OSDN Git Service
Vinson Lee [Fri, 27 Apr 2012 04:27:37 +0000 (21:27 -0700)]
glsl: Remove unused member predicate from ir_dead_functions_visitor.
Fix uninitialized pointer field defect reported by Coverity.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Kenneth Graunke [Tue, 24 Apr 2012 21:09:13 +0000 (14:09 -0700)]
i965/fs: Fix FB writes that tried to use the non-existent m16 register.
A little analysis shows that the worst-case value for "nr" is 17:
- base_mrf = 2 ... 2
- header present (say gen == 5) ... 4
- aa_dest_stencil_reg (stencil test) ... 5
- SIMD16 mode: += 4 * reg_width ... 13
- source_depth_to_render_target ... 15
- dest_depth_reg ... 17
This resulted in us setting base_mrf to 2 and mlen to 15. In other
words, we'd try to use m2..m16. But m16 doesn't exist pre-Gen6. Also,
the instruction scheduler data structures use arrays of size 16, so this
would cause us to access them out of bounds.
While the debugger system routine may need m0 and m1, we don't use it
today, so the simplest solution is just to move base_mrf back to 1.
That way, our worst case message fits in m1..m15, which is legal.
An alternative would be to fail on SIMD16 in this case, but that seems
a bit unfortunate if there's no real need to reserve m0 and m1.
Fixes new piglit test shaders/depth-test-and-write on Ironlake.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=48218
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
Vinson Lee [Thu, 26 Apr 2012 05:49:12 +0000 (22:49 -0700)]
glsl: Remove unused member mem_ctx from ir_dead_functions_visitor.
Fix uninitialized pointer field defect reported by Coverity.
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Ben Skeggs [Thu, 26 Apr 2012 23:26:13 +0000 (09:26 +1000)]
nv30: properly init window information
Should fix >2k rendering issues reported on nv4x.
Signed-off-by: Ben Skeggs <bskeggs@redhat.com>
Tom Stellard [Wed, 25 Apr 2012 22:09:05 +0000 (18:09 -0400)]
radeonsi/llvm: Silence a warning
Tom Stellard [Wed, 25 Apr 2012 16:20:51 +0000 (12:20 -0400)]
radeon/llvm: Remove unused header files
Tom Stellard [Wed, 25 Apr 2012 16:16:27 +0000 (12:16 -0400)]
radeon/llvm: Remove AMDILMachineFunctionInfo.cpp
Tom Stellard [Wed, 25 Apr 2012 15:55:01 +0000 (11:55 -0400)]
radeon/llvm: Remove AMDILModuleInfo.cpp
José Fonseca [Wed, 25 Apr 2012 15:30:06 +0000 (16:30 +0100)]
gallivm: Use lp_build_alloca instead of LLVMBuildAlloca on the loop limiter.
To ensure that the alloca is at the top of the function body, otherwise
LLVM will not eliminate them, causing stack misalignment on 32bits.
Reviewed-by: James Benton <jbenton@vmware.com>
Tom Stellard [Wed, 25 Apr 2012 02:04:43 +0000 (22:04 -0400)]
radeon/llvm: Remove AMDILELFWriterInfo.cpp
Tom Stellard [Wed, 25 Apr 2012 01:57:14 +0000 (21:57 -0400)]
radeon/llvm: Remove AMDILLiteralManager.cpp
Tom Stellard [Wed, 25 Apr 2012 01:52:22 +0000 (21:52 -0400)]
radeon/llvm: Remove AMDILInliner.cpp
Tom Stellard [Wed, 25 Apr 2012 01:48:14 +0000 (21:48 -0400)]
radeon/llvm: Remove AMDILBarrierDetect.cpp
Tom Stellard [Wed, 25 Apr 2012 01:44:11 +0000 (21:44 -0400)]
radeon/llvm: Remove AMDILPrintfConvert.cpp
Tom Stellard [Wed, 25 Apr 2012 01:39:11 +0000 (21:39 -0400)]
radeon/llvm: Remove GlobalManager and KernelManager
Tom Stellard [Wed, 25 Apr 2012 01:20:26 +0000 (21:20 -0400)]
radeon/llvm: Remove AsmPrinter files
Tom Stellard [Wed, 25 Apr 2012 01:15:11 +0000 (21:15 -0400)]
radeon/llvm: Remove IOExpansion files
Tom Stellard [Tue, 24 Apr 2012 23:54:22 +0000 (19:54 -0400)]
radeon/llvm: Remove AMDILPointerManager.cpp
Tom Stellard [Wed, 25 Apr 2012 13:03:21 +0000 (09:03 -0400)]
radeonsi/llvm: Fix initialization of SIMachineFunctionInfo
SIMachineFunctionInfo needs to be initialized before any of the
AMDIL passes.
José Fonseca [Wed, 25 Apr 2012 10:41:13 +0000 (11:41 +0100)]
mesa/st: Fix derreference after free.
Marek Olšák [Tue, 24 Apr 2012 21:10:46 +0000 (23:10 +0200)]
gallium/docs: document the new vertex fetch CAPs
Christoph Bumiller [Tue, 24 Apr 2012 18:04:48 +0000 (20:04 +0200)]
nv50: fix typo in nv50_fragprog_assign_slots
Eric Anholt [Wed, 18 Apr 2012 18:00:44 +0000 (11:00 -0700)]
glsl: Add implementation of inverse() for mat2/3/4.
This is taken from the ogl-math project, with Inverse renamed to adj
(since it's not actually the inverse), transposed, and our types
plugged in. There are potential CSE opportunities in this code
(particularly for hardware with RCP but not DIV), but we should be
doing CSE anyway, so don't hand-optimize.
Fixes piglit inverse tests.
Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Eric Anholt [Wed, 18 Apr 2012 17:51:23 +0000 (10:51 -0700)]
glsl: Add support for generating builtin code from GLSL instead of IR.
This takes advantage of the builtin compiler to generate IR into a
string, the same way we read GLSL for function prototypes for our
profiles.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Eric Anholt [Wed, 18 Apr 2012 21:43:12 +0000 (14:43 -0700)]
glsl: Make ir_reader parse the "temporary" variable qualifier.
This lets ir_reader eat the output of builtin_compiler on actual
function definitions.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Eric Anholt [Wed, 18 Apr 2012 21:40:10 +0000 (14:40 -0700)]
glsl: Add doxygen explaining what main.cpp is for.
I keep getting lost in the Makefile trying to figure out what to edit
to work on builtin_compiler or glsl_compiler.
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Eric Anholt [Tue, 17 Apr 2012 18:56:26 +0000 (11:56 -0700)]
i965/vs: Fix texelFetchOffset()
It appears that when using 'ld' with the offset bits, address bounds
checking happens before the offset is applied, so parts of the drawing
in piglit texelFetchOffset() with a negative texcoord go black.
Eric Anholt [Tue, 17 Apr 2012 01:43:48 +0000 (18:43 -0700)]
i965/fs: Fix texelFetchOffset()
It appears that when using 'ld' with the offset bits, address bounds
checking happens before the offset is applied, so parts of the drawing
in piglit texelFetchOffset() with a negative texcoord go black.
Christoph Bumiller [Tue, 24 Apr 2012 11:34:36 +0000 (13:34 +0200)]
nv50,nvc0: handle new vertex fetch caps
Christoph Bumiller [Tue, 24 Apr 2012 11:35:44 +0000 (13:35 +0200)]
nv50,nvc0: don't initialize the draw module, we don't use it
But some day we might (e.g. for blending 16 bpp formats on nv50).
Christoph Bumiller [Mon, 23 Apr 2012 18:08:54 +0000 (20:08 +0200)]
nv50: enable shader watchdog timer
Jeremy Huddleston [Mon, 23 Apr 2012 23:43:22 +0000 (16:43 -0700)]
darwin: Eliminate a pthread mutex leak
Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>
Tested-by: Charles Davis <cdavis@mines.edu>
Marek Olšák [Mon, 23 Apr 2012 23:23:33 +0000 (01:23 +0200)]
radeonsi: stop using u_vbuf and adapt to gallium interface changes
Marek Olšák [Thu, 19 Apr 2012 17:54:23 +0000 (19:54 +0200)]
u_vbuf: don't call u_vbuf_upload_buffers if there's nothing to upload
Marek Olšák [Thu, 19 Apr 2012 03:28:27 +0000 (05:28 +0200)]
u_vbuf: don't upload user buffers if the driver supports them
Marek Olšák [Thu, 19 Apr 2012 01:47:03 +0000 (03:47 +0200)]
u_vbuf: optimize u_vbuf_translate_find_free_vb_slots
Marek Olšák [Wed, 18 Apr 2012 22:56:44 +0000 (00:56 +0200)]
u_vbuf: turn bool arrays into bitmasks and loops into bitwise operations
Optimized functions:
- u_vbuf_need_minmax_index
- u_vbuf_mapping_vertex_buffer_blocks
Marek Olšák [Wed, 18 Apr 2012 23:41:46 +0000 (01:41 +0200)]
u_vbuf: take advantage of all new vertex fetch caps
Marek Olšák [Wed, 18 Apr 2012 23:24:32 +0000 (01:24 +0200)]
u_vbuf: simplify u_vbuf_set_vertex_buffers
Marek Olšák [Wed, 18 Apr 2012 23:12:55 +0000 (01:12 +0200)]
u_vbuf: rename native_attribs -> driver_attribs
Marek Olšák [Sun, 1 Apr 2012 21:54:44 +0000 (23:54 +0200)]
r600g: move updating vertex buffer state into set_vertex_buffers
Marek Olšák [Wed, 11 Apr 2012 10:55:23 +0000 (12:55 +0200)]
st/mesa: always expose ARB_ES2_compatibility
u_vbuf translates GL_FIXED too if needed.
Marek Olšák [Wed, 11 Apr 2012 10:54:10 +0000 (12:54 +0200)]
st/mesa: always expose ARB_half_float_vertex
u_vbuf kicks in and translates it to float if it's unsupported.
Marek Olšák [Tue, 10 Apr 2012 04:00:17 +0000 (06:00 +0200)]
gallium: make user vertex buffers optional
This couldn't be split because it would break bisecting.
Summary:
* r300g,r600g: stop using u_vbuf
* r300g,r600g: also report that the FIXED vertex type is unsupported
* u_vbuf: refactor for use in the state tracker
* cso: wire up u_vbuf with cso_context
* st/mesa: conditionally install u_vbuf
Marek Olšák [Wed, 11 Apr 2012 13:16:09 +0000 (15:16 +0200)]
i915g: report that all vertex formats are supported
So that u_vbuf isn't enabled.
Marek Olšák [Wed, 11 Apr 2012 14:00:09 +0000 (16:00 +0200)]
r600g: don't share u_upload_mgr with u_vbuf, create its own
Marek Olšák [Wed, 11 Apr 2012 14:00:09 +0000 (16:00 +0200)]
r300g: don't share u_upload_mgr with u_vbuf, create its own
Marek Olšák [Wed, 11 Apr 2012 13:56:59 +0000 (15:56 +0200)]
u_vbuf: pull u_vbuf_draw_max_vertex_count into r300g
Marek Olšák [Tue, 10 Apr 2012 03:14:26 +0000 (05:14 +0200)]
u_vbuf: make use of the new CAPs to determine what to do
This adds the ability to initialize u_vbuf_caps before creating u_vbuf itself.
It will be useful for determining if u_vbuf should be used or not.
Also adapt r300g and r600g.
Marek Olšák [Wed, 11 Apr 2012 13:40:00 +0000 (15:40 +0200)]
gallium drivers: report that user vertex buffers are supported
Marek Olšák [Tue, 10 Apr 2012 03:12:22 +0000 (05:12 +0200)]
gallium: add CAPs for vertex fetcher
Supported vertex formats will be queried using
is_format_supported(.., PIPE_BIND_VERTEX_BUFFER, ..).
Marek Olšák [Tue, 10 Apr 2012 19:29:06 +0000 (21:29 +0200)]
u_vbuf: remove u_vbuf_resource
Marek Olšák [Tue, 10 Apr 2012 02:37:52 +0000 (04:37 +0200)]
u_vbuf: use user_ptr from pipe_resource
Marek Olšák [Tue, 10 Apr 2012 02:33:23 +0000 (04:33 +0200)]
gallium: add user_ptr in pipe_resource
I need to access the pointer in st/mesa when I only have pipe_resource.
Marek Olšák [Sat, 31 Mar 2012 12:52:35 +0000 (14:52 +0200)]
u_vbuf: override draw_vbo
Marek Olšák [Sat, 31 Mar 2012 04:01:39 +0000 (06:01 +0200)]
u_vbuf: override create/bind/destroy_vertex_elements_state
Marek Olšák [Sat, 31 Mar 2012 03:32:25 +0000 (05:32 +0200)]
u_vbuf: override set_vertex_buffers
Marek Olšák [Sat, 31 Mar 2012 02:56:48 +0000 (04:56 +0200)]
u_vbuf: override set_index_buffer
This makes u_vbuf_mgr call the driver instead of the other way around.
Marek Olšák [Wed, 11 Apr 2012 14:08:22 +0000 (16:08 +0200)]
st/mesa: use cso_set_index_buffer and cso_draw_vbo
Reviewed-by: Brian Paul <brianp@vmware.com>
Marek Olšák [Wed, 11 Apr 2012 13:51:52 +0000 (15:51 +0200)]
gallium/util: use cso_draw_arrays in util_draw_vertex_buffer
Reviewed-by: Brian Paul <brianp@vmware.com>
Marek Olšák [Wed, 11 Apr 2012 13:51:13 +0000 (15:51 +0200)]
cso: add set_index_buffer and draw_vbo passthrough functions
v2: use util_draw_init_info
Reviewed-by: Brian Paul <brianp@vmware.com>
Jeremy Huddleston [Mon, 23 Apr 2012 23:13:51 +0000 (16:13 -0700)]
darwin: Don't build gallium
fe3f98ed70be01a49dd2e93870907d092fe4a283 intentionally broke it, so
don't build it.
Signed-off-by: Jeremy Huddleston <jeremyhu@apple.com>
Jonas Maebe [Mon, 23 Apr 2012 23:02:16 +0000 (16:02 -0700)]
apple: Fix a use after free
Reviewed-by: Jeremy Huddleston <jeremyhu@apple.com>
Marek Olšák [Sun, 22 Apr 2012 17:21:45 +0000 (19:21 +0200)]
vbo: call UpdateState directly when notifying a driver about _NEW_ARRAY
Core Mesa doesn't need to know about this.
This also removes the hack in recalculate_input_bindings.
Reviewed-by: Brian Paul <brianp@vmware.com>
Marek Olšák [Sun, 22 Apr 2012 14:36:38 +0000 (16:36 +0200)]
vbo: do FLUSH_CURRENT before validating and updating state
This fixes an assertion failure since:
commit
81afdd20f3f574ce29559d8ad77df5c77652009e
vbo: don't check twice whether it's valid to render
FLUSH_CURRENT may set _NEW_CURRENT_ATTRIB.
Reviewed-by: Brian Paul <brianp@vmware.com>
Marek Olšák [Thu, 19 Apr 2012 02:33:43 +0000 (04:33 +0200)]
r300g/automake: add $ARCH_FLAGS and $OPT_FLAGS
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Marek Olšák [Thu, 19 Apr 2012 02:33:43 +0000 (04:33 +0200)]
r600g/automake: add $ARCH_FLAGS and $OPT_FLAGS
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Brian Paul [Mon, 23 Apr 2012 19:29:06 +0000 (13:29 -0600)]
r600g: init vars to silence warnings
Han Shen(沈涵) [Mon, 23 Apr 2012 14:28:12 +0000 (08:28 -0600)]
bin/mklib: remove '-m32' for arm linux
-m32 is not a valid option for ARM.
NOTE: This is a candidate for the 8.0 branch.
Signed-off-by: Brian Paul <brianp@vmware.com>
Brian Paul [Fri, 20 Apr 2012 22:45:12 +0000 (16:45 -0600)]
st/mesa: no-op glCopyPixels if source region is out of bounds
If the source region for a glCopyPixels is completely outside the
source buffer bounds, no-op the copy. Fixes a failed assertion.
NOTE: This is a candidate for the 8.0 branch.
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Tom Stellard [Fri, 20 Apr 2012 13:12:47 +0000 (09:12 -0400)]
radeon/llvm: Don't print an error message when there is no error
A blank line with an empty error message was being printed even when the
target lookup succeeded.
Tom Stellard [Thu, 19 Apr 2012 14:14:41 +0000 (10:14 -0400)]
radeon/llvm: Lower VCREATE_v4f32 for R600 and SI
Tom Stellard [Thu, 19 Apr 2012 14:09:52 +0000 (10:09 -0400)]
r600g/llvm: Let ISel handle lowering to {INSERT,EXTRACT}_SUBREG
Tom Stellard [Thu, 19 Apr 2012 15:47:08 +0000 (11:47 -0400)]
r600g/llvm: Only emit an instruction's explicit operands
Tom Stellard [Fri, 20 Apr 2012 13:07:37 +0000 (09:07 -0400)]
r600g/llvm: Handle copies between vector registers
Tom Stellard [Thu, 19 Apr 2012 19:19:21 +0000 (15:19 -0400)]
r600g/llvm: Remove debugging hack from R600InstrInfo::copyPhysReg()
Tom Stellard [Thu, 19 Apr 2012 14:12:05 +0000 (10:12 -0400)]
r600g/llvm: Tell the code emitter to ignore KILL and BUNDLE
Tom Stellard [Tue, 28 Feb 2012 22:32:28 +0000 (17:32 -0500)]
r600/llvm: Add LOAD_VTX instruction
Tom Stellard [Tue, 17 Apr 2012 13:26:28 +0000 (09:26 -0400)]
r600g: Add hooks for the LLVM shader compiler
The LLVM backend can now be enabled for r600g by using the
--enable-r600-llvm-compiler configure flag. If you configure with this
flag, you can still use the default compiler by setting the envrionment
variable R600_USE_LLVM=0
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Tom Stellard [Tue, 17 Apr 2012 12:37:38 +0000 (08:37 -0400)]
r600g: Add TGSI->LLVM implementation v2
v2: Add case for ARUBA in r600_llvm_gpu_string()
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Tom Stellard [Mon, 16 Apr 2012 21:48:44 +0000 (17:48 -0400)]
radeon: Move radeon_llvm_emit.cpp declarations into their own header
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Tom Stellard [Mon, 16 Apr 2012 21:47:04 +0000 (17:47 -0400)]
radeon: Remove HAVE_LLVM ifdefs
Only LLVM 3.1 is supported, so these are not necessary.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Vadim Girlin [Fri, 16 Dec 2011 03:32:07 +0000 (07:32 +0400)]
r600g: fix gpr number calculation
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Tom Stellard <thomas.stellard@amd.com>
Tom Stellard [Tue, 17 Apr 2012 13:24:29 +0000 (09:24 -0400)]
configure.ac: Move HAVE_LLVM definition into configure.ac
Otherwise HAVE_LLVM won't be included in the $(DEFINES) variable for
Automake generated Makefiles.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Dave Airlie [Mon, 23 Apr 2012 09:20:10 +0000 (10:20 +0100)]
r600g: split add_reg into add_reg and add_reg_bo variants
This shaves 2k off the final dri.so, and removes lots of pointless
NULL, 0 passing.
most like pointless - but it looked nicer to me.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Jonas Maebe [Mon, 23 Apr 2012 03:39:32 +0000 (20:39 -0700)]
glapi: Correct size of allocated _glapi_table struct
The __glapi_gentable_set_remaining_noop() routine treats the _glapi_struct
as an array of _glapi_get_dispatch_table_size() pointers, so we have to
allocate _glapi_get_dispatch_table_size()*sizeof(void*) bytes rather
than sizeof(struct _glapi_struct) bytes.
Reviewed-by: Jeremy Huddleston <jeremyhu@apple.com>
Dave Airlie [Sun, 22 Apr 2012 07:09:05 +0000 (08:09 +0100)]
r600g: enable GLSL130 on all cards
Alexandre Demers sent me some cayman results with no major problems.
I'll rip out the env var in a week or so.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Dave Airlie [Sat, 21 Apr 2012 16:46:35 +0000 (17:46 +0100)]
r600g: enable dual src blending on r600 cards
tested on my rv610 and it passes the tests with no hangs.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Dave Airlie [Sat, 21 Apr 2012 16:40:32 +0000 (17:40 +0100)]
r600g: enable GLSL 1.30 for r600 class
Full piglit run on my rv610 with no regressions.
This only leaves cayman, however my cayman is resisting my attempt
to get through a full piglit run.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Dave Airlie [Sat, 21 Apr 2012 09:37:08 +0000 (10:37 +0100)]
r600: enable glsl 1.30 on r700
I've done a piglit run on rv740 and confirmed no regressions.
We don't get GL3 on r700 due to transform feedback being busted still.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Dave Airlie [Fri, 20 Apr 2012 18:33:22 +0000 (19:33 +0100)]
r600g: report INTEGERS cap if glsl130 is on.
This cap is used by u_blitter to decide if it can use integers
in vertex data.
fixes some crashes with glsl130 in piglit
Signed-off-by: Dave Airlie <airlied@redhat.com>
Dave Airlie [Fri, 20 Apr 2012 18:01:57 +0000 (19:01 +0100)]
r600g: enable glsl 130 on evergreen.
I've done a piglit run on my SUMO machine and I see no regressions.
Lots of things to fix (skip->fail), but hey maybe we can fix them
if we can see them.
I'll try and work my way across r600,700,cayman sometime if nobody
else gets to them.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Vadim Girlin [Mon, 9 Apr 2012 15:35:40 +0000 (19:35 +0400)]
r600g: disable I2F conversion for InstanceID if integers are supported
Signed-off-by: Vadim Girlin <vadimgirlin@gmail.com>
Reviewed-by: Michel Dänzer <michel@daenzer.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Vadim Girlin [Mon, 9 Apr 2012 15:35:39 +0000 (19:35 +0400)]
r600g: store glsl_feature_level in the r600_screen
Signed-off-by: Vadim Girlin <vadimgirlin@gmail.com>
Reviewed-by: Michel Dänzer <michel@daenzer.net>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Brian Paul [Wed, 18 Apr 2012 16:47:10 +0000 (10:47 -0600)]
mesa: move gl_array_attrib::NewState to gl_array_object::NewArrays
The field wasn't actually used before and it's not used now either.
But this is a more logical place for it and will hopefully allow
doing smarter draw/array validation (per array object) in the future.
Reviewed-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Brian Paul [Thu, 19 Apr 2012 20:38:45 +0000 (14:38 -0600)]
docs: recommend 2GB of RAM for Viewperf tests
Eric Anholt [Wed, 11 Apr 2012 20:24:22 +0000 (13:24 -0700)]
i965: Convert live interval computation to using live variable analysis.
Our previous live interval analysis just said that anything in a loop
was live for the whole loop. If you had to spill a reg in a loop,
then we would consider the unspilled value live across the loop too,
so you never made progress by spilling. Eventually it would consider
everything in the loop unspillable and fail out.
With the new analysis, things completely deffed and used inside the
loop won't be marked live across the loop, so even if you
spill/unspill something that used to be live across the loop, you
reduce register pressure. But you usually don't even have to spill
any more, since our intervals are smaller than before.
This fixes assertion failure trying to compile the shader for the
"glyphy" text rasterier and piglit glsl-fs-unroll-explosion.
Improves Unigine Tropics performance 1.3% +/- 0.2% (n=5), by allowing
more shaders to be compiled in 16-wide mode.
Eric Anholt [Wed, 11 Apr 2012 20:14:27 +0000 (13:14 -0700)]
i965: Move the old live interval analysis code next to the new live vars code.
I'm about to replace the insides of this using the new analysis.
Eric Anholt [Wed, 11 Apr 2012 20:12:33 +0000 (13:12 -0700)]
i965: Add support for live variable analysis using dataflow analysis.
Eric Anholt [Tue, 10 Apr 2012 19:01:50 +0000 (12:01 -0700)]
i965: Add basic block generator.
This takes the fs_inst list generated by the visitor, and generates a
list of basic blocks with edges between them. This is a building
block for data-flow analysis.