.5, otherwise the compiler +module is used. +""" + +from mako import exceptions, util, compat +from mako.compat import arg_stringname +import operator + +if compat.py3k: + # words that cannot be assigned to (notably + # smaller than the total keys in __builtins__) + reserved = set(['True', 'False', 'None', 'print']) + + # the "id" attribute on a function node + arg_id = operator.attrgetter('arg') +else: + # words that cannot be assigned to (notably + # smaller than the total keys in __builtins__) + reserved = set(['True', 'False', 'None']) + + # the "id" attribute on a function node + arg_id = operator.attrgetter('id') + +import _ast +util.restore__ast(_ast) +from mako import _ast_util + + +def parse(code, mode='exec', **exception_kwargs): + """Parse an expression into AST""" + + try: + return _ast_util.parse(code, '', mode) + except Exception: + raise exceptions.SyntaxException( + "(%s) %s (%r)" % ( + compat.exception_as().__class__.__name__, + compat.exception_as(), + code[0:50] + ), **exception_kwargs) + + +class FindIdentifiers(_ast_util.NodeVisitor): + + def __init__(self, listener, **exception_kwargs): + self.in_function = False + self.in_assign_targets = False + self.local_ident_stack = set() + self.listener = listener + self.exception_kwargs = exception_kwargs + + def _add_declared(self, name): + if not self.in_function: + self.listener.declared_identifiers.add(name) + else: + self.local_ident_stack.add(name) + + def visit_ClassDef(self, node): + self._add_declared(node.name) + + def visit_Assign(self, node): + + # flip around the visiting of Assign so the expression gets + # evaluated first, in the case of a clause like "x=x+5" (x + # is undeclared) + + self.visit(node.value) + in_a = self.in_assign_targets + self.in_assign_targets = True + for n in node.targets: + self.visit(n) + self.in_assign_targets = in_a + + if compat.py3k: + + # ExceptHandler is in Python 2, but this block only works in + # Python 3 (and is required there) + + def visit_ExceptHandler(self, node): + if node.name is not None: + self._add_declared(node.name) + if node.type is not None: + self.visit(node.type) + for statement in node.body: + self.visit(statement) + + def visit_Lambda(self, node, *args): + self._visit_function(node, True) + + def visit_FunctionDef(self, node): + self._add_declared(node.name) + self._visit_function(node, False) + + def _expand_tuples(self, args): + for arg in args: + if isinstance(arg, _ast.Tuple): + for n in arg.elts: + yield n + else: + yield arg + + def _visit_function(self, node, islambda): + + # push function state onto stack. dont log any more + # identifiers as "declared" until outside of the function, + # but keep logging identifiers as "undeclared". track + # argument names in each function header so they arent + # counted as "undeclared" + + inf = self.in_function + self.in_function = True + + local_ident_stack = self.local_ident_stack + self.local_ident_stack = local_ident_stack.union([ + arg_id(arg) for arg in self._expand_tuples(node.args.args) + ]) + if islambda: + self.visit(node.body) + else: + for n in node.body: + self.visit(n) + self.in_function = inf + self.local_ident_stack = local_ident_stack + + def visit_For(self, node): + + # flip around visit + + self.visit(node.iter) + self.visit(node.target) + for statement in node.body: + self.visit(statement) + for statement in node.orelse: + self.visit(statement) + + def visit_Name(self, node): + if isinstance(node.ctx, _ast.Store): + # this is eqiuvalent to visit_AssName in + # compiler + self._add_declared(node.id) + elif node.id not in reserved and node.id \ + not in self.listener.declared_identifiers and node.id \ + not in self.local_ident_stack: + self.listener.undeclared_identifiers.add(node.id) + + def visit_Import(self, node): + for name in node.names: + if name.asname is not None: + self._add_declared(name.asname) + else: + self._add_declared(name.name.split('.')[0]) + + def visit_ImportFrom(self, node): + for name in node.names: + if name.asname is not None: + self._add_declared(name.asname) + else: + if name.name == '*': + raise exceptions.CompileException( + "'import *' is not supported, since all identifier " + "names must be explicitly declared. Please use the " + "form 'from import , , " + "...' instead.", **self.exception_kwargs) + self._add_declared(name.name) + + +class FindTuple(_ast_util.NodeVisitor): + + def __init__(self, listener, code_factory, **exception_kwargs): + self.listener = listener + self.exception_kwargs = exception_kwargs + self.code_factory = code_factory + + def visit_Tuple(self, node): + for n in node.elts: + p = self.code_factory(n, **self.exception_kwargs) + self.listener.codeargs.append(p) + self.listener.args.append(ExpressionGenerator(n).value()) + self.listener.declared_identifiers = \ + self.listener.declared_identifiers.union( + p.declared_identifiers) + self.listener.undeclared_identifiers = \ + self.listener.undeclared_identifiers.union( + p.undeclared_identifiers) + + +class ParseFunc(_ast_util.NodeVisitor): + + def __init__(self, listener, **exception_kwargs): + self.listener = listener + self.exception_kwargs = exception_kwargs + + def visit_FunctionDef(self, node): + self.listener.funcname = node.name + + argnames = [arg_id(arg) for arg in node.args.args] + if node.args.vararg: + argnames.append(arg_stringname(node.args.vararg)) + + if compat.py2k: + # kw-only args don't exist in Python 2 + kwargnames = [] + else: + kwargnames = [arg_id(arg) for arg in node.args.kwonlyargs] + if node.args.kwarg: + kwargnames.append(arg_stringname(node.args.kwarg)) + self.listener.argnames = argnames + self.listener.defaults = node.args.defaults # ast + self.listener.kwargnames = kwargnames + if compat.py2k: + self.listener.kwdefaults = [] + else: + self.listener.kwdefaults = node.args.kw_defaults + self.listener.varargs = node.args.vararg + self.listener.kwargs = node.args.kwarg + +class ExpressionGenerator(object): + + def __init__(self, astnode): + self.generator = _ast_util.SourceGenerator(' ' * 4) + self.generator.visit(astnode) + + def value(self): + return ''.join(self.generator.result) diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/runtime.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/runtime.py new file mode 100644 index 00000000000..6b6a35a9215 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/runtime.py @@ -0,0 +1,878 @@ +# mako/runtime.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + +"""provides runtime services for templates, including Context, +Namespace, and various helper functions.""" + +from mako import exceptions, util, compat +from mako.compat import compat_builtins +import sys + + +class Context(object): + """Provides runtime namespace, output buffer, and various + callstacks for templates. + + See :ref:`runtime_toplevel` for detail on the usage of + :class:`.Context`. + + """ + + def __init__(self, buffer, **data): + self._buffer_stack = [buffer] + + self._data = data + + self._kwargs = data.copy() + self._with_template = None + self._outputting_as_unicode = None + self.namespaces = {} + + # "capture" function which proxies to the + # generic "capture" function + self._data['capture'] = compat.partial(capture, self) + + # "caller" stack used by def calls with content + self.caller_stack = self._data['caller'] = CallerStack() + + def _set_with_template(self, t): + self._with_template = t + illegal_names = t.reserved_names.intersection(self._data) + if illegal_names: + raise exceptions.NameConflictError( + "Reserved words passed to render(): %s" % + ", ".join(illegal_names)) + + @property + def lookup(self): + """Return the :class:`.TemplateLookup` associated + with this :class:`.Context`. + + """ + return self._with_template.lookup + + @property + def kwargs(self): + """Return the dictionary of top level keyword arguments associated + with this :class:`.Context`. + + This dictionary only includes the top-level arguments passed to + :meth:`.Template.render`. It does not include names produced within + the template execution such as local variable names or special names + such as ``self``, ``next``, etc. + + The purpose of this dictionary is primarily for the case that + a :class:`.Template` accepts arguments via its ``<%page>`` tag, + which are normally expected to be passed via :meth:`.Template.render`, + except the template is being called in an inheritance context, + using the ``body()`` method. :attr:`.Context.kwargs` can then be + used to propagate these arguments to the inheriting template:: + + ${next.body(**context.kwargs)} + + """ + return self._kwargs.copy() + + def push_caller(self, caller): + """Push a ``caller`` callable onto the callstack for + this :class:`.Context`.""" + + + self.caller_stack.append(caller) + + def pop_caller(self): + """Pop a ``caller`` callable onto the callstack for this + :class:`.Context`.""" + + del self.caller_stack[-1] + + def keys(self): + """Return a list of all names established in this :class:`.Context`.""" + + return list(self._data.keys()) + + def __getitem__(self, key): + if key in self._data: + return self._data[key] + else: + return compat_builtins.__dict__[key] + + def _push_writer(self): + """push a capturing buffer onto this Context and return + the new writer function.""" + + buf = util.FastEncodingBuffer() + self._buffer_stack.append(buf) + return buf.write + + def _pop_buffer_and_writer(self): + """pop the most recent capturing buffer from this Context + and return the current writer after the pop. + + """ + + buf = self._buffer_stack.pop() + return buf, self._buffer_stack[-1].write + + def _push_buffer(self): + """push a capturing buffer onto this Context.""" + + self._push_writer() + + def _pop_buffer(self): + """pop the most recent capturing buffer from this Context.""" + + return self._buffer_stack.pop() + + def get(self, key, default=None): + """Return a value from this :class:`.Context`.""" + + return self._data.get(key, compat_builtins.__dict__.get(key, default)) + + def write(self, string): + """Write a string to this :class:`.Context` object's + underlying output buffer.""" + + self._buffer_stack[-1].write(string) + + def writer(self): + """Return the current writer function.""" + + return self._buffer_stack[-1].write + + def _copy(self): + c = Context.__new__(Context) + c._buffer_stack = self._buffer_stack + c._data = self._data.copy() + c._kwargs = self._kwargs + c._with_template = self._with_template + c._outputting_as_unicode = self._outputting_as_unicode + c.namespaces = self.namespaces + c.caller_stack = self.caller_stack + return c + + def _locals(self, d): + """Create a new :class:`.Context` with a copy of this + :class:`.Context`'s current state, + updated with the given dictionary. + + The :attr:`.Context.kwargs` collection remains + unaffected. + + + """ + + if not d: + return self + c = self._copy() + c._data.update(d) + return c + + def _clean_inheritance_tokens(self): + """create a new copy of this :class:`.Context`. with + tokens related to inheritance state removed.""" + + c = self._copy() + x = c._data + x.pop('self', None) + x.pop('parent', None) + x.pop('next', None) + return c + +class CallerStack(list): + def __init__(self): + self.nextcaller = None + + def __nonzero__(self): + return self.__bool__() + + def __bool__(self): + return len(self) and self._get_caller() and True or False + + def _get_caller(self): + # this method can be removed once + # codegen MAGIC_NUMBER moves past 7 + return self[-1] + + def __getattr__(self, key): + return getattr(self._get_caller(), key) + + def _push_frame(self): + frame = self.nextcaller or None + self.append(frame) + self.nextcaller = None + return frame + + def _pop_frame(self): + self.nextcaller = self.pop() + + +class Undefined(object): + """Represents an undefined value in a template. + + All template modules have a constant value + ``UNDEFINED`` present which is an instance of this + object. + + """ + def __str__(self): + raise NameError("Undefined") + + def __nonzero__(self): + return self.__bool__() + + def __bool__(self): + return False + +UNDEFINED = Undefined() + +class LoopStack(object): + """a stack for LoopContexts that implements the context manager protocol + to automatically pop off the top of the stack on context exit + """ + + def __init__(self): + self.stack = [] + + def _enter(self, iterable): + self._push(iterable) + return self._top + + def _exit(self): + self._pop() + return self._top + + @property + def _top(self): + if self.stack: + return self.stack[-1] + else: + return self + + def _pop(self): + return self.stack.pop() + + def _push(self, iterable): + new = LoopContext(iterable) + if self.stack: + new.parent = self.stack[-1] + return self.stack.append(new) + + def __getattr__(self, key): + raise exceptions.RuntimeException("No loop context is established") + + def __iter__(self): + return iter(self._top) + + +class LoopContext(object): + """A magic loop variable. + Automatically accessible in any ``% for`` block. + + See the section :ref:`loop_context` for usage + notes. + + :attr:`parent` -> :class:`.LoopContext` or ``None`` + The parent loop, if one exists. + :attr:`index` -> `int` + The 0-based iteration count. + :attr:`reverse_index` -> `int` + The number of iterations remaining. + :attr:`first` -> `bool` + ``True`` on the first iteration, ``False`` otherwise. + :attr:`last` -> `bool` + ``True`` on the last iteration, ``False`` otherwise. + :attr:`even` -> `bool` + ``True`` when ``index`` is even. + :attr:`odd` -> `bool` + ``True`` when ``index`` is odd. + """ + + def __init__(self, iterable): + self._iterable = iterable + self.index = 0 + self.parent = None + + def __iter__(self): + for i in self._iterable: + yield i + self.index += 1 + + @util.memoized_instancemethod + def __len__(self): + return len(self._iterable) + + @property + def reverse_index(self): + return len(self) - self.index - 1 + + @property + def first(self): + return self.index == 0 + + @property + def last(self): + return self.index == len(self) - 1 + + @property + def even(self): + return not self.odd + + @property + def odd(self): + return bool(self.index % 2) + + def cycle(self, *values): + """Cycle through values as the loop progresses. + """ + if not values: + raise ValueError("You must provide values to cycle through") + return values[self.index % len(values)] + + +class _NSAttr(object): + def __init__(self, parent): + self.__parent = parent + def __getattr__(self, key): + ns = self.__parent + while ns: + if hasattr(ns.module, key): + return getattr(ns.module, key) + else: + ns = ns.inherits + raise AttributeError(key) + +class Namespace(object): + """Provides access to collections of rendering methods, which + can be local, from other templates, or from imported modules. + + To access a particular rendering method referenced by a + :class:`.Namespace`, use plain attribute access: + + .. sourcecode:: mako + + ${some_namespace.foo(x, y, z)} + + :class:`.Namespace` also contains several built-in attributes + described here. + + """ + + def __init__(self, name, context, + callables=None, inherits=None, + populate_self=True, calling_uri=None): + self.name = name + self.context = context + self.inherits = inherits + if callables is not None: + self.callables = dict([(c.__name__, c) for c in callables]) + + callables = () + + module = None + """The Python module referenced by this :class:`.Namespace`. + + If the namespace references a :class:`.Template`, then + this module is the equivalent of ``template.module``, + i.e. the generated module for the template. + + """ + + template = None + """The :class:`.Template` object referenced by this + :class:`.Namespace`, if any. + + """ + + context = None + """The :class:`.Context` object for this :class:`.Namespace`. + + Namespaces are often created with copies of contexts that + contain slightly different data, particularly in inheritance + scenarios. Using the :class:`.Context` off of a :class:`.Namespace` one + can traverse an entire chain of templates that inherit from + one-another. + + """ + + filename = None + """The path of the filesystem file used for this + :class:`.Namespace`'s module or template. + + If this is a pure module-based + :class:`.Namespace`, this evaluates to ``module.__file__``. If a + template-based namespace, it evaluates to the original + template file location. + + """ + + uri = None + """The URI for this :class:`.Namespace`'s template. + + I.e. whatever was sent to :meth:`.TemplateLookup.get_template()`. + + This is the equivalent of :attr:`.Template.uri`. + + """ + + _templateuri = None + + @util.memoized_property + def attr(self): + """Access module level attributes by name. + + This accessor allows templates to supply "scalar" + attributes which are particularly handy in inheritance + relationships. + + .. seealso:: + + :ref:`inheritance_attr` + + :ref:`namespace_attr_for_includes` + + """ + return _NSAttr(self) + + def get_namespace(self, uri): + """Return a :class:`.Namespace` corresponding to the given ``uri``. + + If the given ``uri`` is a relative URI (i.e. it does not + contain a leading slash ``/``), the ``uri`` is adjusted to + be relative to the ``uri`` of the namespace itself. This + method is therefore mostly useful off of the built-in + ``local`` namespace, described in :ref:`namespace_local`. + + In + most cases, a template wouldn't need this function, and + should instead use the ``<%namespace>`` tag to load + namespaces. However, since all ``<%namespace>`` tags are + evaluated before the body of a template ever runs, + this method can be used to locate namespaces using + expressions that were generated within the body code of + the template, or to conditionally use a particular + namespace. + + """ + key = (self, uri) + if key in self.context.namespaces: + return self.context.namespaces[key] + else: + ns = TemplateNamespace(uri, self.context._copy(), + templateuri=uri, + calling_uri=self._templateuri) + self.context.namespaces[key] = ns + return ns + + def get_template(self, uri): + """Return a :class:`.Template` from the given ``uri``. + + The ``uri`` resolution is relative to the ``uri`` of this + :class:`.Namespace` object's :class:`.Template`. + + """ + return _lookup_template(self.context, uri, self._templateuri) + + def get_cached(self, key, **kwargs): + """Return a value from the :class:`.Cache` referenced by this + :class:`.Namespace` object's :class:`.Template`. + + The advantage to this method versus direct access to the + :class:`.Cache` is that the configuration parameters + declared in ``<%page>`` take effect here, thereby calling + up the same configured backend as that configured + by ``<%page>``. + + """ + + return self.cache.get(key, **kwargs) + + @property + def cache(self): + """Return the :class:`.Cache` object referenced + by this :class:`.Namespace` object's + :class:`.Template`. + + """ + return self.template.cache + + def include_file(self, uri, **kwargs): + """Include a file at the given ``uri``.""" + + _include_file(self.context, uri, self._templateuri, **kwargs) + + def _populate(self, d, l): + for ident in l: + if ident == '*': + for (k, v) in self._get_star(): + d[k] = v + else: + d[ident] = getattr(self, ident) + + def _get_star(self): + if self.callables: + for key in self.callables: + yield (key, self.callables[key]) + + def __getattr__(self, key): + if key in self.callables: + val = self.callables[key] + elif self.inherits: + val = getattr(self.inherits, key) + else: + raise AttributeError( + "Namespace '%s' has no member '%s'" % + (self.name, key)) + setattr(self, key, val) + return val + +class TemplateNamespace(Namespace): + """A :class:`.Namespace` specific to a :class:`.Template` instance.""" + + def __init__(self, name, context, template=None, templateuri=None, + callables=None, inherits=None, + populate_self=True, calling_uri=None): + self.name = name + self.context = context + self.inherits = inherits + if callables is not None: + self.callables = dict([(c.__name__, c) for c in callables]) + + if templateuri is not None: + self.template = _lookup_template(context, templateuri, + calling_uri) + self._templateuri = self.template.module._template_uri + elif template is not None: + self.template = template + self._templateuri = template.module._template_uri + else: + raise TypeError("'template' argument is required.") + + if populate_self: + lclcallable, lclcontext = \ + _populate_self_namespace(context, self.template, + self_ns=self) + + @property + def module(self): + """The Python module referenced by this :class:`.Namespace`. + + If the namespace references a :class:`.Template`, then + this module is the equivalent of ``template.module``, + i.e. the generated module for the template. + + """ + return self.template.module + + @property + def filename(self): + """The path of the filesystem file used for this + :class:`.Namespace`'s module or template. + """ + return self.template.filename + + @property + def uri(self): + """The URI for this :class:`.Namespace`'s template. + + I.e. whatever was sent to :meth:`.TemplateLookup.get_template()`. + + This is the equivalent of :attr:`.Template.uri`. + + """ + return self.template.uri + + def _get_star(self): + if self.callables: + for key in self.callables: + yield (key, self.callables[key]) + def get(key): + callable_ = self.template._get_def_callable(key) + return compat.partial(callable_, self.context) + for k in self.template.module._exports: + yield (k, get(k)) + + def __getattr__(self, key): + if key in self.callables: + val = self.callables[key] + elif self.template.has_def(key): + callable_ = self.template._get_def_callable(key) + val = compat.partial(callable_, self.context) + elif self.inherits: + val = getattr(self.inherits, key) + + else: + raise AttributeError( + "Namespace '%s' has no member '%s'" % + (self.name, key)) + setattr(self, key, val) + return val + +class ModuleNamespace(Namespace): + """A :class:`.Namespace` specific to a Python module instance.""" + + def __init__(self, name, context, module, + callables=None, inherits=None, + populate_self=True, calling_uri=None): + self.name = name + self.context = context + self.inherits = inherits + if callables is not None: + self.callables = dict([(c.__name__, c) for c in callables]) + + mod = __import__(module) + for token in module.split('.')[1:]: + mod = getattr(mod, token) + self.module = mod + + @property + def filename(self): + """The path of the filesystem file used for this + :class:`.Namespace`'s module or template. + """ + return self.module.__file__ + + def _get_star(self): + if self.callables: + for key in self.callables: + yield (key, self.callables[key]) + for key in dir(self.module): + if key[0] != '_': + callable_ = getattr(self.module, key) + if compat.callable(callable_): + yield key, compat.partial(callable_, self.context) + + + def __getattr__(self, key): + if key in self.callables: + val = self.callables[key] + elif hasattr(self.module, key): + callable_ = getattr(self.module, key) + val = compat.partial(callable_, self.context) + elif self.inherits: + val = getattr(self.inherits, key) + else: + raise AttributeError( + "Namespace '%s' has no member '%s'" % + (self.name, key)) + setattr(self, key, val) + return val + +def supports_caller(func): + """Apply a caller_stack compatibility decorator to a plain + Python function. + + See the example in :ref:`namespaces_python_modules`. + + """ + + def wrap_stackframe(context, *args, **kwargs): + context.caller_stack._push_frame() + try: + return func(context, *args, **kwargs) + finally: + context.caller_stack._pop_frame() + return wrap_stackframe + +def capture(context, callable_, *args, **kwargs): + """Execute the given template def, capturing the output into + a buffer. + + See the example in :ref:`namespaces_python_modules`. + + """ + + if not compat.callable(callable_): + raise exceptions.RuntimeException( + "capture() function expects a callable as " + "its argument (i.e. capture(func, *args, **kwargs))" + ) + context._push_buffer() + try: + callable_(*args, **kwargs) + finally: + buf = context._pop_buffer() + return buf.getvalue() + +def _decorate_toplevel(fn): + def decorate_render(render_fn): + def go(context, *args, **kw): + def y(*args, **kw): + return render_fn(context, *args, **kw) + try: + y.__name__ = render_fn.__name__[7:] + except TypeError: + # < Python 2.4 + pass + return fn(y)(context, *args, **kw) + return go + return decorate_render + +def _decorate_inline(context, fn): + def decorate_render(render_fn): + dec = fn(render_fn) + def go(*args, **kw): + return dec(context, *args, **kw) + return go + return decorate_render + +def _include_file(context, uri, calling_uri, **kwargs): + """locate the template from the given uri and include it in + the current output.""" + + template = _lookup_template(context, uri, calling_uri) + (callable_, ctx) = _populate_self_namespace( + context._clean_inheritance_tokens(), + template) + callable_(ctx, **_kwargs_for_include(callable_, context._data, **kwargs)) + +def _inherit_from(context, uri, calling_uri): + """called by the _inherit method in template modules to set + up the inheritance chain at the start of a template's + execution.""" + + if uri is None: + return None + template = _lookup_template(context, uri, calling_uri) + self_ns = context['self'] + ih = self_ns + while ih.inherits is not None: + ih = ih.inherits + lclcontext = context._locals({'next': ih}) + ih.inherits = TemplateNamespace("self:%s" % template.uri, + lclcontext, + template=template, + populate_self=False) + context._data['parent'] = lclcontext._data['local'] = ih.inherits + callable_ = getattr(template.module, '_mako_inherit', None) + if callable_ is not None: + ret = callable_(template, lclcontext) + if ret: + return ret + + gen_ns = getattr(template.module, '_mako_generate_namespaces', None) + if gen_ns is not None: + gen_ns(context) + return (template.callable_, lclcontext) + +def _lookup_template(context, uri, relativeto): + lookup = context._with_template.lookup + if lookup is None: + raise exceptions.TemplateLookupException( + "Template '%s' has no TemplateLookup associated" % + context._with_template.uri) + uri = lookup.adjust_uri(uri, relativeto) + try: + return lookup.get_template(uri) + except exceptions.TopLevelLookupException: + raise exceptions.TemplateLookupException(str(compat.exception_as())) + +def _populate_self_namespace(context, template, self_ns=None): + if self_ns is None: + self_ns = TemplateNamespace('self:%s' % template.uri, + context, template=template, + populate_self=False) + context._data['self'] = context._data['local'] = self_ns + if hasattr(template.module, '_mako_inherit'): + ret = template.module._mako_inherit(template, context) + if ret: + return ret + return (template.callable_, context) + +def _render(template, callable_, args, data, as_unicode=False): + """create a Context and return the string + output of the given template and template callable.""" + + if as_unicode: + buf = util.FastEncodingBuffer(as_unicode=True) + elif template.bytestring_passthrough: + buf = compat.StringIO() + else: + buf = util.FastEncodingBuffer( + as_unicode=as_unicode, + encoding=template.output_encoding, + errors=template.encoding_errors) + context = Context(buf, **data) + context._outputting_as_unicode = as_unicode + context._set_with_template(template) + + _render_context(template, callable_, context, *args, + **_kwargs_for_callable(callable_, data)) + return context._pop_buffer().getvalue() + +def _kwargs_for_callable(callable_, data): + argspec = compat.inspect_func_args(callable_) + # for normal pages, **pageargs is usually present + if argspec[2]: + return data + + # for rendering defs from the top level, figure out the args + namedargs = argspec[0] + [v for v in argspec[1:3] if v is not None] + kwargs = {} + for arg in namedargs: + if arg != 'context' and arg in data and arg not in kwargs: + kwargs[arg] = data[arg] + return kwargs + +def _kwargs_for_include(callable_, data, **kwargs): + argspec = compat.inspect_func_args(callable_) + namedargs = argspec[0] + [v for v in argspec[1:3] if v is not None] + for arg in namedargs: + if arg != 'context' and arg in data and arg not in kwargs: + kwargs[arg] = data[arg] + return kwargs + +def _render_context(tmpl, callable_, context, *args, **kwargs): + import mako.template as template + # create polymorphic 'self' namespace for this + # template with possibly updated context + if not isinstance(tmpl, template.DefTemplate): + # if main render method, call from the base of the inheritance stack + (inherit, lclcontext) = _populate_self_namespace(context, tmpl) + _exec_template(inherit, lclcontext, args=args, kwargs=kwargs) + else: + # otherwise, call the actual rendering method specified + (inherit, lclcontext) = _populate_self_namespace(context, tmpl.parent) + _exec_template(callable_, context, args=args, kwargs=kwargs) + +def _exec_template(callable_, context, args=None, kwargs=None): + """execute a rendering callable given the callable, a + Context, and optional explicit arguments + + the contextual Template will be located if it exists, and + the error handling options specified on that Template will + be interpreted here. + """ + template = context._with_template + if template is not None and \ + (template.format_exceptions or template.error_handler): + try: + callable_(context, *args, **kwargs) + except Exception: + _render_error(template, context, compat.exception_as()) + except: + e = sys.exc_info()[0] + _render_error(template, context, e) + else: + callable_(context, *args, **kwargs) + +def _render_error(template, context, error): + if template.error_handler: + result = template.error_handler(context, error) + if not result: + compat.reraise(*sys.exc_info()) + else: + error_template = exceptions.html_error_template() + if context._outputting_as_unicode: + context._buffer_stack[:] = [ + util.FastEncodingBuffer(as_unicode=True)] + else: + context._buffer_stack[:] = [util.FastEncodingBuffer( + error_template.output_encoding, + error_template.encoding_errors)] + + context._set_with_template(error_template) + error_template.render_context(context, error=error) diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/template.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/template.py new file mode 100644 index 00000000000..fb6106289fa --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/template.py @@ -0,0 +1,705 @@ +# mako/template.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + +"""Provides the Template class, a facade for parsing, generating and executing +template strings, as well as template runtime operations.""" + +from mako.lexer import Lexer +from mako import runtime, util, exceptions, codegen, cache, compat +import os +import re +import shutil +import stat +import sys +import tempfile +import types +import weakref + + +class Template(object): + """Represents a compiled template. + + :class:`.Template` includes a reference to the original + template source (via the :attr:`.source` attribute) + as well as the source code of the + generated Python module (i.e. the :attr:`.code` attribute), + as well as a reference to an actual Python module. + + :class:`.Template` is constructed using either a literal string + representing the template text, or a filename representing a filesystem + path to a source file. + + :param text: textual template source. This argument is mutually + exclusive versus the ``filename`` parameter. + + :param filename: filename of the source template. This argument is + mutually exclusive versus the ``text`` parameter. + + :param buffer_filters: string list of filters to be applied + to the output of ``%def``\ s which are buffered, cached, or otherwise + filtered, after all filters + defined with the ``%def`` itself have been applied. Allows the + creation of default expression filters that let the output + of return-valued ``%def``\ s "opt out" of that filtering via + passing special attributes or objects. + + :param bytestring_passthrough: When ``True``, and ``output_encoding`` is + set to ``None``, and :meth:`.Template.render` is used to render, + the `StringIO` or `cStringIO` buffer will be used instead of the + default "fast" buffer. This allows raw bytestrings in the + output stream, such as in expressions, to pass straight + through to the buffer. This flag is forced + to ``True`` if ``disable_unicode`` is also configured. + + .. versionadded:: 0.4 + Added to provide the same behavior as that of the previous series. + + :param cache_args: Dictionary of cache configuration arguments that + will be passed to the :class:`.CacheImpl`. See :ref:`caching_toplevel`. + + :param cache_dir: + + .. deprecated:: 0.6 + Use the ``'dir'`` argument in the ``cache_args`` dictionary. + See :ref:`caching_toplevel`. + + :param cache_enabled: Boolean flag which enables caching of this + template. See :ref:`caching_toplevel`. + + :param cache_impl: String name of a :class:`.CacheImpl` caching + implementation to use. Defaults to ``'beaker'``. + + :param cache_type: + + .. deprecated:: 0.6 + Use the ``'type'`` argument in the ``cache_args`` dictionary. + See :ref:`caching_toplevel`. + + :param cache_url: + + .. deprecated:: 0.6 + Use the ``'url'`` argument in the ``cache_args`` dictionary. + See :ref:`caching_toplevel`. + + :param default_filters: List of string filter names that will + be applied to all expressions. See :ref:`filtering_default_filters`. + + :param disable_unicode: Disables all awareness of Python Unicode + objects. See :ref:`unicode_disabled`. + + :param enable_loop: When ``True``, enable the ``loop`` context variable. + This can be set to ``False`` to support templates that may + be making usage of the name "``loop``". Individual templates can + re-enable the "loop" context by placing the directive + ``enable_loop="True"`` inside the ``<%page>`` tag -- see + :ref:`migrating_loop`. + + :param encoding_errors: Error parameter passed to ``encode()`` when + string encoding is performed. See :ref:`usage_unicode`. + + :param error_handler: Python callable which is called whenever + compile or runtime exceptions occur. The callable is passed + the current context as well as the exception. If the + callable returns ``True``, the exception is considered to + be handled, else it is re-raised after the function + completes. Is used to provide custom error-rendering + functions. + + :param format_exceptions: if ``True``, exceptions which occur during + the render phase of this template will be caught and + formatted into an HTML error page, which then becomes the + rendered result of the :meth:`.render` call. Otherwise, + runtime exceptions are propagated outwards. + + :param imports: String list of Python statements, typically individual + "import" lines, which will be placed into the module level + preamble of all generated Python modules. See the example + in :ref:`filtering_default_filters`. + + :param future_imports: String list of names to import from `__future__`. + These will be concatenated into a comma-separated string and inserted + into the beginning of the template, e.g. ``futures_imports=['FOO', + 'BAR']`` results in ``from __future__ import FOO, BAR``. If you're + interested in using features like the new division operator, you must + use future_imports to convey that to the renderer, as otherwise the + import will not appear as the first executed statement in the generated + code and will therefore not have the desired effect. + + :param input_encoding: Encoding of the template's source code. Can + be used in lieu of the coding comment. See + :ref:`usage_unicode` as well as :ref:`unicode_toplevel` for + details on source encoding. + + :param lookup: a :class:`.TemplateLookup` instance that will be used + for all file lookups via the ``<%namespace>``, + ``<%include>``, and ``<%inherit>`` tags. See + :ref:`usage_templatelookup`. + + :param module_directory: Filesystem location where generated + Python module files will be placed. + + :param module_filename: Overrides the filename of the generated + Python module file. For advanced usage only. + + :param module_writer: A callable which overrides how the Python + module is written entirely. The callable is passed the + encoded source content of the module and the destination + path to be written to. The default behavior of module writing + uses a tempfile in conjunction with a file move in order + to make the operation atomic. So a user-defined module + writing function that mimics the default behavior would be: + + .. sourcecode:: python + + import tempfile + import os + import shutil + + def module_writer(source, outputpath): + (dest, name) = \\ + tempfile.mkstemp( + dir=os.path.dirname(outputpath) + ) + + os.write(dest, source) + os.close(dest) + shutil.move(name, outputpath) + + from mako.template import Template + mytemplate = Template( + filename="index.html", + module_directory="/path/to/modules", + module_writer=module_writer + ) + + The function is provided for unusual configurations where + certain platform-specific permissions or other special + steps are needed. + + :param output_encoding: The encoding to use when :meth:`.render` + is called. + See :ref:`usage_unicode` as well as :ref:`unicode_toplevel`. + + :param preprocessor: Python callable which will be passed + the full template source before it is parsed. The return + result of the callable will be used as the template source + code. + + :param lexer_cls: A :class:`.Lexer` class used to parse + the template. The :class:`.Lexer` class is used by + default. + + .. versionadded:: 0.7.4 + + :param strict_undefined: Replaces the automatic usage of + ``UNDEFINED`` for any undeclared variables not located in + the :class:`.Context` with an immediate raise of + ``NameError``. The advantage is immediate reporting of + missing variables which include the name. + + .. versionadded:: 0.3.6 + + :param uri: string URI or other identifier for this template. + If not provided, the ``uri`` is generated from the filesystem + path, or from the in-memory identity of a non-file-based + template. The primary usage of the ``uri`` is to provide a key + within :class:`.TemplateLookup`, as well as to generate the + file path of the generated Python module file, if + ``module_directory`` is specified. + + """ + + lexer_cls = Lexer + + def __init__(self, + text=None, + filename=None, + uri=None, + format_exceptions=False, + error_handler=None, + lookup=None, + output_encoding=None, + encoding_errors='strict', + module_directory=None, + cache_args=None, + cache_impl='beaker', + cache_enabled=True, + cache_type=None, + cache_dir=None, + cache_url=None, + module_filename=None, + input_encoding=None, + disable_unicode=False, + module_writer=None, + bytestring_passthrough=False, + default_filters=None, + buffer_filters=(), + strict_undefined=False, + imports=None, + future_imports=None, + enable_loop=True, + preprocessor=None, + lexer_cls=None): + if uri: + self.module_id = re.sub(r'\W', "_", uri) + self.uri = uri + elif filename: + self.module_id = re.sub(r'\W', "_", filename) + drive, path = os.path.splitdrive(filename) + path = os.path.normpath(path).replace(os.path.sep, "/") + self.uri = path + else: + self.module_id = "memory:" + hex(id(self)) + self.uri = self.module_id + + u_norm = self.uri + if u_norm.startswith("/"): + u_norm = u_norm[1:] + u_norm = os.path.normpath(u_norm) + if u_norm.startswith(".."): + raise exceptions.TemplateLookupException( + "Template uri \"%s\" is invalid - " + "it cannot be relative outside " + "of the root path." % self.uri) + + self.input_encoding = input_encoding + self.output_encoding = output_encoding + self.encoding_errors = encoding_errors + self.disable_unicode = disable_unicode + self.bytestring_passthrough = bytestring_passthrough or disable_unicode + self.enable_loop = enable_loop + self.strict_undefined = strict_undefined + self.module_writer = module_writer + + if compat.py3k and disable_unicode: + raise exceptions.UnsupportedError( + "Mako for Python 3 does not " + "support disabling Unicode") + elif output_encoding and disable_unicode: + raise exceptions.UnsupportedError( + "output_encoding must be set to " + "None when disable_unicode is used.") + if default_filters is None: + if compat.py3k or self.disable_unicode: + self.default_filters = ['str'] + else: + self.default_filters = ['unicode'] + else: + self.default_filters = default_filters + self.buffer_filters = buffer_filters + + self.imports = imports + self.future_imports = future_imports + self.preprocessor = preprocessor + + if lexer_cls is not None: + self.lexer_cls = lexer_cls + + # if plain text, compile code in memory only + if text is not None: + (code, module) = _compile_text(self, text, filename) + self._code = code + self._source = text + ModuleInfo(module, None, self, filename, code, text) + elif filename is not None: + # if template filename and a module directory, load + # a filesystem-based module file, generating if needed + if module_filename is not None: + path = module_filename + elif module_directory is not None: + path = os.path.abspath( + os.path.join( + os.path.normpath(module_directory), + u_norm + ".py" + ) + ) + else: + path = None + module = self._compile_from_file(path, filename) + else: + raise exceptions.RuntimeException( + "Template requires text or filename") + + self.module = module + self.filename = filename + self.callable_ = self.module.render_body + self.format_exceptions = format_exceptions + self.error_handler = error_handler + self.lookup = lookup + + self.module_directory = module_directory + + self._setup_cache_args( + cache_impl, cache_enabled, cache_args, + cache_type, cache_dir, cache_url + ) + + + @util.memoized_property + def reserved_names(self): + if self.enable_loop: + return codegen.RESERVED_NAMES + else: + return codegen.RESERVED_NAMES.difference(['loop']) + + def _setup_cache_args(self, + cache_impl, cache_enabled, cache_args, + cache_type, cache_dir, cache_url): + self.cache_impl = cache_impl + self.cache_enabled = cache_enabled + if cache_args: + self.cache_args = cache_args + else: + self.cache_args = {} + + # transfer deprecated cache_* args + if cache_type: + self.cache_args['type'] = cache_type + if cache_dir: + self.cache_args['dir'] = cache_dir + if cache_url: + self.cache_args['url'] = cache_url + + def _compile_from_file(self, path, filename): + if path is not None: + util.verify_directory(os.path.dirname(path)) + filemtime = os.stat(filename)[stat.ST_MTIME] + if not os.path.exists(path) or \ + os.stat(path)[stat.ST_MTIME] < filemtime: + data = util.read_file(filename) + _compile_module_file( + self, + data, + filename, + path, + self.module_writer) + module = compat.load_module(self.module_id, path) + del sys.modules[self.module_id] + if module._magic_number != codegen.MAGIC_NUMBER: + data = util.read_file(filename) + _compile_module_file( + self, + data, + filename, + path, + self.module_writer) + module = compat.load_module(self.module_id, path) + del sys.modules[self.module_id] + ModuleInfo(module, path, self, filename, None, None) + else: + # template filename and no module directory, compile code + # in memory + data = util.read_file(filename) + code, module = _compile_text( + self, + data, + filename) + self._source = None + self._code = code + ModuleInfo(module, None, self, filename, code, None) + return module + + @property + def source(self): + """Return the template source code for this :class:`.Template`.""" + + return _get_module_info_from_callable(self.callable_).source + + @property + def code(self): + """Return the module source code for this :class:`.Template`.""" + + return _get_module_info_from_callable(self.callable_).code + + @util.memoized_property + def cache(self): + return cache.Cache(self) + + @property + def cache_dir(self): + return self.cache_args['dir'] + @property + def cache_url(self): + return self.cache_args['url'] + @property + def cache_type(self): + return self.cache_args['type'] + + def render(self, *args, **data): + """Render the output of this template as a string. + + If the template specifies an output encoding, the string + will be encoded accordingly, else the output is raw (raw + output uses `cStringIO` and can't handle multibyte + characters). A :class:`.Context` object is created corresponding + to the given data. Arguments that are explicitly declared + by this template's internal rendering method are also + pulled from the given ``*args``, ``**data`` members. + + """ + return runtime._render(self, self.callable_, args, data) + + def render_unicode(self, *args, **data): + """Render the output of this template as a unicode object.""" + + return runtime._render(self, + self.callable_, + args, + data, + as_unicode=True) + + def render_context(self, context, *args, **kwargs): + """Render this :class:`.Template` with the given context. + + The data is written to the context's buffer. + + """ + if getattr(context, '_with_template', None) is None: + context._set_with_template(self) + runtime._render_context(self, + self.callable_, + context, + *args, + **kwargs) + + def has_def(self, name): + return hasattr(self.module, "render_%s" % name) + + def get_def(self, name): + """Return a def of this template as a :class:`.DefTemplate`.""" + + return DefTemplate(self, getattr(self.module, "render_%s" % name)) + + def _get_def_callable(self, name): + return getattr(self.module, "render_%s" % name) + + @property + def last_modified(self): + return self.module._modified_time + +class ModuleTemplate(Template): + """A Template which is constructed given an existing Python module. + + e.g.:: + + t = Template("this is a template") + f = file("mymodule.py", "w") + f.write(t.code) + f.close() + + import mymodule + + t = ModuleTemplate(mymodule) + print t.render() + + """ + + def __init__(self, module, + module_filename=None, + template=None, + template_filename=None, + module_source=None, + template_source=None, + output_encoding=None, + encoding_errors='strict', + disable_unicode=False, + bytestring_passthrough=False, + format_exceptions=False, + error_handler=None, + lookup=None, + cache_args=None, + cache_impl='beaker', + cache_enabled=True, + cache_type=None, + cache_dir=None, + cache_url=None, + ): + self.module_id = re.sub(r'\W', "_", module._template_uri) + self.uri = module._template_uri + self.input_encoding = module._source_encoding + self.output_encoding = output_encoding + self.encoding_errors = encoding_errors + self.disable_unicode = disable_unicode + self.bytestring_passthrough = bytestring_passthrough or disable_unicode + self.enable_loop = module._enable_loop + + if compat.py3k and disable_unicode: + raise exceptions.UnsupportedError( + "Mako for Python 3 does not " + "support disabling Unicode") + elif output_encoding and disable_unicode: + raise exceptions.UnsupportedError( + "output_encoding must be set to " + "None when disable_unicode is used.") + + self.module = module + self.filename = template_filename + ModuleInfo(module, + module_filename, + self, + template_filename, + module_source, + template_source) + + self.callable_ = self.module.render_body + self.format_exceptions = format_exceptions + self.error_handler = error_handler + self.lookup = lookup + self._setup_cache_args( + cache_impl, cache_enabled, cache_args, + cache_type, cache_dir, cache_url + ) + +class DefTemplate(Template): + """A :class:`.Template` which represents a callable def in a parent + template.""" + + def __init__(self, parent, callable_): + self.parent = parent + self.callable_ = callable_ + self.output_encoding = parent.output_encoding + self.module = parent.module + self.encoding_errors = parent.encoding_errors + self.format_exceptions = parent.format_exceptions + self.error_handler = parent.error_handler + self.enable_loop = parent.enable_loop + self.lookup = parent.lookup + self.bytestring_passthrough = parent.bytestring_passthrough + + def get_def(self, name): + return self.parent.get_def(name) + +class ModuleInfo(object): + """Stores information about a module currently loaded into + memory, provides reverse lookups of template source, module + source code based on a module's identifier. + + """ + _modules = weakref.WeakValueDictionary() + + def __init__(self, + module, + module_filename, + template, + template_filename, + module_source, + template_source): + self.module = module + self.module_filename = module_filename + self.template_filename = template_filename + self.module_source = module_source + self.template_source = template_source + self._modules[module.__name__] = template._mmarker = self + if module_filename: + self._modules[module_filename] = self + + @classmethod + def get_module_source_metadata(cls, module_source, full_line_map=False): + source_map = re.search( + r"__M_BEGIN_METADATA(.+?)__M_END_METADATA", + module_source, re.S).group(1) + source_map = compat.json.loads(source_map) + source_map['line_map'] = dict((int(k), int(v)) + for k, v in source_map['line_map'].items()) + if full_line_map: + f_line_map = source_map['full_line_map'] = [] + line_map = source_map['line_map'] + + curr_templ_line = 1 + for mod_line in range(1, max(line_map)): + if mod_line in line_map: + curr_templ_line = line_map[mod_line] + f_line_map.append(curr_templ_line) + return source_map + + @property + def code(self): + if self.module_source is not None: + return self.module_source + else: + return util.read_python_file(self.module_filename) + + @property + def source(self): + if self.template_source is not None: + if self.module._source_encoding and \ + not isinstance(self.template_source, compat.text_type): + return self.template_source.decode( + self.module._source_encoding) + else: + return self.template_source + else: + data = util.read_file(self.template_filename) + if self.module._source_encoding: + return data.decode(self.module._source_encoding) + else: + return data + +def _compile(template, text, filename, generate_magic_comment): + lexer = template.lexer_cls(text, + filename, + disable_unicode=template.disable_unicode, + input_encoding=template.input_encoding, + preprocessor=template.preprocessor) + node = lexer.parse() + source = codegen.compile(node, + template.uri, + filename, + default_filters=template.default_filters, + buffer_filters=template.buffer_filters, + imports=template.imports, + future_imports=template.future_imports, + source_encoding=lexer.encoding, + generate_magic_comment=generate_magic_comment, + disable_unicode=template.disable_unicode, + strict_undefined=template.strict_undefined, + enable_loop=template.enable_loop, + reserved_names=template.reserved_names) + return source, lexer + +def _compile_text(template, text, filename): + identifier = template.module_id + source, lexer = _compile(template, text, filename, + generate_magic_comment=template.disable_unicode) + + cid = identifier + if not compat.py3k and isinstance(cid, compat.text_type): + cid = cid.encode() + module = types.ModuleType(cid) + code = compile(source, cid, 'exec') + + # this exec() works for 2.4->3.3. + exec(code, module.__dict__, module.__dict__) + return (source, module) + +def _compile_module_file(template, text, filename, outputpath, module_writer): + source, lexer = _compile(template, text, filename, + generate_magic_comment=True) + + if isinstance(source, compat.text_type): + source = source.encode(lexer.encoding or 'ascii') + + if module_writer: + module_writer(source, outputpath) + else: + # make tempfiles in the same location as the ultimate + # location. this ensures they're on the same filesystem, + # avoiding synchronization issues. + (dest, name) = tempfile.mkstemp(dir=os.path.dirname(outputpath)) + + os.write(dest, source) + os.close(dest) + shutil.move(name, outputpath) + +def _get_module_info_from_callable(callable_): + if compat.py3k: + return _get_module_info(callable_.__globals__['__name__']) + else: + return _get_module_info(callable_.func_globals['__name__']) + +def _get_module_info(filename): + return ModuleInfo._modules[filename] + diff --git a/src/gallium/drivers/swr/rasterizer/scripts/mako/util.py b/src/gallium/drivers/swr/rasterizer/scripts/mako/util.py new file mode 100644 index 00000000000..cba2ab7920c --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/mako/util.py @@ -0,0 +1,360 @@ +# mako/util.py +# Copyright (C) 2006-2015 the Mako authors and contributors +# +# This module is part of Mako and is released under +# the MIT License: http://www.opensource.org/licenses/mit-license.php + +import re +import collections +import codecs +import os +from mako import compat +import operator + +def update_wrapper(decorated, fn): + decorated.__wrapped__ = fn + decorated.__name__ = fn.__name__ + return decorated + + +class PluginLoader(object): + def __init__(self, group): + self.group = group + self.impls = {} + + def load(self, name): + if name in self.impls: + return self.impls[name]() + else: + import pkg_resources + for impl in pkg_resources.iter_entry_points( + self.group, + name): + self.impls[name] = impl.load + return impl.load() + else: + from mako import exceptions + raise exceptions.RuntimeException( + "Can't load plugin %s %s" % + (self.group, name)) + + def register(self, name, modulepath, objname): + def load(): + mod = __import__(modulepath) + for token in modulepath.split(".")[1:]: + mod = getattr(mod, token) + return getattr(mod, objname) + self.impls[name] = load + +def verify_directory(dir): + """create and/or verify a filesystem directory.""" + + tries = 0 + + while not os.path.exists(dir): + try: + tries += 1 + os.makedirs(dir, compat.octal("0775")) + except: + if tries > 5: + raise + +def to_list(x, default=None): + if x is None: + return default + if not isinstance(x, (list, tuple)): + return [x] + else: + return x + + +class memoized_property(object): + """A read-only @property that is only evaluated once.""" + def __init__(self, fget, doc=None): + self.fget = fget + self.__doc__ = doc or fget.__doc__ + self.__name__ = fget.__name__ + + def __get__(self, obj, cls): + if obj is None: + return self + obj.__dict__[self.__name__] = result = self.fget(obj) + return result + +class memoized_instancemethod(object): + """Decorate a method memoize its return value. + + Best applied to no-arg methods: memoization is not sensitive to + argument values, and will always return the same value even when + called with different arguments. + + """ + def __init__(self, fget, doc=None): + self.fget = fget + self.__doc__ = doc or fget.__doc__ + self.__name__ = fget.__name__ + + def __get__(self, obj, cls): + if obj is None: + return self + def oneshot(*args, **kw): + result = self.fget(obj, *args, **kw) + memo = lambda *a, **kw: result + memo.__name__ = self.__name__ + memo.__doc__ = self.__doc__ + obj.__dict__[self.__name__] = memo + return result + oneshot.__name__ = self.__name__ + oneshot.__doc__ = self.__doc__ + return oneshot + +class SetLikeDict(dict): + """a dictionary that has some setlike methods on it""" + def union(self, other): + """produce a 'union' of this dict and another (at the key level). + + values in the second dict take precedence over that of the first""" + x = SetLikeDict(**self) + x.update(other) + return x + +class FastEncodingBuffer(object): + """a very rudimentary buffer that is faster than StringIO, + but doesn't crash on unicode data like cStringIO.""" + + def __init__(self, encoding=None, errors='strict', as_unicode=False): + self.data = collections.deque() + self.encoding = encoding + if as_unicode: + self.delim = compat.u('') + else: + self.delim = '' + self.as_unicode = as_unicode + self.errors = errors + self.write = self.data.append + + def truncate(self): + self.data = collections.deque() + self.write = self.data.append + + def getvalue(self): + if self.encoding: + return self.delim.join(self.data).encode(self.encoding, + self.errors) + else: + return self.delim.join(self.data) + +class LRUCache(dict): + """A dictionary-like object that stores a limited number of items, + discarding lesser used items periodically. + + this is a rewrite of LRUCache from Myghty to use a periodic timestamp-based + paradigm so that synchronization is not really needed. the size management + is inexact. + """ + + class _Item(object): + def __init__(self, key, value): + self.key = key + self.value = value + self.timestamp = compat.time_func() + def __repr__(self): + return repr(self.value) + + def __init__(self, capacity, threshold=.5): + self.capacity = capacity + self.threshold = threshold + + def __getitem__(self, key): + item = dict.__getitem__(self, key) + item.timestamp = compat.time_func() + return item.value + + def values(self): + return [i.value for i in dict.values(self)] + + def setdefault(self, key, value): + if key in self: + return self[key] + else: + self[key] = value + return value + + def __setitem__(self, key, value): + item = dict.get(self, key) + if item is None: + item = self._Item(key, value) + dict.__setitem__(self, key, item) + else: + item.value = value + self._manage_size() + + def _manage_size(self): + while len(self) > self.capacity + self.capacity * self.threshold: + bytime = sorted(dict.values(self), + key=operator.attrgetter('timestamp'), reverse=True) + for item in bytime[self.capacity:]: + try: + del self[item.key] + except KeyError: + # if we couldn't find a key, most likely some other thread + # broke in on us. loop around and try again + break + +# Regexp to match python magic encoding line +_PYTHON_MAGIC_COMMENT_re = re.compile( + r'[ \t\f]* \# .* coding[=:][ \t]*([-\w.]+)', + re.VERBOSE) + +def parse_encoding(fp): + """Deduce the encoding of a Python source file (binary mode) from magic + comment. + + It does this in the same way as the `Python interpreter`__ + + .. __: http://docs.python.org/ref/encodings.html + + The ``fp`` argument should be a seekable file object in binary mode. + """ + pos = fp.tell() + fp.seek(0) + try: + line1 = fp.readline() + has_bom = line1.startswith(codecs.BOM_UTF8) + if has_bom: + line1 = line1[len(codecs.BOM_UTF8):] + + m = _PYTHON_MAGIC_COMMENT_re.match(line1.decode('ascii', 'ignore')) + if not m: + try: + import parser + parser.suite(line1.decode('ascii', 'ignore')) + except (ImportError, SyntaxError): + # Either it's a real syntax error, in which case the source + # is not valid python source, or line2 is a continuation of + # line1, in which case we don't want to scan line2 for a magic + # comment. + pass + else: + line2 = fp.readline() + m = _PYTHON_MAGIC_COMMENT_re.match( + line2.decode('ascii', 'ignore')) + + if has_bom: + if m: + raise SyntaxError("python refuses to compile code with both a UTF8" \ + " byte-order-mark and a magic encoding comment") + return 'utf_8' + elif m: + return m.group(1) + else: + return None + finally: + fp.seek(pos) + +def sorted_dict_repr(d): + """repr() a dictionary with the keys in order. + + Used by the lexer unit test to compare parse trees based on strings. + + """ + keys = list(d.keys()) + keys.sort() + return "{" + ", ".join(["%r: %r" % (k, d[k]) for k in keys]) + "}" + +def restore__ast(_ast): + """Attempt to restore the required classes to the _ast module if it + appears to be missing them + """ + if hasattr(_ast, 'AST'): + return + _ast.PyCF_ONLY_AST = 2 << 9 + m = compile("""\ +def foo(): pass +class Bar(object): pass +if False: pass +baz = 'mako' +1 + 2 - 3 * 4 / 5 +6 // 7 % 8 << 9 >> 10 +11 & 12 ^ 13 | 14 +15 and 16 or 17 +-baz + (not +18) - ~17 +baz and 'foo' or 'bar' +(mako is baz == baz) is not baz != mako +mako > baz < mako >= baz <= mako +mako in baz not in mako""", '', 'exec', _ast.PyCF_ONLY_AST) + _ast.Module = type(m) + + for cls in _ast.Module.__mro__: + if cls.__name__ == 'mod': + _ast.mod = cls + elif cls.__name__ == 'AST': + _ast.AST = cls + + _ast.FunctionDef = type(m.body[0]) + _ast.ClassDef = type(m.body[1]) + _ast.If = type(m.body[2]) + + _ast.Name = type(m.body[3].targets[0]) + _ast.Store = type(m.body[3].targets[0].ctx) + _ast.Str = type(m.body[3].value) + + _ast.Sub = type(m.body[4].value.op) + _ast.Add = type(m.body[4].value.left.op) + _ast.Div = type(m.body[4].value.right.op) + _ast.Mult = type(m.body[4].value.right.left.op) + + _ast.RShift = type(m.body[5].value.op) + _ast.LShift = type(m.body[5].value.left.op) + _ast.Mod = type(m.body[5].value.left.left.op) + _ast.FloorDiv = type(m.body[5].value.left.left.left.op) + + _ast.BitOr = type(m.body[6].value.op) + _ast.BitXor = type(m.body[6].value.left.op) + _ast.BitAnd = type(m.body[6].value.left.left.op) + + _ast.Or = type(m.body[7].value.op) + _ast.And = type(m.body[7].value.values[0].op) + + _ast.Invert = type(m.body[8].value.right.op) + _ast.Not = type(m.body[8].value.left.right.op) + _ast.UAdd = type(m.body[8].value.left.right.operand.op) + _ast.USub = type(m.body[8].value.left.left.op) + + _ast.Or = type(m.body[9].value.op) + _ast.And = type(m.body[9].value.values[0].op) + + _ast.IsNot = type(m.body[10].value.ops[0]) + _ast.NotEq = type(m.body[10].value.ops[1]) + _ast.Is = type(m.body[10].value.left.ops[0]) + _ast.Eq = type(m.body[10].value.left.ops[1]) + + _ast.Gt = type(m.body[11].value.ops[0]) + _ast.Lt = type(m.body[11].value.ops[1]) + _ast.GtE = type(m.body[11].value.ops[2]) + _ast.LtE = type(m.body[11].value.ops[3]) + + _ast.In = type(m.body[12].value.ops[0]) + _ast.NotIn = type(m.body[12].value.ops[1]) + + + +def read_file(path, mode='rb'): + fp = open(path, mode) + try: + data = fp.read() + return data + finally: + fp.close() + +def read_python_file(path): + fp = open(path, "rb") + try: + encoding = parse_encoding(fp) + data = fp.read() + if encoding: + data = data.decode(encoding) + return data + finally: + fp.close() + diff --git a/src/gallium/drivers/swr/rasterizer/scripts/templates/knobs.template b/src/gallium/drivers/swr/rasterizer/scripts/templates/knobs.template new file mode 100644 index 00000000000..922117e7e16 --- /dev/null +++ b/src/gallium/drivers/swr/rasterizer/scripts/templates/knobs.template @@ -0,0 +1,141 @@ +<% + max_len = 0 + for knob in knobs: + if len(knob[0]) > max_len: max_len = len(knob[0]) + max_len += len('KNOB_ ') + if max_len % 4: max_len += 4 - (max_len % 4) + + def space_knob(knob): + knob_len = len('KNOB_' + knob) + return ' '*(max_len - knob_len) +%>/****************************************************************************** +* +* Copyright 2015 +* Intel Corporation +* +* Licensed under the Apache License, Version 2.0 (the "License"); +* you may not use this file except in compliance with the License. +* You may obtain a copy of the License at +* +* http ://www.apache.org/licenses/LICENSE-2.0 +* +* Unless required by applicable law or agreed to in writing, software +* distributed under the License is distributed on an "AS IS" BASIS, +* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. +* See the License for the specific language governing permissions and +* limitations under the License. +* +% if gen_header: +* @file ${filename}.h +% else: +* @file ${filename}.cpp +% endif +* +* @brief Dynamic Knobs for Core. +* +* ======================= AUTO GENERATED: DO NOT EDIT !!! ==================== +* +******************************************************************************/ +%if gen_header: +#pragma once +#include + +template +struct Knob +{ + const T& Value() const { return m_Value; } + const T& Value(const T& newValue) { m_Value = newValue; return Value(); } + +protected: + Knob(const T& defaultValue) : m_Value(defaultValue) {} + +private: + T m_Value; +}; + +#define DEFINE_KNOB(_name, _type, _default) \\ + + struct Knob_##_name : Knob<_type> \\ + + { \\ + + Knob_##_name() : Knob<_type>(_default) { } \\ + + static const char* Name() { return "KNOB_" #_name; } \\ + + } _name; + +#define GET_KNOB(_name) g_GlobalKnobs._name.Value() +#define SET_KNOB(_name, _newValue) g_GlobalKnobs._name.Value(_newValue) + +struct GlobalKnobs +{ + % for knob in knobs: + //----------------------------------------------------------- + // KNOB_${knob[0]} + // + % for line in knob[1]['desc']: + // ${line} + % endfor + DEFINE_KNOB(${knob[0]}, ${knob[1]['type']}, ${knob[1]['default']}); + + % endfor + GlobalKnobs(); + std::string ToString(const char* optPerLinePrefix=""); +}; +extern GlobalKnobs g_GlobalKnobs; + +% for knob in knobs: +#define KNOB_${knob[0]}${space_knob(knob[0])}GET_KNOB(${knob[0]}) +% endfor + + +% else: +% for inc in includes: +#include <${inc}> +% endfor + +//======================================================== +// Static Data Members +//======================================================== +GlobalKnobs g_GlobalKnobs; + +//======================================================== +// Knob Initialization +//======================================================== +GlobalKnobs::GlobalKnobs() +{ + % for knob in knobs: + InitKnob(${knob[0]}); + % endfor + +} + +//======================================================== +// Knob Display (Convert to String) +//======================================================== +std::string GlobalKnobs::ToString(const char* optPerLinePrefix) +{ + std::basic_stringstream str; + str << std::showbase << std::setprecision(1) << std::fixed; + + if (optPerLinePrefix == nullptr) { optPerLinePrefix = ""; } + + % for knob in knobs: + str << optPerLinePrefix << "KNOB_${knob[0]}:${space_knob(knob[0])}"; + % if knob[1]['type'] == 'bool': + str << (KNOB_${knob[0]} ? "+\n" : "-\n"); + % elif knob[1]['type'] != 'float': + str << std::hex << std::setw(11) << std::left << KNOB_${knob[0]}; + str << std::dec << KNOB_${knob[0]} << "\n"; + % else: + str << KNOB_${knob[0]} << "\n"; + % endif + % endfor + str << std::ends; + + return str.str(); +} + + +% endif -- 2.11.0