mesa.git
14 years agost/xorg: Add a function to flush pending rendering and damage
Thomas Hellstrom [Wed, 24 Nov 2010 14:15:42 +0000 (15:15 +0100)]
st/xorg: Add a function to flush pending rendering and damage

This is needed to properly sync with host side rendering. For example,
make sure we flush colorkey painting before updating the overlay.

Signed-off-by: Thomas Hellstrom <thellstrom@vmware.com>
14 years agoegl_dri2: Fix one context, multiple surfaces.
Chia-I Wu [Wed, 24 Nov 2010 05:42:06 +0000 (13:42 +0800)]
egl_dri2: Fix one context, multiple surfaces.

When a context was made current to another surface, the old code did
this

  dri2_dpy->core->bindContext(cctx, ddraw, rdraw);
  dri2_dpy->core->unbindContext(old_cctx);

and there will be no current context due to the second line.

unbindContext should be called only when bindContext is not.  This fixes
a regression since d19afc57.  Thanks to Neil Roberts for noticing the
issue and creating a test case.

14 years agoi915: Disallow alpha, red, RG, and sRGB as render targets
Ian Romanick [Wed, 24 Nov 2010 00:17:21 +0000 (16:17 -0800)]
i915: Disallow alpha, red, RG, and sRGB as render targets

Fixes bugzilla #31832

NOTE: This is a candidate for the 7.9 branch.

14 years agoglsl: start restoring some geometry shader code
Brian Paul [Wed, 24 Nov 2010 00:23:42 +0000 (17:23 -0700)]
glsl: start restoring some geometry shader code

14 years agoglsl: better handling of linker failures
Brian Paul [Wed, 24 Nov 2010 00:18:44 +0000 (17:18 -0700)]
glsl: better handling of linker failures

Upon link error, exit translation loop, free program instructions.
Check for null pointers in calling code.

14 years agomesa: use gl_shader_type enum
Brian Paul [Wed, 24 Nov 2010 00:00:08 +0000 (17:00 -0700)]
mesa: use gl_shader_type enum

14 years agomesa: replace #defines with new gl_shader_type enum
Brian Paul [Tue, 23 Nov 2010 17:28:43 +0000 (10:28 -0700)]
mesa: replace #defines with new gl_shader_type enum

14 years agomesa: _mesa_valid_register_index() to validate register indexes
Brian Paul [Tue, 23 Nov 2010 17:12:55 +0000 (10:12 -0700)]
mesa: _mesa_valid_register_index() to validate register indexes

14 years agomesa: rename, make _mesa_register_file_name() non-static
Brian Paul [Tue, 23 Nov 2010 17:11:46 +0000 (10:11 -0700)]
mesa: rename, make _mesa_register_file_name() non-static

Plus remove unused parameter.

14 years agoglsl: use gl_register_file in a few places
Brian Paul [Mon, 22 Nov 2010 23:55:10 +0000 (16:55 -0700)]
glsl: use gl_register_file in a few places

14 years agoglsl: fix off by one in register index assertion
Brian Paul [Mon, 22 Nov 2010 22:06:47 +0000 (15:06 -0700)]
glsl: fix off by one in register index assertion

14 years agogallium/egl: fix r300 vs r600 loading
Alex Deucher [Tue, 23 Nov 2010 20:18:31 +0000 (15:18 -0500)]
gallium/egl: fix r300 vs r600 loading

Should fix:
https://bugs.freedesktop.org/show_bug.cgi?id=31841

14 years agoi965: Use the new embedded compare in SEL on gen6 for VS MIN and MAX opcodes.
Eric Anholt [Mon, 15 Nov 2010 08:01:04 +0000 (16:01 +0800)]
i965: Use the new embedded compare in SEL on gen6 for VS MIN and MAX opcodes.

Cuts the extra CMP instruction that used to precede SEL.

14 years agoi965: Don't upload line smooth params unless we're line smoothing.
Eric Anholt [Fri, 19 Nov 2010 15:04:35 +0000 (23:04 +0800)]
i965: Don't upload line smooth params unless we're line smoothing.

14 years agoi965: Don't upload line stipple pattern unless we're stippling.
Eric Anholt [Fri, 19 Nov 2010 15:02:07 +0000 (23:02 +0800)]
i965: Don't upload line stipple pattern unless we're stippling.

14 years agoi965: Don't upload polygon stipple unless required.
Eric Anholt [Fri, 19 Nov 2010 14:53:31 +0000 (22:53 +0800)]
i965: Don't upload polygon stipple unless required.

14 years agoi965: Move gen4 blend constant color to the gen4 blending file.
Eric Anholt [Fri, 19 Nov 2010 14:58:48 +0000 (22:58 +0800)]
i965: Move gen4 blend constant color to the gen4 blending file.

14 years agor600g: Removed duplicated call to tgsi_split_literal_constant().
Tilman Sauerbeck [Fri, 19 Nov 2010 21:31:43 +0000 (22:31 +0100)]
r600g: Removed duplicated call to tgsi_split_literal_constant().

Signed-off-by: Tilman Sauerbeck <tilman@code-monkey.de>
14 years agor300/compiler: Don't allow presubtract sources to be remapped twice
Tom Stellard [Tue, 23 Nov 2010 07:48:47 +0000 (23:48 -0800)]
r300/compiler: Don't allow presubtract sources to be remapped twice

https://bugs.freedesktop.org/show_bug.cgi?id=31193

NOTE: This is a candidate for the 7.9 branch.

14 years agor600g: Only compare active vertex elements
Mathias Fröhlich [Tue, 23 Nov 2010 07:39:30 +0000 (08:39 +0100)]
r600g: Only compare active vertex elements

Signed-off-by: Tilman Sauerbeck <tilman@code-monkey.de>
14 years agomesa: Clean up header file inclusion in syncobj.h.
Vinson Lee [Tue, 23 Nov 2010 05:51:49 +0000 (21:51 -0800)]
mesa: Clean up header file inclusion in syncobj.h.

14 years agollvmpipe: Remove unnecessary headers.
Vinson Lee [Tue, 23 Nov 2010 05:39:14 +0000 (21:39 -0800)]
llvmpipe: Remove unnecessary headers.

14 years agomesa: fix regression from b4bb6680200b5a898583392f4c831c02f41e63f7
Xiang, Haihao [Tue, 23 Nov 2010 00:52:23 +0000 (08:52 +0800)]
mesa: fix regression from b4bb6680200b5a898583392f4c831c02f41e63f7

Pending commands to the previous context aren't flushed since commit b4bb668

Reported-by: Oleksiy Krivoshey <oleksiyk@gmail.com>
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
14 years agor600c: fix VC flush on cedar and palm
Alex Deucher [Tue, 23 Nov 2010 00:27:58 +0000 (19:27 -0500)]
r600c: fix VC flush on cedar and palm

14 years agor600g: add support for ontario APUs
Alex Deucher [Mon, 22 Nov 2010 22:47:24 +0000 (17:47 -0500)]
r600g: add support for ontario APUs

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
14 years agor600c: add Ontario Fusion APU support
Alex Deucher [Mon, 22 Nov 2010 18:25:42 +0000 (13:25 -0500)]
r600c: add Ontario Fusion APU support

Signed-off-by: Alex Deucher <alexdeucher@gmail.com>
14 years agor300g: Avoid returning values in a static array, fixing a potential race
Mathias Fröhlich [Mon, 22 Nov 2010 22:19:52 +0000 (23:19 +0100)]
r300g: Avoid returning values in a static array, fixing a potential race

(Marek: added the initializion of "vec" in the default statement)

NOTE: This is a candidate for the 7.9 branch.

Signed-off-by: Marek Olšák <maraeo@gmail.com>
14 years agor600g: fix some winsys functions to deal properly with evergreen
Alex Deucher [Mon, 22 Nov 2010 22:39:54 +0000 (17:39 -0500)]
r600g: fix some winsys functions to deal properly with evergreen

Are these functions actually used anywhere?

14 years agor600g: fix additional EVENT_WRITE packet
Alex Deucher [Mon, 22 Nov 2010 22:39:16 +0000 (17:39 -0500)]
r600g: fix additional EVENT_WRITE packet

Add explicit EVENT_TYPE field

14 years agost/mesa: set MaxUniformComponents
Marek Olšák [Sun, 21 Nov 2010 21:02:02 +0000 (22:02 +0100)]
st/mesa: set MaxUniformComponents

Signed-off-by: Brian Paul <brianp@vmware.com>
14 years agoswrast: init alpha value to 1.0 in opt_sample_rgb_2d()
Brian Paul [Mon, 22 Nov 2010 16:04:13 +0000 (09:04 -0700)]
swrast: init alpha value to 1.0 in opt_sample_rgb_2d()

14 years agogallium: add PIPE_SHADER_CAP_SUBROUTINES
Marek Olšák [Sun, 14 Nov 2010 14:34:59 +0000 (15:34 +0100)]
gallium: add PIPE_SHADER_CAP_SUBROUTINES

This fixes piglit/glsl-vs-main-return and glsl-fs-main-return for the drivers
which don't support RET (i915g, r300g, r600g, svga).

ir_to_mesa does not currently generate subroutines, but it's a matter of time
till it's added. It would then break all the drivers which don't implement
them, so this CAP makes sense.

Signed-off-by: Marek Olšák <maraeo@gmail.com>
14 years agoMerge branch 'lp-offset-twoside'
Keith Whitwell [Mon, 22 Nov 2010 10:36:01 +0000 (10:36 +0000)]
Merge branch 'lp-offset-twoside'

14 years agor600g: pick correct color swap for A8 fbos.
Dave Airlie [Mon, 22 Nov 2010 06:03:00 +0000 (16:03 +1000)]
r600g: pick correct color swap for A8 fbos.

This fixes fdo bug 31810.

Signed-off-by: Dave Airlie <airlied@redhat.com>
14 years agor300/compiler: Add a more efficient version of rc_find_free_temporary()
Tom Stellard [Fri, 12 Nov 2010 08:59:13 +0000 (00:59 -0800)]
r300/compiler: Add a more efficient version of rc_find_free_temporary()

14 years agor300/compiler: Enable rename_reg pass for r500 cards
Tom Stellard [Thu, 11 Nov 2010 05:34:18 +0000 (21:34 -0800)]
r300/compiler: Enable rename_reg pass for r500 cards

In addition, the rename_reg pass has been rewritten to use
rc_get_readers().

14 years agor300/compiler: Use presubtract operations as much as possible
Tom Stellard [Tue, 9 Nov 2010 02:49:44 +0000 (18:49 -0800)]
r300/compiler: Use presubtract operations as much as possible

Previously, presubtract operations where only being used by instructions
with less than three source source registers.

14 years agor300/compiler: Convert RGB to alpha in the scheduler
Tom Stellard [Thu, 30 Sep 2010 06:52:49 +0000 (23:52 -0700)]
r300/compiler: Convert RGB to alpha in the scheduler

14 years agor300/compiler: Track readers through branches in rc_get_readers()
Tom Stellard [Sat, 6 Nov 2010 18:30:27 +0000 (11:30 -0700)]
r300/compiler: Track readers through branches in rc_get_readers()

14 years agor300/compiler: Handle BREAK and CONTINUE in rc_get_readers()
Tom Stellard [Sun, 14 Nov 2010 01:00:45 +0000 (17:00 -0800)]
r300/compiler: Handle BREAK and CONTINUE in rc_get_readers()

14 years agor300/compiler: Add rc_get_readers()
Tom Stellard [Sat, 30 Oct 2010 05:27:04 +0000 (22:27 -0700)]
r300/compiler: Add rc_get_readers()

14 years agor300/compiler: Ignore alpha dest register when replicating the result
Tom Stellard [Thu, 11 Nov 2010 09:13:01 +0000 (01:13 -0800)]
r300/compiler: Ignore alpha dest register when replicating the result

When the result of the alpha instruction is being replicated to the RGB
destination register, we do not need to use alpha's destination register.
This fixes an invalid "Too many hardware temporaries used" error in
the case where a transcendent operation writes to a temporary register
greater than max_temp_regs.

NOTE: This is a candidate for the 7.9 branch.

14 years agor300/compiler: Use zero as the register index for unused sources
Tom Stellard [Thu, 11 Nov 2010 09:01:13 +0000 (01:01 -0800)]
r300/compiler: Use zero as the register index for unused sources

This fixes an invalid "Too many hardware temporaries used" error in the
case where a source reads from a temporary register with an index greater
than max_temp_regs and then the source is marked as unused before the
register allocation pass.

NOTE: This is a candidate for the 7.9 branch.

14 years agor300/compiler: Fix instruction scheduling within IF blocks
Tom Stellard [Sun, 14 Nov 2010 01:12:58 +0000 (17:12 -0800)]
r300/compiler: Fix instruction scheduling within IF blocks

Reads of registers that where not written to within the same block were
not being tracked.  So in a situations like this:
0: IF
1: ADD t0, t1, t2
2: MOV t2, t1

Instruction 2 didn't know that instruction 1 read from t2, so
in some cases instruction 2 was being scheduled before instruction 1.

NOTE: This is a candidate for the 7.9 branch.

14 years agor300/compiler: Fix register allocator's handling of loops
Tom Stellard [Sun, 14 Nov 2010 00:57:06 +0000 (16:57 -0800)]
r300/compiler: Fix register allocator's handling of loops

NOTE: This is a candidate for the 7.9 branch.

14 years agor300/compiler: Make sure presubtract sources use supported swizzles
Tom Stellard [Thu, 11 Nov 2010 09:28:44 +0000 (01:28 -0800)]
r300/compiler: Make sure presubtract sources use supported swizzles

NOTE: This is a candidate for the 7.9 branch.

14 years agor600: Remove unnecessary header.
Vinson Lee [Sun, 21 Nov 2010 23:02:51 +0000 (15:02 -0800)]
r600: Remove unnecessary header.

14 years agodocs: add GL 4.1 status
Marek Olšák [Sun, 21 Nov 2010 21:54:33 +0000 (22:54 +0100)]
docs: add GL 4.1 status

14 years agost/mesa: enable ARB_explicit_attrib_location and EXT_separate_shader_objects
Marek Olšák [Sun, 21 Nov 2010 00:53:05 +0000 (01:53 +0100)]
st/mesa: enable ARB_explicit_attrib_location and EXT_separate_shader_objects

Gallium drivers pass all piglit tests for the two (there are 12 tests
for separate_shader_objects and 5 tests for explicit_attrib_location),
and I was told the extensions don't need any driver-specific code.

I made them dependent on PIPE_CAP_GLSL.

Signed-off-by: Brian Paul <brianp@vmware.com>
14 years agomesa: fix get_texture_dimensions() for texture array targets
Brian Paul [Sun, 21 Nov 2010 17:05:47 +0000 (10:05 -0700)]
mesa: fix get_texture_dimensions() for texture array targets

Fixes http://bugs.freedesktop.org/show_bug.cgi?id=31779

14 years agodocs: update some GL 3.0 status
Brian Paul [Sun, 21 Nov 2010 16:31:19 +0000 (09:31 -0700)]
docs: update some GL 3.0 status

14 years agomesa: hook up GL 3.x entrypoints
Brian Paul [Sun, 21 Nov 2010 16:19:23 +0000 (09:19 -0700)]
mesa: hook up GL 3.x entrypoints

Fix up some details in the xml files and regenerate dispatch files.

14 years agoglapi: rename GL3.xml to GL3x.xml as it covers all GL 3.x versions
Brian Paul [Sun, 21 Nov 2010 16:13:59 +0000 (09:13 -0700)]
glapi: rename GL3.xml to GL3x.xml as it covers all GL 3.x versions

14 years agomesa: fix error msg typo
Brian Paul [Fri, 19 Nov 2010 21:34:07 +0000 (14:34 -0700)]
mesa: fix error msg typo

14 years agoi915g: kill idws->pool
Daniel Vetter [Fri, 19 Nov 2010 22:38:23 +0000 (23:38 +0100)]
i915g: kill idws->pool

The drm winsys only ever handles one gem memory manager. Rip out
the unnecessary complication.

Reviewed-by: Jakob Bornecrantz <wallbraker@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jakob Bornecrantz <wallbraker@gmail.com>
14 years agoi915g: kill buf->map_gtt
Daniel Vetter [Fri, 19 Nov 2010 22:38:22 +0000 (23:38 +0100)]
i915g: kill buf->map_gtt

Not using the gtt is considered harmful for performance. And for
partial uploads there's always drm_intel_bo_subdata.

Reviewed-by: Jakob Bornecrantz <wallbraker@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jakob Bornecrantz <wallbraker@gmail.com>
14 years agoi915g: kill RGBA/X formats
Daniel Vetter [Fri, 19 Nov 2010 22:38:21 +0000 (23:38 +0100)]
i915g: kill RGBA/X formats

It's intel, so always little endian!

Reviewed-by: Jakob Bornecrantz <wallbraker@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jakob Bornecrantz <wallbraker@gmail.com>
14 years agoi915g: add pineview pci ids
Daniel Vetter [Fri, 19 Nov 2010 22:38:20 +0000 (23:38 +0100)]
i915g: add pineview pci ids

Reviewed-by: Jakob Bornecrantz <wallbraker@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jakob Bornecrantz <wallbraker@gmail.com>
14 years agoi915g: s/hw_tiled/tiling
Daniel Vetter [Fri, 19 Nov 2010 22:38:19 +0000 (23:38 +0100)]
i915g: s/hw_tiled/tiling

More in line with other intel drivers.

Change to use enum by Jakob Bornecrantz.

Reviewed-by: Jakob Bornecrantz <wallbraker@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jakob Bornecrantz <wallbraker@gmail.com>
14 years agoi915g: rip out ->sw_tiled
Daniel Vetter [Fri, 19 Nov 2010 22:38:18 +0000 (23:38 +0100)]
i915g: rip out ->sw_tiled

It looks like this was meant to facilitate unfenced access to textures/
color/renderbuffers. It's totally incomplete and fundamentally broken
on a few levels:
- broken: The kernel needs to about every tiled bo to fix up bit17
  swizzling on swap-in.
- unflexible: fenced/unfenced relocs from execbuffer2 do the same, much
  simpler.
- unneeded: with relaxed fencing tiled gem bos are as memory-efficient
  as this trick.

Hence kill it.

Reviewed-by: Jakob Bornecrantz <wallbraker@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Jakob Bornecrantz <wallbraker@gmail.com>
14 years agor300g: silence guard band cap errors
Joakim Sindholt [Sun, 21 Nov 2010 12:24:03 +0000 (13:24 +0100)]
r300g: silence guard band cap errors

Somebody should find out what these are. It can be found on Windows
getting a D3DCAPS9 from IDirect3D9::GetCaps() and reading the
GuardBand* values.

14 years agost/vega: Fix vgReadPixels with a subrectangle.
Chia-I Wu [Sun, 21 Nov 2010 10:58:47 +0000 (18:58 +0800)]
st/vega: Fix vgReadPixels with a subrectangle.

Fix a crash when the subrectangle is not inside the fb.  Fix wrong
pipe transfer when sx > 0 or sy + height != fb->height.

This fixes "readpixels" demo.

14 years agost/vega: Set wrap_r for mask and blend samplers.
Chia-I Wu [Sun, 21 Nov 2010 10:46:57 +0000 (18:46 +0800)]
st/vega: Set wrap_r for mask and blend samplers.

These two samplers use non-normalized texture coordinates.  wrap_r
cannot be PIPE_TEX_WRAP_REPEAT (the default).

This fixes

  sp_tex_sample.c:1790:get_linear_unorm_wrap: Assertion `0' failed

assertion failure.

14 years agost/vega: vegaLookupSingle should validate the state.
Chia-I Wu [Sun, 21 Nov 2010 10:44:21 +0000 (18:44 +0800)]
st/vega: vegaLookupSingle should validate the state.

Fix "lookup" demo crash.

14 years agotgsi: Add STENCIL to text parser.
Chia-I Wu [Sun, 21 Nov 2010 10:36:41 +0000 (18:36 +0800)]
tgsi: Add STENCIL to text parser.

Fix OpenVG "filter" demo

  Program received signal SIGSEGV, Segmentation fault.
  0xb7153dc9 in str_match_no_case (pcur=0xbfffe564, str=0x0) at
  tgsi/tgsi_text.c:86
  86         while (*str != '\0' && *str == uprcase( *cur )) {

14 years agomesa: Clean up header file inclusion in stencil.h.
Vinson Lee [Sun, 21 Nov 2010 06:44:33 +0000 (22:44 -0800)]
mesa: Clean up header file inclusion in stencil.h.

14 years agomesa: Clean up header file inclusion in shared.h.
Vinson Lee [Sun, 21 Nov 2010 06:30:27 +0000 (22:30 -0800)]
mesa: Clean up header file inclusion in shared.h.

14 years agomesa: Clean up header file inclusion in shaderapi.h.
Vinson Lee [Sun, 21 Nov 2010 06:17:28 +0000 (22:17 -0800)]
mesa: Clean up header file inclusion in shaderapi.h.

14 years agomesa: Clean up header file inclusion in scissor.h.
Vinson Lee [Sun, 21 Nov 2010 06:01:30 +0000 (22:01 -0800)]
mesa: Clean up header file inclusion in scissor.h.

14 years agomesa: Clean up header file inclusion in renderbuffer.h.
Vinson Lee [Sun, 21 Nov 2010 05:32:07 +0000 (21:32 -0800)]
mesa: Clean up header file inclusion in renderbuffer.h.

14 years agomesa: Clean up header file inclusion in readpix.h.
Vinson Lee [Sun, 21 Nov 2010 05:23:35 +0000 (21:23 -0800)]
mesa: Clean up header file inclusion in readpix.h.

14 years agomesa: Clean up header file inclusion in rastpos.h.
Vinson Lee [Sun, 21 Nov 2010 05:14:06 +0000 (21:14 -0800)]
mesa: Clean up header file inclusion in rastpos.h.

14 years agomesa: Clean up header file inclusion in polygon.h.
Vinson Lee [Sun, 21 Nov 2010 05:06:09 +0000 (21:06 -0800)]
mesa: Clean up header file inclusion in polygon.h.

14 years agointel: Remove unnecessary header.
Vinson Lee [Sun, 21 Nov 2010 04:13:50 +0000 (20:13 -0800)]
intel: Remove unnecessary header.

14 years agor600: Remove unnecesary header.
Vinson Lee [Sun, 21 Nov 2010 03:04:30 +0000 (19:04 -0800)]
r600: Remove unnecesary header.

14 years agoswrast: Remove unnecessary header.
Vinson Lee [Sun, 21 Nov 2010 03:00:18 +0000 (19:00 -0800)]
swrast: Remove unnecessary header.

14 years agost/mesa: Remove unnecessary headers.
Vinson Lee [Sun, 21 Nov 2010 02:48:09 +0000 (18:48 -0800)]
st/mesa: Remove unnecessary headers.

14 years agoscons: Define IN_DRI_DRIVER.
Chia-I Wu [Sun, 21 Nov 2010 01:47:11 +0000 (17:47 -0800)]
scons: Define IN_DRI_DRIVER.

The define is required for DRI drivers.  It is not needed for
libgl-xlib, but the overhead it introduces should be minor.

14 years agonvfx: only expose one rt on nv30
Xavier Chantry [Sat, 20 Nov 2010 21:51:12 +0000 (22:51 +0100)]
nvfx: only expose one rt on nv30

We do not know how to use more, GL_ARB_draw_buffers is not exposed on blob.

14 years agor600g: Fix location for clip plane registers
Owen W. Taylor [Sat, 20 Nov 2010 17:18:56 +0000 (12:18 -0500)]
r600g: Fix location for clip plane registers

The stride between the different clip plane registers was incorrect.

https://bugs.freedesktop.org/show_bug.cgi?id=31788

agd5f: fix evergreen as well.

14 years agor300g: fix rendering with no vertex elements
Marek Olšák [Sun, 14 Nov 2010 17:57:14 +0000 (18:57 +0100)]
r300g: fix rendering with no vertex elements

Fixes glsl-vs-point-size, although I meant to fix glsl-novertexdata.
Since swrast fails glsl-novertexdata too, I guess it's a core issue.

14 years agoi965: Remove duplicate MRF writes in the FS backend.
Eric Anholt [Fri, 19 Nov 2010 07:57:05 +0000 (15:57 +0800)]
i965: Remove duplicate MRF writes in the FS backend.

This is quite common for multitexture sampling, and not only cuts down
on the second and later set of MOVs, but typically also allows
compute-to-MRF on the first set.

No statistically siginficant performance difference in nexuiz (n=3),
but it reduces instruction count in one of its shaders and seems like
a good idea.

14 years agoi965: Improve compute-to-mrf.
Eric Anholt [Thu, 18 Nov 2010 07:03:50 +0000 (15:03 +0800)]
i965: Improve compute-to-mrf.

We were skipping it if the instruction producing the value we were
going to compute-to-mrf used its result reg as a source reg.  This
meant that the typical "write interpolated color to fragment color" or
"texture from interpolated texcoord" shader didn't compute-to-MRF.
Just don't check for the interference cases until after we've checked
if this is the instruction we wanted to compute-to-MRF.

Improves nexuiz high-settings performance on my laptop 0.48% +- 0.08%
(n=3).

14 years agoir_to_mesa: Detect and emit MOV_SATs for saturate constructs.
Eric Anholt [Fri, 19 Nov 2010 10:50:05 +0000 (18:50 +0800)]
ir_to_mesa: Detect and emit MOV_SATs for saturate constructs.

The goal here is to avoid regressing performance on ir_to_mesa drivers
for fixed function fragment shaders requiring saturates.

14 years agoi965: Recognize saturates and turn them into a saturated mov.
Eric Anholt [Fri, 19 Nov 2010 02:36:06 +0000 (10:36 +0800)]
i965: Recognize saturates and turn them into a saturated mov.

On pre-gen6, this turns 4 instructions into 1.  We could still do
better by folding the saturate into the instruction generating the
value if nobody else uses it, but that should be a separate pass.

14 years agoglsl: Add a helper function for determining if an rvalue could be a saturate.
Eric Anholt [Fri, 19 Nov 2010 10:27:41 +0000 (18:27 +0800)]
glsl: Add a helper function for determining if an rvalue could be a saturate.

Hardware pretty commonly has saturate modifiers on instructions, and
this can be used in codegen to produce those, without everyone else
needing to understand clamping other than min and max.

14 years agoi965: Fold constants into the second arg of BRW_SEL as well.
Eric Anholt [Thu, 18 Nov 2010 03:34:54 +0000 (11:34 +0800)]
i965: Fold constants into the second arg of BRW_SEL as well.

This hits a common case with min/max operations.

14 years agoi965: Remove extra \n at the end of every instruction in INTEL_DEBUG=wm.
Eric Anholt [Thu, 18 Nov 2010 03:48:47 +0000 (11:48 +0800)]
i965: Remove extra \n at the end of every instruction in INTEL_DEBUG=wm.

14 years agoi965: Just use memset() to clear most members in FS constructors.
Eric Anholt [Fri, 19 Nov 2010 09:44:35 +0000 (17:44 +0800)]
i965: Just use memset() to clear most members in FS constructors.

This should make it a lot harder to forget to zero things.

14 years agoi965: Fix compute_to_mrf to not move a MRF write up into another live range.
Eric Anholt [Fri, 19 Nov 2010 05:53:28 +0000 (13:53 +0800)]
i965: Fix compute_to_mrf to not move a MRF write up into another live range.

Fixes glsl-fs-copy-propagation-texcoords-1.

14 years agomesa: Include C++ files in the makedepend of DRI drivers.
Eric Anholt [Fri, 19 Nov 2010 09:19:38 +0000 (17:19 +0800)]
mesa: Include C++ files in the makedepend of DRI drivers.

14 years agoglsl: Fix type of label 'default' in switch statement.
Vinson Lee [Sat, 20 Nov 2010 01:28:22 +0000 (17:28 -0800)]
glsl: Fix type of label 'default' in switch statement.

14 years agoglsl: Add lower_vector.cpp to SConscript.
Vinson Lee [Sat, 20 Nov 2010 01:22:23 +0000 (17:22 -0800)]
glsl: Add lower_vector.cpp to SConscript.

14 years agoglsl: Fix matrix constructors with vector parameters
Ian Romanick [Sat, 20 Nov 2010 01:16:12 +0000 (17:16 -0800)]
glsl: Fix matrix constructors with vector parameters

When the semantics of write masks in assignments were changed, this
code was not correctly updated.

Fixes piglit test glsl-mat-from-vec-ctor-01.

14 years agoglsl: Combine many instruction lowering passes into one.
Kenneth Graunke [Fri, 19 Nov 2010 01:54:07 +0000 (17:54 -0800)]
glsl: Combine many instruction lowering passes into one.

This should save on the overhead of tree-walking and provide a
convenient place to add more instruction lowering in the future.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
14 years agoglsl: Simplify a type check by using type->is_integer().
Kenneth Graunke [Fri, 19 Nov 2010 01:15:13 +0000 (17:15 -0800)]
glsl: Simplify a type check by using type->is_integer().

14 years agoglsl: Add ir_quadop_vector expression
Ian Romanick [Tue, 16 Nov 2010 20:01:42 +0000 (12:01 -0800)]
glsl: Add ir_quadop_vector expression

The vector operator collects 2, 3, or 4 scalar components into a
vector.  Doing this has several advantages.  First, it will make
ud-chain tracking for components of vectors much easier.  Second, a
later optimization pass could collect scalars into vectors to allow
generation of SWZ instructions (or similar as operands to other
instructions on R200 and i915).  It also enables an easy way to
generate IR for SWZ instructions in the ARB_vertex_program assembler.

14 years agoglsl: Add unary ir_expression constructor
Ian Romanick [Tue, 9 Nov 2010 22:19:10 +0000 (14:19 -0800)]
glsl: Add unary ir_expression constructor

14 years agoglsl: Add ir_rvalue::is_negative_one predicate
Ian Romanick [Tue, 16 Nov 2010 19:59:22 +0000 (11:59 -0800)]
glsl: Add ir_rvalue::is_negative_one predicate

14 years agoglsl: Eliminate assumptions about size of ir_expression::operands
Ian Romanick [Thu, 11 Nov 2010 00:33:10 +0000 (16:33 -0800)]
glsl: Eliminate assumptions about size of ir_expression::operands

This may grow in the near future.