OSDN Git Service
Chad Versace [Thu, 7 Jan 2016 21:46:20 +0000 (13:46 -0800)]
anv/meta: Fix hardcoded format size in anv_CmdCopy*
When looping through VkBufferImageCopy regions, for each region we
incremented the offset into the VkBuffer assuming the format size was 4.
Fixes CTS tests dEQP-VK.pipeline.image.view_type.cube_array.3d.* on
Skylake.
Chad Versace [Thu, 7 Jan 2016 19:07:44 +0000 (11:07 -0800)]
isl: Add missing break statement in array pitch calculation
Fixes regression in
ed98c374bd3f1952fbab3031afaf5ff4d178ef41.
Chad Versace [Thu, 7 Jan 2016 19:00:29 +0000 (11:00 -0800)]
isl/gen9: Fix array pitch of 3d surfaces
For tiled 3D surfaces, the array pitch must aligned to the tile height.
From the Skylake BSpec >> RENDER_SURFACE_STATE >> Surface QPitch:
Tile Mode != Linear: This field must be set to an integer multiple of
the tile height
Fixes CTS tests 'dEQP-VK.pipeline.image.view_type.3d.format.r8g8b8a8_unorm.*'.
Fixes Crucible tests 'func.miptree.r8g8b8a8-unorm.aspect-color.view-3d.*'.
Chad Versace [Thu, 7 Jan 2016 18:58:29 +0000 (10:58 -0800)]
isl: Refactor func isl_calc_array_pitch_sa_rows
Update the function to calculate the array pitch is *element rows*, and
it rename it accordingly to isl_calc_array_pitch_el_rows.
Jordan Justen [Wed, 6 Jan 2016 23:42:18 +0000 (15:42 -0800)]
isl: Assert that alignments are not 0 for isl_align
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Jordan Justen [Wed, 6 Jan 2016 23:40:01 +0000 (15:40 -0800)]
anv: Assert that alignments are not 0 for align_*
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Jordan Justen [Wed, 6 Jan 2016 23:43:11 +0000 (15:43 -0800)]
isl: Fix image alignment calculation
The previous code was resulting in an alignment of 0.
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Jason Ekstrand [Thu, 7 Jan 2016 03:33:46 +0000 (19:33 -0800)]
anv/state: Pull sampler vk-to-gen maps into genX_state_util.h
Jason Ekstrand [Thu, 7 Jan 2016 01:13:04 +0000 (17:13 -0800)]
nir/spirv: Wrap borrow/carry ops in b2i
NIR specifies them as booleans but SPIR-V wants ints.
Jason Ekstrand [Thu, 7 Jan 2016 01:00:01 +0000 (17:00 -0800)]
nir/spirv/cfg: Only set fall to true at the start of a case
Previously, we were setting it to true at the top of the switch statement.
However, this causes all of the cases to get executed until you hit a
break. Instead, you want to be not executing at the start, start executing
when you hit your case, and end at a break.
Jordan Justen [Wed, 6 Jan 2016 18:27:57 +0000 (10:27 -0800)]
anv: Fix build without VALGRIND
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Jason Ekstrand [Wed, 6 Jan 2016 22:09:47 +0000 (14:09 -0800)]
i965/compiler: Enable more lowering in NIR
We don't need these for GLSL or ARB, but we need them for SPIR-V
Jason Ekstrand [Wed, 6 Jan 2016 23:05:10 +0000 (15:05 -0800)]
nir/algebraic: Add more lowering
This commit adds lowering options for the following opcodes:
- nir_op_fmod
- nir_op_bitfield_insert
- nir_op_uadd_carry
- nir_op_usub_borrow
Jason Ekstrand [Wed, 6 Jan 2016 22:46:31 +0000 (14:46 -0800)]
nir/opcodes: Fix the folding expression for usub_borrow
Jason Ekstrand [Wed, 6 Jan 2016 23:24:38 +0000 (15:24 -0800)]
nir/spirv: Properly implement Modf
Jason Ekstrand [Wed, 6 Jan 2016 23:24:24 +0000 (15:24 -0800)]
nir/builder: Add a helper for storing to a deref
Chad Versace [Tue, 5 Jan 2016 22:35:52 +0000 (14:35 -0800)]
anv/gen9: Teach gen9_image_view_init() about 1D surface qpitch
QPitch is usually expressed as rows of surface elements (where a surface
element is an compression block or a single surface sample. Skylake 1D
is an outlier; there QPitch is expressed as individual surface
elements.
Chad Versace [Tue, 5 Jan 2016 22:30:23 +0000 (14:30 -0800)]
isl: Add isl_surf_get_array_pitch_el()
Will be needed to program SurfaceQPitch for Skylake 1D arrays.
Chad Versace [Tue, 5 Jan 2016 22:30:03 +0000 (14:30 -0800)]
isl/gen9: Support ISL_DIM_LAYOUT_GEN9_1D
Chad Versace [Tue, 5 Jan 2016 22:28:28 +0000 (14:28 -0800)]
isl: Don't align phys_slice0_sa.width twice
It's already aligned to the format's block width. Don't align it again
in isl_calc_row_pitch().
Chad Versace [Tue, 5 Jan 2016 22:26:30 +0000 (14:26 -0800)]
isl: Fix the documented units of isl_surf::row_pitch
It's the pitch between surface elements, not between surface samples.
Chad Versace [Tue, 5 Jan 2016 22:32:44 +0000 (14:32 -0800)]
anv/gen9: Fix oob lookup of surface halign, valign
For 1D surfaces and for surfaces with Yf or Ys tiling, the hardware
ignores SurfaceVerticalAlignment and SurfaceHorizontalAlignment.
Moreover, the anv_halign[] and anv_valign[] lookup tables may not even
contain the surface's actual alignment values. So don't do the lookup
for those surfaces.
Chad Versace [Tue, 5 Jan 2016 22:37:45 +0000 (14:37 -0800)]
anv/meta: Teach meta how to blit from a 1D image
Meta needed a VkShader with a 1D sampler type.
Jason Ekstrand [Wed, 6 Jan 2016 00:18:01 +0000 (16:18 -0800)]
nir/spirv: Fix switch statements with duplicate cases
Jason Ekstrand [Tue, 5 Jan 2016 23:56:59 +0000 (15:56 -0800)]
nir/spirv/cfg: Assert that blocks only ever get added once
This effectively prevents infinite loops in cfg_walk_blocks.
Jason Ekstrand [Tue, 5 Jan 2016 22:59:40 +0000 (14:59 -0800)]
nir/spirv: Simplify phi node handling
Instead of trying to crawl through predecessor chains and build phi nodes,
we just do a poor-man's out-of-ssa on the spot. The into-SSA pass will
deal with putting the actual phi nodes in for us.
Jason Ekstrand [Tue, 5 Jan 2016 21:42:00 +0000 (13:42 -0800)]
anv/pipeline: Lower indirect temporaries and inputs
Jason Ekstrand [Tue, 5 Jan 2016 21:04:07 +0000 (13:04 -0800)]
nir: Add an indirect deref lowering pass
Kristian Høgsberg Kristensen [Tue, 5 Jan 2016 20:00:54 +0000 (12:00 -0800)]
vk: Implement a basic pipeline cache
This is not really a cache yet, but it allows us to share one state
stream for all pipelines, which means we can bump the block size without
wasting a lot of memory.
Kristian Høgsberg Kristensen [Tue, 5 Jan 2016 19:43:25 +0000 (11:43 -0800)]
vk: Destroy device->mutex when destroying the device
Chad Versace [Tue, 5 Jan 2016 19:35:44 +0000 (11:35 -0800)]
isl: Don't force linear for 1d surfaces in gen7_filter_tiling()
gen7_filter_tiling() should filter out only tiling flags that are
incompatible with the surface. It shouldn't make performance decisions,
such as forcing linear for 1D; that's the role of the caller.
Chad Versace [Tue, 5 Jan 2016 19:28:49 +0000 (11:28 -0800)]
isl: Document gen7_filter_tiling()
Chad Versace [Tue, 5 Jan 2016 18:39:21 +0000 (10:39 -0800)]
isl: Prefer linear tiling for 1D surfaces
Chad Versace [Tue, 5 Jan 2016 17:53:11 +0000 (09:53 -0800)]
isl: Remove isl_format_layout::bpb
struct isl_format_layout contained two near-redundant members: bpb (bits
per block) and bs (block size). There do exist some hardware formats for
which bpb != 8 * bs, but Vulkan does not use them. Therefore we don't
need bpb.
Chad Versace [Tue, 5 Jan 2016 17:59:07 +0000 (09:59 -0800)]
anv: Use isl_format_layout::bs instead of ::bpb
For all formats used by Vulkan, 8 * bs == bpb.
(bs=block_size_in_bytes, bpb=bits_per_block)
Chad Versace [Tue, 5 Jan 2016 01:09:11 +0000 (17:09 -0800)]
isl: Align isl_surf::phys_level0_sa to the format's compression block
Chad Versace [Tue, 5 Jan 2016 01:00:49 +0000 (17:00 -0800)]
isl: Fix mis-documented units of isl_surf::phys_level_sa
It's in physical surface samples. Hence the _sa suffix.
Jason Ekstrand [Tue, 5 Jan 2016 06:08:24 +0000 (22:08 -0800)]
nir/spirv: Add support for the ControlBarrier instruction
Jason Ekstrand [Tue, 5 Jan 2016 05:36:11 +0000 (21:36 -0800)]
anv/UpdateDescriptorSets: Use the correct index for the buffer view
Jason Ekstrand [Tue, 5 Jan 2016 04:59:16 +0000 (20:59 -0800)]
nir/spirv: Implement extended add, sub, and mul
Jason Ekstrand [Tue, 5 Jan 2016 01:32:33 +0000 (17:32 -0800)]
nir/spirv: Add support for bitfield operations
Jason Ekstrand [Tue, 5 Jan 2016 01:16:44 +0000 (17:16 -0800)]
nir/spirv: Add support for msb/lsb opcodes
Jason Ekstrand [Tue, 5 Jan 2016 01:04:26 +0000 (17:04 -0800)]
nir/spirv: Add a documenting assert for OpConstantSampler
Jason Ekstrand [Tue, 5 Jan 2016 00:57:01 +0000 (16:57 -0800)]
nir/spirv: Add initial support for ConstantNull
Chad Versace [Tue, 5 Jan 2016 00:31:10 +0000 (16:31 -0800)]
isl: Align single-level 2D surfaces to compression block
This fixes an assertion failure at isl.c:1003.
Reported-by: Nanley Chery <nanley.g.chery@intel.com>
Jason Ekstrand [Fri, 1 Jan 2016 16:45:35 +0000 (08:45 -0800)]
anv/formats: Hand out different formats based on tiled vs. linear
Jason Ekstrand [Fri, 1 Jan 2016 22:09:17 +0000 (14:09 -0800)]
anv/image_view: Separate vulkan and isl formats
Previously, anv_image_view had a anv_format pointer that we used for
everything. This commit replaces that pointer with a VkFormat enum copied
from the API and an isl_format. In order to implement RGB formats, we have
to use a different isl_format for the actual surface state than the obvious
one from the VkFormat. Separating the two helps us keep things streight.
Jason Ekstrand [Fri, 1 Jan 2016 22:12:22 +0000 (14:12 -0800)]
anv_get_isl_format: Support depth+stencil aspect value
You just get the depth format in this case.
Jason Ekstrand [Fri, 1 Jan 2016 21:47:18 +0000 (13:47 -0800)]
anv/image: Do more work in anv_image_view_init
There was a bunch of common code in gen7/8_image_view_init that we really
should be sharing.
Jason Ekstrand [Fri, 1 Jan 2016 16:42:35 +0000 (08:42 -0800)]
anv/formats: Rework GetPhysicalDeviceFormatProperties
It now calls get_isl_format to get both linear and tiled views of the
format and determines linear/tiled properties from that. Buffer properties
are determined from the linear format.
Jason Ekstrand [Fri, 1 Jan 2016 02:31:31 +0000 (18:31 -0800)]
anv/formats: Add a tiling parameter to get_isl_format
Currently, this parameter does nothing.
Jason Ekstrand [Thu, 31 Dec 2015 22:57:51 +0000 (14:57 -0800)]
isl/format: Add some helpers for working with RGB formats
Jason Ekstrand [Thu, 31 Dec 2015 22:30:51 +0000 (14:30 -0800)]
isl: Add a file for format helpers
Jason Ekstrand [Thu, 31 Dec 2015 21:51:15 +0000 (13:51 -0800)]
genX/state: Pull some generic helpers into a shared header
Jason Ekstrand [Thu, 31 Dec 2015 21:22:43 +0000 (13:22 -0800)]
meta/blit: Rework how format and aspect choices are made
This commit does two things. First, it introduces choose_* functions for
chosing formats and aspects. Second, it changes the copy (not blit) code
to use appropreately sized UINT formats for everything except depth. There
are two main reasons for this: First, it means that compressed and other
non-renderable texture upload should "just work" because it won't be
tripping over non-renderable formats. Second, it allows us to easly copy
an RGB buffer to and from an RGBX image because the formats will get
switched over to their UINT variants and the shader will deal with the
extra channel for us.
Jason Ekstrand [Thu, 31 Dec 2015 20:39:34 +0000 (12:39 -0800)]
anv/image: Add a vk_format field
We've been trying to move away from anv_format for a while and this should
help with the transition. There are cases (mostly in meta) where we need
the original format for the image and not the isl_format. These will be
moved over to the new vk_format and everythign else will use the isl_format
from the particular anv_surface.
Chad Versace [Mon, 4 Jan 2016 23:30:27 +0000 (15:30 -0800)]
isl: Document mnemonic in Yf and Ys tiling
The 'f' means "four K". The 's' means "sixty-four K".
Kristian Høgsberg Kristensen [Mon, 4 Jan 2016 18:53:47 +0000 (10:53 -0800)]
isl: Use isl_align_npot for row_pitch
Many formats are not power-of-two bytes per pixels and we need the
non-power-of-two align macro here.
This reverts the revert from
4f9a211b, but keeps the change from
a827b553
that fixed the yuv if-else mix-up.
Kristian Høgsberg Kristensen [Mon, 4 Jan 2016 18:42:50 +0000 (10:42 -0800)]
vk: Don't leak pipeline if initialization fails
Kristian Høgsberg Kristensen [Mon, 4 Jan 2016 06:58:00 +0000 (22:58 -0800)]
vk: Allocate subpass attachment in one big block
This avoids making a lot of small allocations and handles allocation
failure correctly.
Fixes dEQP-VK.api.object_management.alloc_callback_fail.* failures.
Kristian Høgsberg Kristensen [Mon, 4 Jan 2016 06:43:47 +0000 (22:43 -0800)]
vk: Handle allocation failures in meta init paths
Fixes dEQP-VK.api.object_management.alloc_callback_fail.* failures.
Kristian Høgsberg Kristensen [Mon, 4 Jan 2016 06:42:01 +0000 (22:42 -0800)]
vk: Handle allocation failure in anv_pipeline_init()
Fixes dEQP-VK.api.object_management.alloc_callback_fail.* failures.
Kristian Høgsberg Kristensen [Mon, 4 Jan 2016 05:31:22 +0000 (21:31 -0800)]
vk: Call vk_error when we generate a VK_ERROR
Kristian Høgsberg Kristensen [Mon, 4 Jan 2016 05:15:29 +0000 (21:15 -0800)]
vk: Only finish wayland wsi if we created it
Failure during instance creation will leave instance->wayland_wsi
undefined. When we then try to clean that up we crash. Set
instance->wayland_wsi to NULL on failure and only clean it up if it's
non-NULL.
Fixes part of dEQP-VK.api.object_management.alloc_callback_fail.*
Chad Versace [Fri, 1 Jan 2016 17:52:57 +0000 (09:52 -0800)]
isl: Fix row pitch for linear buffers
isl always aligned the row pitch to the surface's image alignment. This
was sometimes wrong when the surface backed a VkBuffer. For a VkBuffer,
the surface's row pitch is set by VkBufferImageCopy::bufferRowLength,
whose required alignment is only that of the VkFormat.
In particular, VkBuffer rows are packed in many dEQP and Crucible tests.
And packed rows are rarely aligned to the surface's image alignment.
Fixes: dEQP-VK.pipeline.image.view_type.2d.format.r8g8b8a8_unorm.size.13x13
Chad Versace [Fri, 1 Jan 2016 18:37:25 +0000 (10:37 -0800)]
isl: Fix swapped if-else in isl_calc_row_pitch
The YUV case was applied to non-YUV formats. Oops.
Jason Ekstrand [Sat, 2 Jan 2016 19:51:40 +0000 (11:51 -0800)]
nir/spirv: Fix group decorations
They were completely bogus before. For one thing, OpDecorationGroup
created a value of type undef rather than decoration_group. Also
OpGroupMemberDecorate didn't properly apply the decoration to the different
members of the different groups. It *should* be correct now but there's no
good way to test it yet.
Jason Ekstrand [Sat, 2 Jan 2016 15:52:22 +0000 (07:52 -0800)]
anv/device: Only allocate whole pages in AllocateMemory
The kernel is going to give us whole pages anyway, so allocating part of a
page doesn't help. And this ensures that we can always work with whole
pages.
Jason Ekstrand [Fri, 1 Jan 2016 17:26:06 +0000 (09:26 -0800)]
anv/device: Handle non-4k-aligned calls to MapMemory
As per the spec:
minMemoryMapAlignment is the minimum required alignment, in bytes, of
host-visible memory allocations within the host address space. When
mapping a memory allocation with vkMapMemory, subtracting offset bytes
from the returned pointer will always produce a multiple of the value of
this limit.
Jason Ekstrand [Thu, 31 Dec 2015 20:07:23 +0000 (12:07 -0800)]
anv/format: Get rid of num_channels
Jason Ekstrand [Thu, 31 Dec 2015 20:07:10 +0000 (12:07 -0800)]
anv/cmd_buffer: Fix a pointer-cast typo
Chad Versace [Thu, 31 Dec 2015 20:02:04 +0000 (12:02 -0800)]
isl: Document some isl_surf::phys_level0_sa invariants
isl_dim_layout restricts the range of isl_surf::phys_level0_sa.
Jason Ekstrand [Thu, 31 Dec 2015 20:00:58 +0000 (12:00 -0800)]
anv/pipeline: Better vertex input channel setup
First off, it now uses isl formats instead of anv_format. Also, it
properly handles integer vs. floating-point default channels and can
properly handle alpha-only channels. (Not sure if those are allowed).
Jason Ekstrand [Thu, 31 Dec 2015 19:58:24 +0000 (11:58 -0800)]
anv/pipeline: Move vk_to_gen tables into a shared header
Chad Versace [Thu, 31 Dec 2015 19:56:13 +0000 (11:56 -0800)]
isl: Better document surface units
Logical pixels, physical surface samples, and physical surface elements.
Requested-by: Jason Ekstrand <jason.ekstrand@intel.com>
Chad Versace [Thu, 31 Dec 2015 19:55:48 +0000 (11:55 -0800)]
isl: Document the 3D block extent of isl_format
Jason Ekstrand [Thu, 31 Dec 2015 05:57:04 +0000 (21:57 -0800)]
nir/spirv: Add a missing break statement in handle_image
Jason Ekstrand [Thu, 31 Dec 2015 05:01:55 +0000 (21:01 -0800)]
Revert "isl: Fix assertion failure for npot pixel formats"
This reverts commit
96d1baa88d37c51c94579f650cfd9465d28634f4.
Jason Ekstrand [Thu, 31 Dec 2015 04:54:27 +0000 (20:54 -0800)]
nir/spirv: Handle push constants after decorations
Jason Ekstrand [Thu, 31 Dec 2015 03:32:41 +0000 (19:32 -0800)]
anv/device: Place memory types at heapIndex == 0
Previously, they were at heapIndex == 1 even though we only advertised one
heap.
Jason Ekstrand [Thu, 31 Dec 2015 02:51:57 +0000 (18:51 -0800)]
nir/spirv: Fix constant num_elements and allocation
Thanks to the addition of nir_clone, we now have a num_elements field in
nir_constant which we weren't setting. Also, constants have to be parented
to the variable they initialize, so we have to make a copy.
Jason Ekstrand [Thu, 31 Dec 2015 02:51:06 +0000 (18:51 -0800)]
nir/lower_outputs_to_temporaries: Reparent constant initializers
Jason Ekstrand [Thu, 31 Dec 2015 02:44:19 +0000 (18:44 -0800)]
nir/clone: Expose nir_constant_clone
Jason Ekstrand [Thu, 31 Dec 2015 01:43:00 +0000 (17:43 -0800)]
nir/gather_info: Add support for end_primitive_with_counter
Jason Ekstrand [Thu, 31 Dec 2015 01:32:03 +0000 (17:32 -0800)]
nir/spirv: Handle OpLine
Jason Ekstrand [Thu, 31 Dec 2015 01:31:51 +0000 (17:31 -0800)]
nir/spirv: Let OpEntryPoint act as an OpName
Jason Ekstrand [Thu, 31 Dec 2015 01:31:19 +0000 (17:31 -0800)]
nir/lower_outputs_to_temporaries: Take a nir_function entrypoint
Jason Ekstrand [Thu, 31 Dec 2015 01:17:12 +0000 (17:17 -0800)]
nir/spirv: Add support for multiple entrypoints per shader
This is done by passing the entrypoint name into spirv_to_nir. It will
then process the shader as if that were the only entrypoint we care about.
Instead of returning a nir_shader, it now returns a nir_function.
Jason Ekstrand [Thu, 31 Dec 2015 01:00:16 +0000 (17:00 -0800)]
nir/spirv: Get the shader stage from the SPIR-V
Previously, we depended on it being passed in.
Jason Ekstrand [Thu, 31 Dec 2015 00:48:59 +0000 (16:48 -0800)]
nir/spirv: Use shader stage for determining variable locations
Jason Ekstrand [Thu, 31 Dec 2015 00:41:54 +0000 (16:41 -0800)]
nir/spirv: Get rid of default GS info
shaderc has been fixed for a while now.
Jason Ekstrand [Thu, 31 Dec 2015 00:38:16 +0000 (16:38 -0800)]
nir/spirv: Handle execution modes as decorations
They're basically the same thing.
Jason Ekstrand [Thu, 31 Dec 2015 00:35:34 +0000 (16:35 -0800)]
nir/spirv: Separate handling of preamble from type/var/const instructions
Chad Versace [Thu, 31 Dec 2015 00:26:10 +0000 (16:26 -0800)]
isl: Fix assertion failure for npot pixel formats
When aligning to isl_format_layout::bs (which is the number of bytes in
the pixel), use isl_align_npot() instead of isl_align(), because
isl_align() works only for power-of-2 alignment.
Fixes assertion in
dEQP-VK.pipeline.image.view_type.1d.format.r16g16b16_sfloat.size.512x1.
Jason Ekstrand [Wed, 30 Dec 2015 22:41:13 +0000 (14:41 -0800)]
nir/spirv/GLSL450: Add support for SAbs
Kenneth Graunke [Wed, 30 Dec 2015 00:21:09 +0000 (16:21 -0800)]
nir/spirv: Implement IsInf and IsNan built-ins.
Jason Ekstrand [Wed, 30 Dec 2015 22:02:20 +0000 (14:02 -0800)]
isl: Tile-align height in image size calculation
This fixes a bunch of gpu hangs on the dEQP-VK.glsl.ShaderExecutor.common
group of CTS tests.
Kenneth Graunke [Wed, 30 Dec 2015 21:22:32 +0000 (13:22 -0800)]
Revert "nir/spirv: Update to the 1.0 GLSL.std.450 header"
This reverts commit
b33f5d388979f23c583c78f10a18a941cb04ce04,
and also removes the (empty) case statements for the new built-ins.
It doesn't look like glslang has updated yet, so updating the header
just breaks everything, as we no longer agree on opcode numbers.
Jason Ekstrand [Wed, 30 Dec 2015 19:40:38 +0000 (11:40 -0800)]
anv/allocator: Rework state streams again
If we're going to hav valgrind verify state streams then we need to ensure
that once we choose a pointer into a block we always use that pointer until
the block is freed. I was trying to do this with the "current_map" thing.
However, that breaks down because you have to use the map from the block
pool to get to the stream_block to get at current_map. Instead, this
commit changes things to track the stream_block by pointer instead of by
offset into the block pool.
Jason Ekstrand [Wed, 30 Dec 2015 18:37:50 +0000 (10:37 -0800)]
gen7/8/cmd_buffer: Allocate the correct ammount for COLOR_CALC_STATE
We were allocating 6 bytes when we should have been allocating 6 dwords.