mesa.git
10 years agowinsys/radeon: remove some unused code
Marek Olšák [Wed, 16 Apr 2014 12:02:26 +0000 (14:02 +0200)]
winsys/radeon: remove some unused code

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
10 years agowinsys/radeon: remove is_handle_added array
Marek Olšák [Mon, 14 Apr 2014 13:32:39 +0000 (15:32 +0200)]
winsys/radeon: remove is_handle_added array

Use index -1 if a buffer is not added.

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
10 years agowinsys/radeon: remove local variable reloc from radeon_get_reloc
Marek Olšák [Mon, 14 Apr 2014 13:25:43 +0000 (15:25 +0200)]
winsys/radeon: remove local variable reloc from radeon_get_reloc

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
10 years agowinsys/radeon: remove parameter reloc from radeon_get_reloc
Marek Olšák [Mon, 14 Apr 2014 13:24:04 +0000 (15:24 +0200)]
winsys/radeon: remove parameter reloc from radeon_get_reloc

Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
10 years agoutil: Add __declspec(noreturn) to _debug_assert_fail().
José Fonseca [Wed, 16 Apr 2014 16:45:56 +0000 (17:45 +0100)]
util: Add __declspec(noreturn) to _debug_assert_fail().

Mostly for consistency; as MSVC's static source code analysis doesn't
seem to rely on assertions, but instead on different kind of source
annotations( http://msdn.microsoft.com/en-us/library/hh916383.aspx ).

Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agoauxiliary/os,auxiliary/util: Fix the `‘noreturn’ function does return` warning.
José Fonseca [Wed, 16 Apr 2014 16:36:51 +0000 (17:36 +0100)]
auxiliary/os,auxiliary/util: Fix the `‘noreturn’ function does return` warning.

Now that _debug_assert_fail() has the noreturn attribute, it is better
that execution truly never returns.  Not just for sake of silencing the
warning, but because the code at the return IP address may be invalid or
lead to inconsistent results.

This removes support for the GALLIUM_ABORT_ON_ASSERT debugging
environment variable, but between the usefulness of
GALLIUM_ABORT_ON_ASSERT and better static code analysis I think better
static code analysis wins.

Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agoscons: Enable building through Clang Static Analyzer.
José Fonseca [Wed, 16 Apr 2014 16:26:05 +0000 (17:26 +0100)]
scons: Enable building through Clang Static Analyzer.

Same intent as commit a45a50a4828e1357e9555474bc127c5585b3a420,
but this the C compiler is detected via C-preprocessor macros,
similar to how autotools do it, as that seems to be the most
reliable method.

Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agogallium glsl: Fix crash with piglit fs-deref-literal-array-of-structs.shader_test
Maarten Lankhorst [Thu, 17 Apr 2014 08:34:10 +0000 (10:34 +0200)]
gallium glsl: Fix crash with piglit fs-deref-literal-array-of-structs.shader_test

This allows the following shader code to work without a weird crash:

struct Foo {
  int value[1];
};

int actual_value = Foo[2](Foo(int[1](100)), Foo(int[1](200)))[i].value[0];

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@canonical.com>
10 years agonouveau/vdec: small fixes to h264 handling
Maarten Lankhorst [Tue, 15 Apr 2014 15:35:31 +0000 (17:35 +0200)]
nouveau/vdec: small fixes to h264 handling

nouveau_vp3_inter_sizes requires sliec_count as argument just
as the other places that call it from h264 code do. Hopefully
fixes something.

Fix the status_vp code to allow status == 0 too, when processing
hasn't started yet.

set h264->second_field correctly.

10 years agost/xa: Cache render target surface
Thomas Hellstrom [Tue, 8 Apr 2014 20:48:51 +0000 (16:48 -0400)]
st/xa: Cache render target surface

Otherwise it will trick the gallium driver into thinking that the render
target has actually changed (due to different pipe_surface pointing to
same underlying pipe_resource).  This is really badness for tiling GPUs
like adreno.

This also appears to fix a rendering error with Motif on vmwgfx.
Why that is is still under investigation.

Based on an idea by Rob Clark.

Cc: "10.0 10.1" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Rob Clark <robclark@freedesktop.org>
10 years agost/xa: scissor to help tilers
Rob Clark [Tue, 8 Apr 2014 20:48:50 +0000 (16:48 -0400)]
st/xa: scissor to help tilers

Keep track of the maximal bounds of all the operations and set scissor
accordingly.  For tiling GPU's this can be a big win by reducing the
memory bandwidth spent moving pixels from system memory to tile buffer
and back.

You could imagine being more sophisticated and splitting up disjoint
operations.  But this simplistic approach is good enough for the common
cases.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
10 years agost/xa: remove unneeded args
Rob Clark [Tue, 8 Apr 2014 20:48:49 +0000 (16:48 -0400)]
st/xa: remove unneeded args

Signed-off-by: Rob Clark <robclark@freedesktop.org>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
10 years agoglsl: Small optimization for constant conditionals
Iago Toral Quiroga [Tue, 15 Apr 2014 10:30:40 +0000 (12:30 +0200)]
glsl: Small optimization for constant conditionals

Once the relevant branch has been identified do not iterate over the
instructions in the branch, do a linked list insertion instead to avoid the
loop.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoglsl: Fix incorrect indentation.
Iago Toral Quiroga [Mon, 14 Apr 2014 07:14:23 +0000 (09:14 +0200)]
glsl: Fix incorrect indentation.

Acked-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agometa: Clip src/dest rects in BlitFramebuffer, using the scissor
Chris Forbes [Wed, 16 Apr 2014 02:15:27 +0000 (14:15 +1200)]
meta: Clip src/dest rects in BlitFramebuffer, using the scissor

Fixes piglit's fbo-blit-stretch test on drivers which use the meta path.
(i965: should fix Broadwell, but also fixes Sandybridge/Ivybridge/Haswell
since this test falls off the blorp path now due to format conversion)

V2: Use scissor instead of just mangling the rects, to avoid texcoord
rounding problems. (Thanks Marek)

V3: Rebase on Eric's CTSI meta changes; re-add _mesa_update_state in the
CTSI path so that _mesa_clip_blit sees the correct bounds.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=77414
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Tested-by: Anuj Phogat <anuj.phogat@gmail.com>
10 years agomesa: fix check for dummy renderbuffer in _mesa_FramebufferRenderbufferEXT()
Samuel Iglesias Gonsalvez [Thu, 3 Apr 2014 06:30:06 +0000 (08:30 +0200)]
mesa: fix check for dummy renderbuffer in _mesa_FramebufferRenderbufferEXT()

According to the spec:
<renderbuffertarget> must be RENDERBUFFER and <renderbuffer>
should be set to the name of the renderbuffer object to be
attached to the framebuffer.  <renderbuffer> must be either
zero or the name of an existing renderbuffer object of type
<renderbuffertarget>, otherwise an INVALID_OPERATION error is
generated.

This patch changes the previous returned GL_INVALID_VALUE to
GL_INVALID_OPERATION.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=76894

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Samuel Iglesias Gonsalvez <siglesias@igalia.com>
10 years agoi965: Don't make instructions with a null dest a barrier to scheduling.
Matt Turner [Wed, 9 Apr 2014 20:38:14 +0000 (13:38 -0700)]
i965: Don't make instructions with a null dest a barrier to scheduling.

Now that we properly track accumulator dependencies, the scheduler is
able to schedule instructions between the mach and mov in the common
the integer multiplication pattern:

   mul  acc0, x, y
   mach null, x, y
   mov  dest, acc0

Since a null destination implies no dependency on the destination, we
can also safely schedule instructions (that don't write the accumulator)
between the mul and mach.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/fs: Change fs_visitor::emit_lrp to use MAC for gen<6
Juha-Pekka Heikkila [Fri, 28 Mar 2014 13:28:33 +0000 (15:28 +0200)]
i965/fs: Change fs_visitor::emit_lrp to use MAC for gen<6

This allows us to emit ADD/MUL/MAC instead of MUL/ADD/MUL/ADD,
saving one instruction and two temporary registers.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
10 years agoi965/fs: Add support for the MAC instruction.
Juha-Pekka Heikkila [Fri, 28 Mar 2014 13:28:32 +0000 (15:28 +0200)]
i965/fs: Add support for the MAC instruction.

This allows us to generate the MAC (multiply-accumulate) instruction,
which can be used to implement some expressions in fewer instructions
than doing a series of MUL and ADDs.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
10 years agoi965/vec4: Change vec4_visitor::emit_lrp to use MAC for gen<6
Juha-Pekka Heikkila [Fri, 28 Mar 2014 13:28:31 +0000 (15:28 +0200)]
i965/vec4: Change vec4_visitor::emit_lrp to use MAC for gen<6

This allows us to emit ADD/MUL/MAC instead of MUL/ADD/MUL/ADD,
saving one instruction and two temporary registers.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
10 years agoi965/vec4: Add support for the MAC instruction.
Juha-Pekka Heikkila [Fri, 28 Mar 2014 13:28:30 +0000 (15:28 +0200)]
i965/vec4: Add support for the MAC instruction.

This allows us to generate the MAC (multiply-accumulate) instruction,
which can be used to implement some expressions in fewer instructions
than doing a series of MUL and ADDs.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
10 years agoi965: Add writes_accumulator flag
Juha-Pekka Heikkila [Fri, 4 Apr 2014 13:51:59 +0000 (16:51 +0300)]
i965: Add writes_accumulator flag

Our hardware has an "accumulator" register, which can be used to store
intermediate results across multiple instructions.  Many instructions
can implicitly write a value to the accumulator in addition to their
normal destination register.  This is enabled by the "AccWrEn" flag.

This patch introduces a new flag, inst->writes_accumulator, which
allows us to express the AccWrEn notion in the IR.  It also creates a
n ALU2_ACC macro to easily define emitters for instructions that
implicitly write the accumulator.

Previously, we only supported implicit accumulator writes from the
ADDC, SUBB, and MACH instructions.  We always enabled them on those
instructions, and left them disabled for other instructions.

To take advantage of the MAC (multiply-accumulate) instruction, we
need to be able to set AccWrEn on other types of instructions.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
10 years agoi965: Add is_accumulator() function.
Juha-Pekka Heikkila [Wed, 9 Apr 2014 19:08:12 +0000 (12:08 -0700)]
i965: Add is_accumulator() function.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
10 years agoi965: Add reads_accumulator_implicitly() function.
Matt Turner [Wed, 9 Apr 2014 19:01:49 +0000 (12:01 -0700)]
i965: Add reads_accumulator_implicitly() function.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agomesa: Add error condition for integer formats in glGetTexImage()
Anuj Phogat [Wed, 19 Mar 2014 00:01:19 +0000 (17:01 -0700)]
mesa: Add error condition for integer formats in glGetTexImage()

OpenGL 4.0 spec, page 306 suggests an INVALID_OPERATION in glGetTexImage
if :
   "format is one of the integer formats in table 3.3 and the internal
    format of the texture image is not integer, or format is not one of
    the integer formats in table 3.3 and the internal format is integer."

V2: Use helper function _mesa_is_format_integer()

Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agomesa: Add helper function _mesa_is_format_integer()
Anuj Phogat [Tue, 25 Mar 2014 20:00:12 +0000 (13:00 -0700)]
mesa: Add helper function _mesa_is_format_integer()

This function will be used in the following patch.

Cc: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agomesa: Fix glGetVertexAttribi(GL_VERTEX_ATTRIB_ARRAY_SIZE)
Anuj Phogat [Thu, 13 Mar 2014 01:16:21 +0000 (18:16 -0700)]
mesa: Fix glGetVertexAttribi(GL_VERTEX_ATTRIB_ARRAY_SIZE)

mesa currently returns 4 when GL_VERTEX_ATTRIB_ARRAY_SIZE is queried
for a vertex array initially set up with size=GL_BGRA. This patch
makes changes to return size=GL_BGRA as required by the spec.

Fixes Khronos OpenGL CTS test: vertex_array_bgra_basic.test

V2: Use array->Format instead of adding a new variable

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: <mesa-stable@lists.freedesktop.org>
10 years agoglsl: Fix copy-paste error in linker_warning()
Anuj Phogat [Sat, 8 Mar 2014 00:48:35 +0000 (16:48 -0800)]
glsl: Fix copy-paste error in linker_warning()

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agor600g: Disable LLVM by default at runtime for graphics
Michel Dänzer [Wed, 16 Apr 2014 06:00:09 +0000 (15:00 +0900)]
r600g: Disable LLVM by default at runtime for graphics

For graphics, the LLVM compiler backend currently has many shortcomings
compared to the non-LLVM one. E.g. it can't handle geometry shaders yet,
but that's just the tip of the iceberg.

So building Mesa with --enable-r600-llvm-compiler is currently not
recommended for anyone who doesn't want to work on fixing those issues.
However, for protection of users who end up enabling it anyway for some
reason, let's disable the LLVM backend at runtime by default. It can be
enabled with the environment variable R600_DEBUG=llvm.

Cc: "10.1" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
10 years agogallivm: fix compilation with llvm 3.5 r206241+
Roland Scheidegger [Wed, 16 Apr 2014 01:46:20 +0000 (03:46 +0200)]
gallivm: fix compilation with llvm 3.5 r206241+

Just adjust to the ever-changing API, pass in MCContext when creating the
MCDisassembler.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
10 years agoRevert "scons: Enable building through Clang Static Analyzer."
José Fonseca [Wed, 16 Apr 2014 12:13:48 +0000 (13:13 +0100)]
Revert "scons: Enable building through Clang Static Analyzer."

This reverts commit a45a50a4828e1357e9555474bc127c5585b3a420.

Unfortunately gcc dumps argv[0] as the first word of --version, so it is
unreliable for detecting gcc.

In particular `cc --version` and `i686-w64-mingw32-gcc --version` give
wrong results.

A better solution needs to be found -- most likely using C-preprocessing
like autotools does.  Revert for now.

10 years agor600g,radeonsi: share some of gfx flush code
Marek Olšák [Sat, 12 Apr 2014 16:12:11 +0000 (18:12 +0200)]
r600g,radeonsi: share some of gfx flush code

Reviewed-by: Christian König <christian.koenig@amd.com>
10 years agor600g,radeonsi: share r600_flush_from_st
Marek Olšák [Sat, 12 Apr 2014 15:57:32 +0000 (17:57 +0200)]
r600g,radeonsi: share r600_flush_from_st

Reviewed-by: Christian König <christian.koenig@amd.com>
10 years agor600g: merge r600_flush with r600_context_flush
Marek Olšák [Sat, 12 Apr 2014 15:53:57 +0000 (17:53 +0200)]
r600g: merge r600_flush with r600_context_flush

Reviewed-by: Christian König <christian.koenig@amd.com>
10 years agoradeonsi: merge si_flush with si_context_flush
Marek Olšák [Sat, 12 Apr 2014 15:34:44 +0000 (17:34 +0200)]
radeonsi: merge si_flush with si_context_flush

This also removes si_flush_gfx_ring.

Reviewed-by: Christian König <christian.koenig@amd.com>
10 years agogallium/radeon: create and return a fence in the flush function
Marek Olšák [Sat, 12 Apr 2014 15:01:52 +0000 (17:01 +0200)]
gallium/radeon: create and return a fence in the flush function

All flush functions get a fence parameter. cs_create_fence is removed.

Reviewed-by: Christian König <christian.koenig@amd.com>
10 years agor600g: remove redundant r600_flush_dma_from_winsys
Marek Olšák [Fri, 11 Apr 2014 20:40:14 +0000 (22:40 +0200)]
r600g: remove redundant r600_flush_dma_from_winsys

Reviewed-by: Christian König <christian.koenig@amd.com>
10 years agowinsys/radeon: fold cs_set_flush_callback into cs_create
Marek Olšák [Fri, 11 Apr 2014 20:14:27 +0000 (22:14 +0200)]
winsys/radeon: fold cs_set_flush_callback into cs_create

Reviewed-by: Christian König <christian.koenig@amd.com>
10 years agoradeonsi: cleanup redundant computation of flush flags and rename a function
Marek Olšák [Fri, 11 Apr 2014 19:51:12 +0000 (21:51 +0200)]
radeonsi: cleanup redundant computation of flush flags and rename a function

Reviewed-by: Christian König <christian.koenig@amd.com>
10 years agor600g: remove redundant r600_flush_from_winsys
Marek Olšák [Fri, 11 Apr 2014 19:41:05 +0000 (21:41 +0200)]
r600g: remove redundant r600_flush_from_winsys

Reviewed-by: Christian König <christian.koenig@amd.com>
10 years agowinsys/radeon: remove cs_write_reloc, add simpler cs_get_reloc
Marek Olšák [Fri, 11 Apr 2014 18:34:46 +0000 (20:34 +0200)]
winsys/radeon: remove cs_write_reloc, add simpler cs_get_reloc

The only difference is that it doesn't write to the CS and only returns
the index.

Reviewed-by: Christian König <christian.koenig@amd.com>
10 years agowinsys/radeon: consolidate hash table lookup
Marek Olšák [Fri, 11 Apr 2014 11:20:02 +0000 (13:20 +0200)]
winsys/radeon: consolidate hash table lookup

I should have done this long ago.

Reviewed-by: Christian König <christian.koenig@amd.com>
10 years agoscons: Add an analyze option.
José Fonseca [Mon, 14 Apr 2014 13:31:02 +0000 (14:31 +0100)]
scons: Add an analyze option.

For Clang static code analyzer, the scan-build script will produce more
comprehensive output.  Nevertheless you can invoke it as

  CC=clang CXX=clang++ scons analyze=1

For MSVC this is the best way to use its static code analysis.  Simply
invoke as

  scons analyze=1

Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agoutil/u_debug: Add noreturn attribute to _debug_assert_fail().
José Fonseca [Mon, 14 Apr 2014 11:13:32 +0000 (12:13 +0100)]
util/u_debug: Add noreturn attribute to _debug_assert_fail().

As recommended by
http://clang-analyzer.llvm.org/annotations.html#attr_noreturn

Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agoscons: Enable building through Clang Static Analyzer.
José Fonseca [Mon, 14 Apr 2014 11:10:25 +0000 (12:10 +0100)]
scons: Enable building through Clang Static Analyzer.

By accurately detecting gcc/clang through --version option instead
of executable name.

Clang Static Analyzer reports many issues, most false positives, but it
found at least one real and subtle use-after-free issue
in st_texture_get_sampler_view():

  http://people.freedesktop.org/~jrfonseca/scan-build-2014-04-14-1/report-869047.html#EndPath

Reviewed-by: Brian Paul <brianp@vmware.com>
10 years agoglsl: Properly handle blocks that define the same field name.
Iago Toral Quiroga [Fri, 4 Apr 2014 13:11:15 +0000 (15:11 +0200)]
glsl: Properly handle blocks that define the same field name.

Currently we can have name space collisions between blocks that define the same
fields. For example:

in block
{
    vec4 Color;
} In[];

out block
{
    vec4 Color;
} Out;

These two blocks will assign the same interface name (block.Color) to the Color
field in flatten_named_interface_blocks_declarations.cpp, leading to havoc.
This was breaking badly the gl-320-primitive-shading test from ogl-samples.

The patch uses the block instance name to avoid collisions, producing names
like block.In.Color and block.Out.Color to avoid the name clash.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=76394
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agor600g/radeonsi: Map transfer staging texture unsynchronized when possible
Michel Dänzer [Tue, 15 Apr 2014 05:45:13 +0000 (14:45 +0900)]
r600g/radeonsi: Map transfer staging texture unsynchronized when possible

The transfer staging texture is always freshly allocated, so for write-only
transfers we don't need to explicitly wait for the BO to become idle.

Squeezes a few hundered MB/s more out of x11perf -shmput500 with glamor.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agoRevert "i965/fs: Only sweep NOPs if register coalescing made progress."
Matt Turner [Wed, 16 Apr 2014 00:27:55 +0000 (17:27 -0700)]
Revert "i965/fs: Only sweep NOPs if register coalescing made progress."

This reverts commit f092e8951ce5212ba3cbb382ce3a6666eb6c9bed.

Didn't mean to push this...

10 years agoi965/fs: Only sweep NOPs if register coalescing made progress.
Matt Turner [Tue, 15 Apr 2014 23:28:04 +0000 (16:28 -0700)]
i965/fs: Only sweep NOPs if register coalescing made progress.

Otherwise there's nothing to do.

10 years agoi965: Fix buffer overruns in MSAA MCS buffer clearing.
Eric Anholt [Mon, 14 Apr 2014 23:52:43 +0000 (16:52 -0700)]
i965: Fix buffer overruns in MSAA MCS buffer clearing.

This manifested as rendering failures or sometimes GPU hangs in
compositors when they accidentally got MSAA visuals due to a bug in the X
Server.  Today we decided that the problem in compositors was equivalent
to a corruption bug we'd noticed recently in resizing MSAA-visual
glxgears, and debugging got a lot easier.

When we allocate our MCS MT, libdrm takes the size we request, aligns it
to Y tile size (blowing it up from 300x300=900000 bytes to 384*320=122880
bytes, 30 pages), then puts it into a power-of-two-sized BO (131072 bytes,
32 pages).  Because it's Y tiled, we attach a 384-byte-stride fence to it.
When we memset by the BO size in Mesa, between bytes 122880 and 131072 the
data gets stored to the first 20 or so scanlines of each of the 3 tiled
pages in that row, even though only 2 of those pages were allocated by
libdrm.  In the glxgears case, the missing 3rd page happened to
consistently be the static VBO that got mapped right after the first MCS
allocation, so corruption only appeared once window resize made us throw
out the old MCS and then allocate the same BO to back the new MCS.

Instead, just memset the amount of data we actually asked libdrm to
allocate for, which will be smaller (more efficient) and not overrun.
Thanks go to Kenneth for doing most of the hard debugging to eliminate a
lot of the search space for the bug.

Cc: "10.0 10.1" <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=77207
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agometa: Add support for MSAA resolves from 2D_MS_ARRAY textures.
Eric Anholt [Sat, 1 Mar 2014 00:39:03 +0000 (16:39 -0800)]
meta: Add support for MSAA resolves from 2D_MS_ARRAY textures.

We don't have any piglit tests for this currently.

v2: Use vec3s for the texcoords so it has some hope of working.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agometa: Add an accelerated glCopyTexSubImage using glBlitFramebuffer.
Eric Anholt [Fri, 28 Feb 2014 21:23:25 +0000 (13:23 -0800)]
meta: Add an accelerated glCopyTexSubImage using glBlitFramebuffer.

You'll note from the previous commits that there's something of a loop
here: You call CTSI, which calls BlitFB, then if things go wrong that
falls back to CTSI.  As a result, meta CTSI reaches over into blitfb to
tell it "no, don't try that fallback".

v2: Drop the _mesa_update_state(), which was only necessary due to use of
    _mesa_clip_blit() in _mesa_meta_BlitFramebuffer() in another patch
    series.
v3: Drop an _EXT suffix I copy-and-pasted.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com> (v2)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agometa: Add support for CUBE_MAP_ARRAY to generatemipmap.
Eric Anholt [Tue, 8 Apr 2014 20:34:39 +0000 (13:34 -0700)]
meta: Add support for CUBE_MAP_ARRAY to generatemipmap.

I added support to bind_fbo_image in the process of building meta
CopyTexSubImage, and found that it broke generatemipmap because previously
we would just throw a GL error there and then end up with an incomplete
FBO and fallback.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agometa: Infer bind_fbo_image parameters from an incoming image.
Eric Anholt [Mon, 17 Mar 2014 23:45:48 +0000 (16:45 -0700)]
meta: Infer bind_fbo_image parameters from an incoming image.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agometa: Move bind_fbo_image() code back to meta.c, to reuse it elsewhere.
Eric Anholt [Mon, 17 Mar 2014 23:26:42 +0000 (16:26 -0700)]
meta: Move bind_fbo_image() code back to meta.c, to reuse it elsewhere.

I need to do the same code again for CopyTexSubImage().

v2: Drop incorrect, not-terribly-useful comment (review by Ken)

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com> (v1)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agometa: Refactor the BlitFramebuffer depth CopyTexImage fallback.
Eric Anholt [Fri, 28 Feb 2014 20:47:54 +0000 (12:47 -0800)]
meta: Refactor the BlitFramebuffer depth CopyTexImage fallback.

This avoids a ReadPixels() if there's accelerated CopyTexImage present.
It now requires GLSL as opposed to just fragment programs, but we don't
have any drivers that do ARB_fp but not GLSL.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agometa: Refactor the BlitFramebuffer color CopyTexImage fallback.
Eric Anholt [Fri, 28 Feb 2014 20:36:02 +0000 (12:36 -0800)]
meta: Refactor the BlitFramebuffer color CopyTexImage fallback.

There shouldn't be anything special about copying out a subset of the src
rb to a temp before texturing from it, so just do it when we're figuring
out our src texture binding.

This drops Anuj's change to copy an extra border of 1 pixel around the src
area.  I can't see how that change could be valid, and presumably if
there's some filtering problem at edges we just need to set the right
wrap mode.

v2: Don't fall back to swrast on non-2D/RECT/2D_MS textures when we can
    still CopyTexSubImage.  Fixes a segfault regression on i965 with
    gl-3.2-layered-rendering-blit.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com> (v1)
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org> (v1)
Tested-by: Anuj Phogat <anuj.phogat@gmail.com>
10 years agometa: Drop blit src size fallback.
Eric Anholt [Fri, 28 Feb 2014 20:19:09 +0000 (12:19 -0800)]
meta: Drop blit src size fallback.

I think we can assert that renderbuffer size is <= maximum 2D texture
size.  Our source coordinates should have already been clipped to the src
renderbuffer size, but haven't actually (so we could potentially have
trouble if there's scaling, and we're in the CopyTexImage path that tries
to use src size).  However, this texture size dependency was blocking the
next refactors, so I'm not sure if we want to go ahead with this series
before we get the clipping sorted out or not.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965: Avoid dependency hints on math opcodes
Mike Stroyan [Thu, 13 Feb 2014 00:24:55 +0000 (17:24 -0700)]
i965: Avoid dependency hints on math opcodes

  Putting NoDDClr and NoDDChk dependency control on instruction
sequences that include math opcodes can cause corruption of channels.
Treat math opcodes like send opcodes and suppress dependency hinting.

Signed-off-by: Mike Stroyan <mike@LunarG.com>
Tested-by: Tony Bertapelli <anthony.p.bertapelli@intel.com>
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Matt Turner <mattst88@gmail.com>
10 years agoi965: Expand INTEL_DEBUG to uint64_t.
Matt Turner [Sat, 12 Apr 2014 05:27:31 +0000 (22:27 -0700)]
i965: Expand INTEL_DEBUG to uint64_t.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agodri: Expand driParseDebugString return value to uint64_t.
Matt Turner [Sat, 12 Apr 2014 04:11:38 +0000 (21:11 -0700)]
dri: Expand driParseDebugString return value to uint64_t.

Users will downcast if they don't have >32 debug flags.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoi965/fs: Remove dead_code_eliminate_local().
Matt Turner [Mon, 14 Apr 2014 18:18:59 +0000 (11:18 -0700)]
i965/fs: Remove dead_code_eliminate_local().

Subsumed by the new dead_code_eliminate() function. No shader-db
changes.

Reviewed-by: Eric Anholt <eric@anholt.net>
10 years agoi965/fs: Clear variable from live-set if it's completely overwritten.
Matt Turner [Mon, 14 Apr 2014 18:17:13 +0000 (11:17 -0700)]
i965/fs: Clear variable from live-set if it's completely overwritten.

One program affected:

instructions in affected programs:     246 -> 244 (-0.81%)

Reviewed-by: Eric Anholt <eric@anholt.net>
10 years agoi965/fs: Reimplement dead_code_elimination().
Matt Turner [Sun, 13 Apr 2014 00:40:18 +0000 (17:40 -0700)]
i965/fs: Reimplement dead_code_elimination().

total instructions in shared programs: 1653399 -> 1651790 (-0.10%)
instructions in affected programs:     92157 -> 90548 (-1.75%)
GAINED:                                2
LOST:                                  2

Also significantly reduces the number of optimization loop iterations:

total loop iterations in shared programs: 39724 -> 31651 (-20.32%)
loop iterations in affected programs:     21617 -> 13544 (-37.35%)

Including some great pathological cases, like 29 -> 3 in Strike Suit
Zero and 24 -> 3 in Dota2.

Reviewed-by: Eric Anholt <eric@anholt.net>
10 years agoi965/vec4: Let DCE eliminate dead writes in other basic blocks.
Matt Turner [Mon, 24 Mar 2014 20:23:13 +0000 (13:23 -0700)]
i965/vec4: Let DCE eliminate dead writes in other basic blocks.

We previously stopped searching for unread writes after encountering
control flow, but we can instead just search backwards until we hit
control flow.

instructions in affected programs:     22854 -> 22194 (-2.89%)

10 years agoi965/gs: Add dummy source to prepare_channel_masks instruction.
Matt Turner [Sun, 30 Mar 2014 00:44:45 +0000 (17:44 -0700)]
i965/gs: Add dummy source to prepare_channel_masks instruction.

The generator uses its destination as a source implicitly, which breaks
some assumptions in dead code elimination. Giving the instruction a
source allows us to reason about it better.

10 years agoglsl: Use M_PI_* macros.
Matt Turner [Sun, 6 Apr 2014 05:03:04 +0000 (22:03 -0700)]
glsl: Use M_PI_* macros.

Notice our multiple values for M_PI_2, which rounded ...32 up to
...4 and ...5.

10 years agoi965: Disable Z16 in all APIs.
Kenneth Graunke [Sun, 13 Apr 2014 21:19:03 +0000 (14:19 -0700)]
i965: Disable Z16 in all APIs.

We originally thought that GL 3.0 required GL_DEPTH_COMPONENT16 to map
exactly to Z16.  However, we misread the specification, thanks in part
to LaTeX reordering the tables in the PDF.

Page 180 of the GL 3.0 specification (glspec30.20080923.pdf) says:
"[...] memory allocation per texture component is assigned by the GL to
match the allocations listed in tables 3.16-3.18 as closely as possible.
[...]

Required Texture Formats
[...]
In addition, implementations are required to support the following sized
internal formats.  Requesting one of these internal formats for any
texture type will allocate exactly the internal component sizes and
types shown for that format in tables 3.16-3.17:"

Notably, however, GL_DEPTH_COMPONENT16 does /not/ appear in table 3.16
or table 3.17.  It appears in table 3.18, where the "exact" rule doesn't
apply, and it falls back to the "closely as possible" rule.

The confusing part is that the ordering of the tables in the PDF is:

Table 3.16 (pages 182-184)
Table 3.18 (bottom of page 184 to top of 185)
Table 3.17 (page 185)

Presumably, people saw table 3.16, then saw the table immediately
following with DEPTH_COMPONENT* formats, and assumed it was 3.17.

Based on a patch by Chia-I Wu, but without the driconf option to force
Z16 to be used.  It's not required, and there's apparently no benefit
to actually using it.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Chia-I Wu <olv@lunarg.com>
10 years agoi965: Update comments about Z16 being slow.
Kenneth Graunke [Sun, 13 Apr 2014 21:15:49 +0000 (14:15 -0700)]
i965: Update comments about Z16 being slow.

We've learned a few things since we originally disabled Z16; this attempts
to summarize the issue.  I am no expert on this subject, though, so the
comment may not be totally accurate.

I did some benchmarking on GM45 and Ironlake, and discovered that for
GLBenchmark 2.7 EgyptHD, using Z16 was 3% slower on GM45 (n=15), and
4.5% slower on Ironlake (n=95).  So, we can drop the "on Ivybridge"
aspect of the comment - it's always slower.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Chia-I Wu <olv@lunarg.com>
10 years agor600g/radeonsi: Use caching buffer manager for textures as well
Michel Dänzer [Thu, 10 Apr 2014 09:15:55 +0000 (18:15 +0900)]
r600g/radeonsi: Use caching buffer manager for textures as well

Significantly reduces BO allocation / destruction overhead for transfers,
e.g. measurable via x11perf -shm{ge,pu}t* with glamor.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
10 years agoi965/gen8: add debug code to show FS disasm with jump locations
Jordan Justen [Sat, 12 Apr 2014 00:22:46 +0000 (17:22 -0700)]
i965/gen8: add debug code to show FS disasm with jump locations

Copied from similar code in gen8_vec4_generator.cpp.

Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoilo: remove GPE state size estimation
Chia-I Wu [Thu, 10 Apr 2014 06:59:05 +0000 (14:59 +0800)]
ilo: remove GPE state size estimation

Use size defines from genhw.

10 years agoilo: remove GPE command size estimation
Chia-I Wu [Wed, 9 Apr 2014 18:45:23 +0000 (02:45 +0800)]
ilo: remove GPE command size estimation

Use size defines from genhw.

10 years agoilo: remove unused headers
Chia-I Wu [Sun, 13 Apr 2014 16:55:01 +0000 (00:55 +0800)]
ilo: remove unused headers

Remove intel_*.h.  brw_*.h is still needed by the state dumper and
disassembler.

10 years agoilo: use only defines from genhw headers
Chia-I Wu [Sun, 13 Apr 2014 12:07:40 +0000 (20:07 +0800)]
ilo: use only defines from genhw headers

Stop including classic driver headers in genhw.h, with some formatting fixes.

10 years agoilo: scripted conversion to genhw headers
Chia-I Wu [Sun, 13 Apr 2014 15:37:04 +0000 (23:37 +0800)]
ilo: scripted conversion to genhw headers

Hopefully my four hundred line sed script is correct.

10 years agoilo: add genhw headers
Chia-I Wu [Sat, 12 Apr 2014 16:33:00 +0000 (00:33 +0800)]
ilo: add genhw headers

All except genhw.h are generated by https://github.com/olvaffe/envytools/.
intel_chipset.h is deprecated.

10 years agoilo: avoid brw_wm_barycentric_interp_mode in compiler
Chia-I Wu [Fri, 11 Apr 2014 19:17:33 +0000 (03:17 +0800)]
ilo: avoid brw_wm_barycentric_interp_mode in compiler

In preparation for genhw.

10 years agoilo: add TOY_OPCODE_DO
Chia-I Wu [Fri, 11 Apr 2014 18:26:46 +0000 (02:26 +0800)]
ilo: add TOY_OPCODE_DO

We used to give BRW_OPCODE_DO a special meaning, while we should have used
TOY_OPCODE_DO.

10 years agogtest: Update to 1.7.0.
Vinson Lee [Thu, 27 Feb 2014 06:54:24 +0000 (22:54 -0800)]
gtest: Update to 1.7.0.

This patch fixes gtest build errors on Mac OS X 10.9.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=73106
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Tested-by: Ian Romanick <ian.d.romanick@intel.com>
Acked-by: Ian Romanick <ian.d.romanick@intel.com>
10 years agomesa: Consider gl_VertexID and gl_InstanceID active attribs
Chris Forbes [Sat, 12 Apr 2014 09:20:43 +0000 (21:20 +1200)]
mesa: Consider gl_VertexID and gl_InstanceID active attribs

Fixes piglit's spec/gl-3.2/get-active-attrib-returns-all-inputs.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agomesa: Extract is_active_attrib() in shaderapi
Chris Forbes [Sat, 12 Apr 2014 09:20:42 +0000 (21:20 +1200)]
mesa: Extract is_active_attrib() in shaderapi

The rules are about to get a bit more complex to account for
gl_InstanceID and gl_VertexID, which are system values.

Extracting this first avoids introducing duplication.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoglsl: Fix typo in interface block comment
Chris Forbes [Sun, 13 Apr 2014 05:01:07 +0000 (17:01 +1200)]
glsl: Fix typo in interface block comment

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
10 years agoegl-static: fix build after recent radeon winsys changes
Simone Scanzoni [Sun, 13 Apr 2014 00:36:20 +0000 (02:36 +0200)]
egl-static: fix build after recent radeon winsys changes

Signed-off-by: Marek Olšák <marek.olsak@amd.com>
10 years agomesa: Fix typo in error message
Chris Forbes [Sat, 12 Apr 2014 23:53:16 +0000 (11:53 +1200)]
mesa: Fix typo in error message

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
10 years agoi965: glClearBuffer() should only clear a single buffer.
Iago Toral Quiroga [Mon, 31 Mar 2014 12:17:08 +0000 (14:17 +0200)]
i965: glClearBuffer() should only clear a single buffer.

glClearBuffer() is currently clearing all active draw color buffers (all
buffers that have not been set to GL_NONE when calling glDrawBuffers) instead
of only clearing the one it receives as parameter. Altough brw_clear()
receives a bit mask indicating the color buffers that should be cleared,
this mask is ignored when calling brw_blorp_clear_color().

This was breaking the 'fbo-drawbuffers-none glClearBuffer' piglit test.

The patch provides the bit mask to brw_blorp_clear_color() so it can limit
clearing to the color buffers present in the mask.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=76832
Reviewed-by: Eric Anholt <eric@anholt.net>
10 years agoi965: Add comment to explain the weird-looking shadow compares.
Chris Forbes [Sat, 12 Apr 2014 10:32:21 +0000 (22:32 +1200)]
i965: Add comment to explain the weird-looking shadow compares.

This always looks crazy when I stumble across it, until I remember
what the hardware is doing. Describing it ought to short-circuit
that process next time :)

V2: Fix indents to 6 spaces, not 7.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
10 years agoglsl: Ignore loop-too-large heuristic if there's bad variable indexing.
Kenneth Graunke [Wed, 9 Apr 2014 01:09:43 +0000 (18:09 -0700)]
glsl: Ignore loop-too-large heuristic if there's bad variable indexing.

Many shaders use a pattern such as:

for (int i = 0; i < NUM_LIGHTS; i++) {
   ...access a uniform array, or shader input/output array...
}

where NUM_LIGHTS is a small constant (such as 2, 4, or 8).

The expectation is that the compiler will unroll those loops, turning
the array access into constant indexing, which is more efficient, and
which may enable array splitting and other optimizations.

In many cases, our heuristic fails - either there's another tiny nested
loop inside, or the estimated number of instructions is just barely
beyond the threshold.  So, we fail to unroll the loop, leaving the
variable indexing in place.

Drivers which don't support the particular flavor of variable indexing
will call lower_variable_index_to_cond_assign(), which generates piles
and piles of immensely inefficient code.  We'd like to avoid generating
that.

This patch detects unsupported forms of variable-indexing in loops, where
the array index is a loop induction variable.  In that case, it bypasses
the loop-too-large heuristic and forces unrolling.

Improves performance in various microbenchmarks: Gl32PSBump8 by 47%,
Gl32ShMapVsm by 80%, and Gl32ShMapPcf by 27%.  No changes in shader-db.

v2: Check ir->array for being an array or matrix, rather than the
    ir_dereference_array itself.
v3: Fix and expand statistics in commit message.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
10 years agoglsl: Rename loop_unroll_count::fail to "nested_loop."
Kenneth Graunke [Tue, 8 Apr 2014 22:33:27 +0000 (15:33 -0700)]
glsl: Rename loop_unroll_count::fail to "nested_loop."

The "fail" flag is set if loop_unroll_count encounters a nested loop;
calling the flag "nested_loop" is a bit clearer.

The original reasoning was that count is inaccurate (too small) if there
are nested loops, as we don't do any sort of analysis on the inner loop.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
10 years agoglsl: Pass gl_shader_compiler_optimizations to unroll_loops().
Kenneth Graunke [Wed, 9 Apr 2014 02:58:36 +0000 (19:58 -0700)]
glsl: Pass gl_shader_compiler_optimizations to unroll_loops().

Loop unrolling will need to know a few more options in the future.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
10 years agoglsl: Drop do_common_optimization's max_unroll_iterations parameter.
Kenneth Graunke [Tue, 8 Apr 2014 22:43:46 +0000 (15:43 -0700)]
glsl: Drop do_common_optimization's max_unroll_iterations parameter.

Now that we pass in gl_shader_compiler_options, it makes sense to just
use options->MaxUnrollIterations, rather than passing a separate
parameter.

Half of the invocations already passed options->MaxUnrollIterations,
while the other half passed in a hardcoded value of 32.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
10 years agoi965: Use EmitNoIndirect flags in lower_variable_index_to_cond_assign.
Kenneth Graunke [Wed, 9 Apr 2014 01:31:19 +0000 (18:31 -0700)]
i965: Use EmitNoIndirect flags in lower_variable_index_to_cond_assign.

This will prevent the two from getting out of sync again.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
10 years agoi965: Correct EmitNoIndirect shader compiler option flags.
Kenneth Graunke [Wed, 9 Apr 2014 01:29:01 +0000 (18:29 -0700)]
i965: Correct EmitNoIndirect shader compiler option flags.

These were out of sync with the flags used to control
lower_variable_index_to_cond_assign in brw_shader.cpp.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
10 years agoi965/fs: Reset reg_from when we can't coalesce.
Matt Turner [Wed, 9 Apr 2014 21:04:10 +0000 (14:04 -0700)]
i965/fs: Reset reg_from when we can't coalesce.

Not setting this would prevented coalescing after a failed attempt if
the sources for both MOVs were the same.

total instructions in shared programs: 1654531 -> 1650224 (-0.26%)
instructions in affected programs:     423167 -> 418860 (-1.02%)
GAINED:                                2
LOST:                                  0

Reviewed-by: Eric Anholt <eric@anholt.net>
10 years agoi965: Fill in a bunch of gen7/hsw data cache-related disasm.
Eric Anholt [Mon, 31 Mar 2014 16:23:24 +0000 (09:23 -0700)]
i965: Fill in a bunch of gen7/hsw data cache-related disasm.

This gets us disasm of atomic ops.

v2: Fix fallthrough on pre-gen7.  (bug caught by Ilia Mirkin).

Reviewed-by: Matt Turner <mattst88@gmail.com>
10 years agoi965: Stop setting up a 1:1 "attrib" member in our vertex inputs.
Eric Anholt [Sun, 6 Apr 2014 19:08:37 +0000 (12:08 -0700)]
i965: Stop setting up a 1:1 "attrib" member in our vertex inputs.

It's just the array index, so we can just go look at the array and see
which element we are.

No significant performance difference (n=140)

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
10 years agoi965: Skip a bunch of IB BO refcount twiddling.
Eric Anholt [Sun, 6 Apr 2014 18:16:48 +0000 (11:16 -0700)]
i965: Skip a bunch of IB BO refcount twiddling.

Improves cairo performance on glamor by 1.64828% +/- 1.04742% (n=65).

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
10 years agoi965/gen7: Skip repeated NULL depth/stencil state emits.
Eric Anholt [Sun, 6 Apr 2014 17:49:49 +0000 (10:49 -0700)]
i965/gen7: Skip repeated NULL depth/stencil state emits.

Improves cairo performance on glamor by 2.87752% +/- 0.966977 (n=57).

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
10 years agodocs: Fix ubo indexing description
Chris Forbes [Fri, 11 Apr 2014 20:31:05 +0000 (08:31 +1200)]
docs: Fix ubo indexing description

Ian points out that this being unrestricted was an oversight in the
spec, and is corrected in GLSL4.40.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
10 years agodraw: remove unused 'start' variable in draw_stats_clipper_primitives()
Brian Paul [Fri, 11 Apr 2014 17:46:55 +0000 (11:46 -0600)]
draw: remove unused 'start' variable in draw_stats_clipper_primitives()

It was computed, but never actually used.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>