mesa.git
11 years agomesa: remove __QUICKDRAW__ tests
Brian Paul [Tue, 5 Nov 2013 00:47:19 +0000 (17:47 -0700)]
mesa: remove __QUICKDRAW__ tests

Reviewed-by: Matt Turner <mattst88@gmail.com>
11 years agomesa: remove WGLAPI macro
Brian Paul [Tue, 5 Nov 2013 00:47:19 +0000 (17:47 -0700)]
mesa: remove WGLAPI macro

WGLAPI was defined in glheader.h but wasn't used anywhere.

Reviewed-by: Matt Turner <mattst88@gmail.com>
11 years agoi965: Expose brw_reg_from_fs_reg() to other files.
Kenneth Graunke [Fri, 1 Nov 2013 20:29:37 +0000 (13:29 -0700)]
i965: Expose brw_reg_from_fs_reg() to other files.

This will be useful for Broadwell code as well.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
11 years agoi965: Combine gen6_clip_state.c and gen7_clip_state.c.
Kenneth Graunke [Fri, 1 Nov 2013 23:21:01 +0000 (16:21 -0700)]
i965: Combine gen6_clip_state.c and gen7_clip_state.c.

The changes between Gen6-7 are minimal, and can easily be solved with
an extra generation check.  This cuts a lot of duplicated code.

It also helps prevent even more duplication for Broadwell.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agodri/nouveau: Fix nouveau_init_screen2 breakage.
Francisco Jerez [Mon, 4 Nov 2013 19:58:10 +0000 (11:58 -0800)]
dri/nouveau: Fix nouveau_init_screen2 breakage.

Fix incorrect init ordering in nouveau_init_screen2 caused by
083f66fdd6451648fe355b64b02b29a6a4389f0d.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=71172

11 years agoi965/gen7: Add instruction latency estimates for untyped atomics and reads.
Francisco Jerez [Fri, 1 Nov 2013 18:29:13 +0000 (11:29 -0700)]
i965/gen7: Add instruction latency estimates for untyped atomics and reads.

The latency information has been obtained empirically from
measurements taken on Haswell and Ivy Bridge.

Acked-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
11 years agoi965/gen7: Handle atomic instructions from the VEC4 back-end.
Francisco Jerez [Wed, 25 Sep 2013 23:31:35 +0000 (16:31 -0700)]
i965/gen7: Handle atomic instructions from the VEC4 back-end.

This can deal with all the 15 32-bit untyped atomic operations the
hardware supports, but only INC and PREDEC are going to be exposed
through the API for now.

v2: Represent atomics as GLSL intrinsics.  Add support for variably
    indexed atomic counter arrays.
v3: Add comment on why we don't need to assign uniform storage for
    atomic counters.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/gen7: Handle atomic instructions from the FS back-end.
Francisco Jerez [Wed, 25 Sep 2013 23:30:20 +0000 (16:30 -0700)]
i965/gen7: Handle atomic instructions from the FS back-end.

This can deal with all the 15 32-bit untyped atomic operations the
hardware supports, but only INC and PREDEC are going to be exposed
through the API for now.

v2: Represent atomics as GLSL intrinsics.  Add support for variably
    indexed atomic counter arrays.  Fix interaction with fragment
    discard.
v3: Add comment on why we don't need to assign uniform storage for
    atomic counters.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965: Add a 'has_side_effects' back-end instruction predicate.
Francisco Jerez [Sun, 20 Oct 2013 21:02:08 +0000 (14:02 -0700)]
i965: Add a 'has_side_effects' back-end instruction predicate.

This patch fixes the three dead code elimination passes and the
VEC4/FS instruction scheduling passes so they leave instructions with
side effects alone.

At some point it might be interesting to have the instruction
scheduler calculate the exact memory dependencies between atomic ops,
but they're rare enough that it seems unlikely that it will make any
practical difference.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoclover: Calculate optimal work group size when it's not specified by the user.
Francisco Jerez [Mon, 4 Nov 2013 19:26:13 +0000 (11:26 -0800)]
clover: Calculate optimal work group size when it's not specified by the user.

Inspired by a patch sent to the mailing list by Tom Stellard, but
using a different algorithm to calculate the optimal block size that
has been found to be considerably more effective.

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
11 years agoclover: Constify some command_queue arguments.
Francisco Jerez [Mon, 4 Nov 2013 19:24:10 +0000 (11:24 -0800)]
clover: Constify some command_queue arguments.

11 years agoclover: Workaround compiler bug present in GCC 4.7.0-4.7.2.
Francisco Jerez [Wed, 30 Oct 2013 18:11:06 +0000 (11:11 -0700)]
clover: Workaround compiler bug present in GCC 4.7.0-4.7.2.

Variadic template aliases make these versions of GCC very confused,
write down the full type spec instead.

11 years agost/xorg: handle updates to DamageUnregister API
Emil Velikov [Fri, 1 Nov 2013 16:44:10 +0000 (16:44 +0000)]
st/xorg: handle updates to DamageUnregister API

xserver 1.14.99.2 simplified the DamageUnregister API, by
dropping the drawable argument.
Follow xf86-video-intel and xf86-video-vmware approach and
handle the new API by checking XORG_VERSION_CURRENT.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=71110
Reported-by: Michał Górny <mgorny@gentoo.org>
Reported-by: Vinson Lee <vlee@freedesktop.org>
Tested-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Emil Velikov <emil.l.velikov@gmail.com>
11 years agomesa: remove Watcom C support
Brian Paul [Mon, 4 Nov 2013 14:33:41 +0000 (07:33 -0700)]
mesa: remove Watcom C support

Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agomesa: remove Centerline C support from gl.h
Brian Paul [Mon, 4 Nov 2013 14:26:54 +0000 (07:26 -0700)]
mesa: remove Centerline C support from gl.h

Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agomesa: remove BUILD_FOR_SNAP bits
Brian Paul [Mon, 4 Nov 2013 14:29:57 +0000 (07:29 -0700)]
mesa: remove BUILD_FOR_SNAP bits

Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agomesa: remove SciTech stuff from gl.h
Brian Paul [Mon, 4 Nov 2013 14:25:22 +0000 (07:25 -0700)]
mesa: remove SciTech stuff from gl.h

Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agor600g: properly unbind a DSA state being deleted in r600_delete_dsa_state
Marek Olšák [Sun, 3 Nov 2013 19:27:28 +0000 (20:27 +0100)]
r600g: properly unbind a DSA state being deleted in r600_delete_dsa_state

Tested-by: Christian König <christian.koenig@amd.com>
11 years agodocs/GL3: document radeonsi support, minor cleanup
Marek Olšák [Thu, 31 Oct 2013 14:49:36 +0000 (15:49 +0100)]
docs/GL3: document radeonsi support, minor cleanup

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
11 years agoradeonsi: implement ARB_vertex_type_2_10_10_10_rev
Marek Olšák [Thu, 31 Oct 2013 14:20:06 +0000 (15:20 +0100)]
radeonsi: implement ARB_vertex_type_2_10_10_10_rev

11 years agor600g,radeonsi: properly expose texture buffer formats
Marek Olšák [Thu, 31 Oct 2013 14:32:30 +0000 (15:32 +0100)]
r600g,radeonsi: properly expose texture buffer formats

This exposes GL_ARB_texture_buffer_object_rgb32.

11 years agoradeonsi: implement texture buffer objects
Marek Olšák [Thu, 31 Oct 2013 14:08:49 +0000 (15:08 +0100)]
radeonsi: implement texture buffer objects

GLSL 1.40 is done.

11 years agoradeonsi: report our border color behavior
Marek Olšák [Wed, 30 Oct 2013 20:44:07 +0000 (21:44 +0100)]
radeonsi: report our border color behavior

11 years agoradeonsi: bind a dummy constant buffer in place of NULL buffers
Marek Olšák [Wed, 30 Oct 2013 19:44:23 +0000 (20:44 +0100)]
radeonsi: bind a dummy constant buffer in place of NULL buffers

11 years agoradeonsi: implement uniform buffer objects
Marek Olšák [Fri, 25 Oct 2013 09:45:47 +0000 (11:45 +0200)]
radeonsi: implement uniform buffer objects

11 years agotgsi/scan: set maximum index for each constant buffer
Marek Olšák [Wed, 30 Oct 2013 13:24:27 +0000 (14:24 +0100)]
tgsi/scan: set maximum index for each constant buffer

11 years agoradeonsi: try to fix IA_MULTI_VGT_PARAM programming
Marek Olšák [Tue, 29 Oct 2013 23:36:58 +0000 (00:36 +0100)]
radeonsi: try to fix IA_MULTI_VGT_PARAM programming

This doesn't make any difference on Bonaire, but it might help on Hawaii.

11 years agowinsys/radeon: use type-3 NOPs for CS padding on CIK
Marek Olšák [Tue, 29 Oct 2013 23:22:01 +0000 (00:22 +0100)]
winsys/radeon: use type-3 NOPs for CS padding on CIK

The type-2 NOPs are said to be unstable. It doesn't make a difference here.

11 years agoclover: fix build with LLVM 3.4
Aaron Watry [Fri, 1 Nov 2013 15:25:43 +0000 (10:25 -0500)]
clover: fix build with LLVM 3.4

dso_list was added as an argument for createInternalizePass in 3.4, and then
it was removed again in the same llvm version.

Tested-by: Mike Lothian <mike@fireburn.co.uk>
Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
11 years agodraw: move type construction out of loop
Brian Paul [Fri, 1 Nov 2013 23:07:55 +0000 (17:07 -0600)]
draw: move type construction out of loop

We can create clip_ptr_type once instead of n times inside the loop.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
11 years agoi965: Add driconf option clamp_max_samples
Chad Versace [Sun, 3 Nov 2013 21:14:50 +0000 (13:14 -0800)]
i965: Add driconf option clamp_max_samples

The new option clamps GL_MAX_SAMPLES to a hardware-supported MSAA mode.
If negative, then no clamping occurs.

v2: (for Paul)
  - Add option to i965 only, not to all DRI drivers.
  - Do not realy on int->uint cast to convert negative
    values to large positive values. Explicitly check for
    clamp_max_samples < 0.
v3: (for Ken)
   - Don't allow clamp_max_samples to alter context version.
   - Use clearer for-loop and correct comment.
   - Rename variables.
v4: (for Ken)
   - Merge identical if-branches.

Reviewed-and-tested-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
11 years agoi965: Fix logic_op check.
Vinson Lee [Sun, 3 Nov 2013 22:43:53 +0000 (14:43 -0800)]
i965: Fix logic_op check.

Fixes "Macro compares unsigned to 0" defect reported by Coverity.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
11 years agoi915: Fix logic_op check.
Vinson Lee [Sun, 3 Nov 2013 22:42:18 +0000 (14:42 -0800)]
i915: Fix logic_op check.

Fixes "Macro compares unsigned to 0" defect reported by Coverity.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
11 years agoi965: Initialize vec4_visitor member variables.
Vinson Lee [Sat, 26 Oct 2013 07:10:25 +0000 (00:10 -0700)]
i965: Initialize vec4_visitor member variables.

Fixes "Uninitialized pointer field" defect reported by Coverity.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agogallium/targets: remove vdpau-softpipe
Marek Olšák [Sat, 2 Nov 2013 11:20:29 +0000 (12:20 +0100)]
gallium/targets: remove vdpau-softpipe

Reviewed-by: Christian König <christian.koenig@amd.com>
11 years agogallium/targets: remove xvmc-softpipe
Marek Olšák [Sat, 2 Nov 2013 11:18:44 +0000 (12:18 +0100)]
gallium/targets: remove xvmc-softpipe

Reviewed-by: Christian König <christian.koenig@amd.com>
11 years agogallium/targets: remove r300/vdpau
Marek Olšák [Sat, 2 Nov 2013 11:07:42 +0000 (12:07 +0100)]
gallium/targets: remove r300/vdpau

Reviewed-by: Christian König <christian.koenig@amd.com>
11 years agogallium/targets: remove r300/xvmc
Marek Olšák [Sat, 2 Nov 2013 11:03:42 +0000 (12:03 +0100)]
gallium/targets: remove r300/xvmc

Reviewed-by: Christian König <christian.koenig@amd.com>
11 years agogallium/targets: remove radeonsi/xorg
Marek Olšák [Fri, 1 Nov 2013 18:42:47 +0000 (19:42 +0100)]
gallium/targets: remove radeonsi/xorg

Reviewed-by: Christian König <christian.koenig@amd.com>
11 years agogallium/targets: remove r600/xorg
Marek Olšák [Fri, 1 Nov 2013 18:36:12 +0000 (19:36 +0100)]
gallium/targets: remove r600/xorg

Reviewed-by: Christian König <christian.koenig@amd.com>
11 years agofreedreno/a3xx/texture: min/max lod
Rob Clark [Fri, 1 Nov 2013 23:46:55 +0000 (19:46 -0400)]
freedreno/a3xx/texture: min/max lod

Signed-off-by: Rob Clark <robclark@freedesktop.org>
11 years agofreedreno/a3xx: update envytools headers
Rob Clark [Fri, 1 Nov 2013 23:45:02 +0000 (19:45 -0400)]
freedreno/a3xx: update envytools headers

Signed-off-by: Rob Clark <robclark@freedesktop.org>
11 years agofreedreno/a3xx: fix VS out / FS in linking
Rob Clark [Fri, 1 Nov 2013 14:11:27 +0000 (10:11 -0400)]
freedreno/a3xx: fix VS out / FS in linking

Actually link VS out / FS in based on semantic info, keeping in mind
that position/pointsize can also be an input to the FS.  This fixes a
few fragment shaders which were using gl_Position.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
11 years agofreedreno/a3xx: allow num_samplers != num_textures
Rob Clark [Fri, 1 Nov 2013 14:09:39 +0000 (10:09 -0400)]
freedreno/a3xx: allow num_samplers != num_textures

Signed-off-by: Rob Clark <robclark@freedesktop.org>
11 years agofreedreno/a3xx/compiler: highp frag shader
Rob Clark [Thu, 31 Oct 2013 13:59:49 +0000 (09:59 -0400)]
freedreno/a3xx/compiler: highp frag shader

Fixes use of full-precision in fragment shader (ie. don't clobber r0.x
since that can be used by future bary instructions for varying fetch).
And makes use of full-precision the default in fragment shader (but can
be overriden via FD_MESA_DEBUG=fraghalf).

Seems like half precision is often not enough for texture coordinates.
The blob compiler is clever enough to keep texture coords in full
precision registers while using half precision for everything else.  But
we aren't quite that clever yet, so better to default to full precision.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
11 years agofreedreno/a3xx/compiler: relative addressing fixes.
Rob Clark [Sun, 27 Oct 2013 14:19:58 +0000 (10:19 -0400)]
freedreno/a3xx/compiler: relative addressing fixes.

Handle some relative addressing constraints: cannot handle const or
relative in cat5 and src2 of cat3.

Signed-off-by: Rob Clark <robclark@freedesktop.org>
11 years agofreedreno: we do actually support sqrt
Rob Clark [Fri, 25 Oct 2013 15:48:24 +0000 (11:48 -0400)]
freedreno: we do actually support sqrt

Signed-off-by: Rob Clark <robclark@freedesktop.org>
11 years agoi965: Enable ARB_sample_shading on intel hardware >= gen6
Anuj Phogat [Fri, 30 Aug 2013 20:13:15 +0000 (13:13 -0700)]
i965: Enable ARB_sample_shading on intel hardware >= gen6

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Ken Graunke <kenneth@whitecape.org>
11 years agoi965/gen7: Enable the features required for GL_ARB_sample_shading
Anuj Phogat [Mon, 7 Oct 2013 19:45:44 +0000 (12:45 -0700)]
i965/gen7: Enable the features required for GL_ARB_sample_shading

- Enable GEN7_WM_MSDISPMODE_PERSAMPLE, GEN7_WM_POSOFFSET_SAMPLE,
  GEN7_WM_OMASK_TO_RENDER_TARGET as per extension's specification.
- Only enable one of GEN7_WM_8_DISPATCH_ENABLE or GEN7_WM_16_DISPATCH_ENABLE
  when GEN7_WM_MSDISPMODE_PERSAMPLE is enabled. Refer IVB PRM Vol. 2, Part 1,
  Page 288 for details.

V2:
    - Use shared function _mesa_get_min_invocations_per_fragment().
    - Use brw_wm_prog_data variables: uses_pos_offset, uses_omask.

V3:
    - Enable simd16 dispatch with per sample shading.
    - Make changes to give preference to 'simd16 only' mode over
      'simd8 only' mode in case of non 1x per sample shading.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/gen6: Enable the features required for GL_ARB_sample_shading
Anuj Phogat [Mon, 7 Oct 2013 19:05:56 +0000 (12:05 -0700)]
i965/gen6: Enable the features required for GL_ARB_sample_shading

- Enable GEN6_WM_MSDISPMODE_PERSAMPLE, GEN6_WM_POSOFFSET_SAMPLE,
  GEN6_WM_OMASK_TO_RENDER_TARGET as per extension's specification.
- Only enable one of GEN6_WM_8_DISPATCH_ENABLE or GEN6_WM_16_DISPATCH_ENABLE
  when GEN6_WM_MSDISPMODE_PERSAMPLE is enabled.
  Refer SNB PRM Vol. 2, Part 1, Page 279 for details.

V2:
    - Use shared function _mesa_get_min_invocations_per_fragment().
    - Use brw_wm_prog_data variables: uses_pos_offset, uses_omask.

V3:
    - Enable simd16 dispatch with per sample shading.
    - Make changes to give preference to 'simd16 only' mode over
      'simd8 only' mode in case of non 1x per sample shading.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965: Add FS backend for builtin gl_SampleMask[]
Anuj Phogat [Thu, 24 Oct 2013 23:21:13 +0000 (16:21 -0700)]
i965: Add FS backend for builtin gl_SampleMask[]

V2:
   - Update comments
   - Add a special backend instructions to compute sample_mask.
   - Add a new variable uses_omask in brw_wm_prog_data.

V3:
   - Make changes to support simd16 mode.
   - Delete redundant AND instruction and handle the register
     stride in FS backend instruction.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965: Add FS backend for builtin gl_SampleID
Anuj Phogat [Thu, 24 Oct 2013 23:17:08 +0000 (16:17 -0700)]
i965: Add FS backend for builtin gl_SampleID

V2:
   - Update comments
   - Add compute_sample_id variables in brw_wm_prog_key
   - Add a special backend instruction to compute sample_id.

V3:
   - Make changes to support simd16 mode.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965: Add FS backend for builtin gl_SamplePosition
Anuj Phogat [Thu, 24 Oct 2013 22:53:05 +0000 (15:53 -0700)]
i965: Add FS backend for builtin gl_SamplePosition

V2:
   - Update comments.
   - Add compute_pos_offset variable in brw_wm_prog_key.
   - Add variable uses_pos_offset in brw_wm_prog_data.

V3:
   - Make changes to support simd16 mode.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965: Don't do vector splitting for ir_var_system_value
Anuj Phogat [Tue, 22 Oct 2013 19:00:11 +0000 (12:00 -0700)]
i965: Don't do vector splitting for ir_var_system_value

This is required while adding builtin system value vec{2, 3, 4}
variables. For example:
(declare (sys) vec2 gl_SamplePosition)

Without this patch above glsl ir splits in to:
(declare (temporary) float gl_SamplePosition_x)
(declare (temporary) float gl_SamplePosition_y)

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agomesa: Add a helper function _mesa_get_min_invocations_per_fragment()
Anuj Phogat [Thu, 17 Oct 2013 00:22:18 +0000 (17:22 -0700)]
mesa: Add a helper function _mesa_get_min_invocations_per_fragment()

This function is used to test if we need to do per sample shading or
per fragment shading.

V2: Use MAX2() to make sure the function returns a number >= 1.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoglsl: Add new builtins required by GL_ARB_sample_shading
Anuj Phogat [Fri, 30 Aug 2013 20:10:54 +0000 (13:10 -0700)]
glsl: Add new builtins required by GL_ARB_sample_shading

New builtins added by GL_ARB_sample_shading:
in vec2 gl_SamplePosition
in int gl_SampleID
in int gl_NumSamples
out int gl_SampleMask[]

V2: - Use SWIZZLE_XXXX for STATE_NUM_SAMPLES.
    - Use "result.samplemask" in arb_output_attrib_string.
    - Add comment to explain the size of gl_SampleMask[] array.
    - Make gl_SampleID and gl_SamplePosition system values.
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
11 years agomesa: Pass number of samples as a program state variable
Anuj Phogat [Thu, 3 Oct 2013 01:02:20 +0000 (18:02 -0700)]
mesa: Pass number of samples as a program state variable

Number of samples will be required in fragment shader program by new
GLSL builtin uniform "gl_NumSamples".

V2: Use "state.numsamples" in place of "state.num.samples"
    Use _NEW_BUFFERS flag in place of _NEW_MULTISAMPLE

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <idr@freedesktop.org>
Reviewed-by: Ken Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agomesa: Add new functions and enums required by GL_ARB_sample_shading
Anuj Phogat [Fri, 30 Aug 2013 19:52:38 +0000 (12:52 -0700)]
mesa: Add new functions and enums required by GL_ARB_sample_shading

New functions added by GL_ARB_sample_shading:
glMinSampleShadingARB()

New enums:
GL_SAMPLE_SHADING_ARB
GL_MIN_SAMPLE_SHADING_VALUE_ARB

V2: Update comments.
    Create new GL4x.xml.
    Remove redundant code in get.c.
    Update the API_XML list in Makefile.am.
    Add extra_gl40_ARB_sample_shading predicate to get.c.

V3:
   Fix make check failure.
   Add checks for desktop GL.
   Use GLfloat in place of GLclampf in glMinSampleShading().
Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ken Graunke <kenneth@whitecape.org>
11 years agomesa: Add infrastructure for GL_ARB_sample_shading
Anuj Phogat [Fri, 30 Aug 2013 19:34:36 +0000 (12:34 -0700)]
mesa: Add infrastructure for GL_ARB_sample_shading

This patch implements the common support code required for the
GL_ARB_sample_shading extension.

V2: Move GL_ARB_sample_shading to ARB extension list.

Signed-off-by: Anuj Phogat <anuj.phogat@gmail.com>
Reviewed-by: Ian Romanick <idr@freedesktop.org>
Reviewed-by: Ken Graunke <kenneth@whitecape.org>
11 years agoi965/fs: Optimize saturating SEL.G(E) with imm val <= 0.0f.
Matt Turner [Mon, 28 Oct 2013 04:26:36 +0000 (21:26 -0700)]
i965/fs: Optimize saturating SEL.G(E) with imm val <= 0.0f.

Only one program's instruction count is changed, but a shader in Tropics
is also affected.

instructions in affected programs:     326 -> 320 (-1.84%)

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/fs: Optimize saturating SEL.L(E) with imm val >= 1.0.
Matt Turner [Mon, 28 Oct 2013 03:03:48 +0000 (20:03 -0700)]
i965/fs: Optimize saturating SEL.L(E) with imm val >= 1.0.

total instructions in shared programs: 1409124 -> 1406971 (-0.15%)
instructions in affected programs:     158376 -> 156223 (-1.36%)

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/fs: Optimize OR with identical sources into a MOV.
Matt Turner [Mon, 28 Oct 2013 02:34:48 +0000 (19:34 -0700)]
i965/fs: Optimize OR with identical sources into a MOV.

Helps a lot of Steam games.

total instructions in shared programs: 1409360 -> 1409124 (-0.02%)
instructions in affected programs:     20842 -> 20606 (-1.13%)

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoglsl: Add a CSE pass.
Eric Anholt [Thu, 17 Oct 2013 17:28:40 +0000 (10:28 -0700)]
glsl: Add a CSE pass.

This only operates on constant/uniform values for now, because otherwise I'd
have to deal with killing my available CSE entries when assignments happen,
and getting even this working in the tree ir was painful enough.

As is, it has the following effect in shader-db:

total instructions in shared programs: 1524077 -> 1521964 (-0.14%)
instructions in affected programs:     50629 -> 48516 (-4.17%)
GAINED:                                0
LOST:                                  0

And, for tropics, that accounts for most of the effect, the FPS
improvement is 11.67% +/- 0.72% (n=3).

v2: Use read_only field of the variable, manually check the lod_info union
    members, use get_num_operands(), rename cse_operands_visitor to
    is_cse_candidate_visitor, move all is-a-candidate logic to that
    function, and call it before checking for CSE on a given rvalue, more
    comments, use private keyword.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/vec4: Don't overwrite op[1] when doing a UBO load.
Eric Anholt [Thu, 31 Oct 2013 00:09:53 +0000 (17:09 -0700)]
i965/vec4: Don't overwrite op[1] when doing a UBO load.

Prior to the GLSL CSE pass, all of our testing happened to have a freshly
computed temporary in op[1], from the multiply by 16 to get a byte offset.
As of CSE you'll get var_refs of a reused value when you've got multiple
loads from the same offset.

Make a proper temporary for computing our temporary value, to avoid
shifting the value farther and farther down.  Avoids a regression in
gs-float-array-variable-index

Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agost/mesa: fix _mesa_init_transform_feedback_object() argument
Brian Paul [Fri, 1 Nov 2013 14:43:22 +0000 (08:43 -0600)]
st/mesa: fix _mesa_init_transform_feedback_object() argument

Need to pass a pointer of the base type, not the st type.
Fixes a compiler warning.

11 years agoi965: Fix brw_store_register_mem64 to stay within a single batch.
Kenneth Graunke [Wed, 30 Oct 2013 23:06:06 +0000 (16:06 -0700)]
i965: Fix brw_store_register_mem64 to stay within a single batch.

Previously, the write of each 32-bit half might land in separate batch
buffers, which is insane.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisf@ijw.co.nz>
11 years agodocs: List transfom_feedback{2,3,instanced} for i965 in release notes.
Kenneth Graunke [Thu, 31 Oct 2013 18:10:20 +0000 (11:10 -0700)]
docs: List transfom_feedback{2,3,instanced} for i965 in release notes.

11 years agoi965: Enable the ARB_transform_feedback_instanced extension on Gen7+.
Kenneth Graunke [Sat, 26 Oct 2013 20:27:18 +0000 (13:27 -0700)]
i965: Enable the ARB_transform_feedback_instanced extension on Gen7+.

This depends on ARB_transform_feedback2, so I've predicated it on the
ability to do register writes.

It also depends on ARB_transform_feedback3, which is the only reason we
couldn't expose it previously.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
11 years agoi965: Enable the ARB_transform_feedback3 extension on Gen7+.
Kenneth Graunke [Sat, 26 Oct 2013 20:20:02 +0000 (13:20 -0700)]
i965: Enable the ARB_transform_feedback3 extension on Gen7+.

This extension is written a bit strangely.  Although it introduces the
concept of multiple transform feedback streams, it doesn't actually
provide more than a single stream.

The ARB_gpu_shader5 extension is what introduces the ability to write to
streams other than stream #0 and increases the required number of streams.

Since we don't yet support ARB_gpu_shader5, we can safely enable
ARB_transform_feedback3 even though we only support a single stream.
This does provide some useful functionality: applications can now use
more than one interleaved transform feedback buffer.

v2: Only expose the extension if ARB_transform_feedback2 is also
    available, to avoid confusing applications (suggested by Ian).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
11 years agoi965: Add support for gl_SkipComponents[1234].
Kenneth Graunke [Sat, 26 Oct 2013 18:34:11 +0000 (11:34 -0700)]
i965: Add support for gl_SkipComponents[1234].

ARB_transform_feedback3 allows applications to insert blank space
between interleaved varyings by adding fake 1, 2, 3, or 4-component
varyings named gl_SkipComponents[1234].

Mesa's core data structures don't explicitly track these, instead simply
tracking the buffer offset for each real varying.  If there is padding
due to gl_SkipComponents, these will not be contiguous.

Our hardware takes the specification quite literally.  Instead of
specifying offsets for each varying, it assumes they're all contiguous
and requires you to program fake varyings for each "hole".

This patch adds support for emitting SO_DECL structures for these holes.
Although we've lost the information about exactly how the application
specified their padding (i.e. gl_SkipComponents2, gl_SkipComponents2
vs. a single gl_SkipComponents4), it shouldn't matter.  We just need to
emit the right amount of space.  This patch emits the minimal number of
hole SO_DECL structures.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
11 years agoi965: Explicitly maintain a count of SO_DECL structures emitted.
Kenneth Graunke [Sat, 26 Oct 2013 18:14:55 +0000 (11:14 -0700)]
i965: Explicitly maintain a count of SO_DECL structures emitted.

Currently, we emit one SO_DECL structure per output, so we use the index
in the Outputs[] array as the index into the so_decl[] array as well.

In order to support the fake "gl_SkipComponents[1234]" varyings from
ARB_transform_feedback3, we'll need to emit SO_DECLs to fill in the
holes between successive outputs.  This means we'll likely emit more
SO_DECLs than there are outputs, so we need to count it explicitly.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
11 years agoi965: Create a temporary for transform feedback output components.
Kenneth Graunke [Sat, 26 Oct 2013 18:12:48 +0000 (11:12 -0700)]
i965: Create a temporary for transform feedback output components.

This is a bit shorter.

v2: Mark the temporary const (requested by Ian).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
11 years agoi965: Enable ARB_transform_feedback2 on Gen7+ if register writes work.
Kenneth Graunke [Tue, 28 May 2013 03:09:56 +0000 (20:09 -0700)]
i965: Enable ARB_transform_feedback2 on Gen7+ if register writes work.

With Linux 3.12, register writes work on Ivybridge and Baytrail, but not
Haswell.  That will be fixed in a future kernel revision, at which point
this extension will automatically be enabled.

v2: Use I915_GEM_DOMAIN_INSTRUCTION for the register read, and also
    correctly set the writeable flag when mapping (caught by Eric).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Initialize batchbuffer and state modules before extensions.
Kenneth Graunke [Sat, 26 Oct 2013 05:44:19 +0000 (22:44 -0700)]
i965: Initialize batchbuffer and state modules before extensions.

We only want to enable ARB_transform_feedback2 if we can write to
registers from batchbuffers.  In order to test that, we need to be able
to submit batches.  And for batches to work, we need to program the
initial pipeline state (like PIPELINE_SELECT), which is done from
brw_state_init().

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Implement glDrawTransformFeedback().
Kenneth Graunke [Fri, 6 Sep 2013 23:59:31 +0000 (16:59 -0700)]
i965: Implement glDrawTransformFeedback().

Implementing the GetTransformFeedbackVertexCount() driver hook allows
the VBO module to call us with the right number of vertices.

The hardware doesn't directly count the number of vertices written by
SOL, so we instead use the SO_NUM_PRIMS_WRITTEN(n) counters and multiply
by the number of vertices per primitive.

Unfortunately, counting the number of primitives generated is tricky:
a program might pause a transform feedback operation, start a second one
with a different object, then switch back and resume.  Both transform
feedback operations share the SO_NUM_PRIMS_WRITTEN counters.

To work around this, we save the counter values at Begin, Pause, Resume,
and End.  This "bookends" each section where transform feedback is
active for the current object.  Adding up differences of pairs gives
us the number of primitives generated.  (This is similar to what we
do for occlusion queries on platforms without hardware contexts.)

v2: Fix missing parenthesis in assertion (caught by Eric Anholt).
v3: Reuse prim_count_bo rather than freeing it and immediately
    allocating a new one (suggested by Topi Pohjolainen).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Mark brw_draw_prims tfb_vertcount parameter as unused.
Kenneth Graunke [Mon, 27 May 2013 18:41:21 +0000 (11:41 -0700)]
i965: Mark brw_draw_prims tfb_vertcount parameter as unused.

Renaming it makes it obvious that it isn't used, and the assertion
verifies that the VBO module never passes us such an object.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agomesa: Add a new GetTransformFeedbackVertexCount() driver hook.
Kenneth Graunke [Mon, 27 May 2013 18:27:18 +0000 (11:27 -0700)]
mesa: Add a new GetTransformFeedbackVertexCount() driver hook.

DrawTransformFeedback() needs to obtain the number of vertices written
to a particular stream during the last Begin/EndTransformFeedback block.
The new driver hook returns exactly that information.

Gallium drivers already implement this by passing the transform feedback
object to the drawing function, counting the number of vertices written
on the GPU, and using draw indirect.  This is efficient, but doesn't
always work:

If vertex data comes from user arrays, then the VBO module needs to
know how many vertices to upload, so we need to synchronously count.
Gallium drivers are currently broken in this case.

It also doesn't work if primitive restart is done in software.  For
normal drawing, vbo_draw_arrays() performs software primitive restart,
splitting the draw call in two.  vbo_draw_transform_feedback() currently
doesn't because it has no idea how many vertices need to be drawn.

The new driver hook gives it that information, allowing us to reuse
the existing vbo_draw_arrays() code to do everything right.

On Intel hardware (at least Ivybridge), using the draw indirect approach
is difficult since the hardware counts primitives, rather than vertices,
which requires doing some simple math.  So we always use this hook.

Gallium drivers will likely want to use this hook in some cases, but
want to use the existing draw indirect approach where possible.  Hence,
I've added a flag to allow drivers to opt-in to this call.

v2: Make it possible to implement this hook but only use this path
    when necessary (suggested by Marek).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
11 years agoi965: Implement Pause/ResumeTransformfeedback driver hooks on Gen7+.
Kenneth Graunke [Mon, 27 May 2013 00:53:14 +0000 (17:53 -0700)]
i965: Implement Pause/ResumeTransformfeedback driver hooks on Gen7+.

The ARB_transform_feedback2 extension introduces the ability to pause
and resume transform feedback sessions.  Although only one can be active
at a time, it's possible to switch between multiple transform feedback
objects while paused.

In order to facilitate this, we need to save/restore the SO_WRITE_OFFSET
registers so that after resuming, the GPU continues writing where it
left off.

This functionality also exists in ES 3.0, but somehow we completely
forgot to implement it.

v2: Reduce alignment from 4096 to 64 (it seemed excessive).
v3: Use I915_GEM_DOMAIN_INSTRUCTION instead of RENDER, for consistency
    with other writes.  It shouldn't matter on IVB+.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Create a new brw_transform_feedback_object subclass.
Kenneth Graunke [Fri, 24 May 2013 08:29:14 +0000 (01:29 -0700)]
i965: Create a new brw_transform_feedback_object subclass.

This adds the basic driver hooks to allocate/free the brw variant.
It doesn't contain any additional information yet, but it will soon.

v2: Use the new _mesa_init_transform_feedback_object helper function
    (requested by Eric and Ian).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agost/mesa: Use the new _mesa_init_transform_feedback_object() helper.
Kenneth Graunke [Fri, 25 Oct 2013 22:58:46 +0000 (15:58 -0700)]
st/mesa: Use the new _mesa_init_transform_feedback_object() helper.

This picks up a missing obj->EverBound = GL_FALSE line, and will catch
any new fields that get added in the future.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agomesa: Separate transform feedback object initialization from allocation.
Kenneth Graunke [Fri, 25 Oct 2013 22:54:10 +0000 (15:54 -0700)]
mesa: Separate transform feedback object initialization from allocation.

Both Gallium and i965 subclass gl_transform_feedback_object, which
requires implementing a custom NewTransformFeedback hook.  Creating a
helper function to initialize the fields avoids code duplication and
divergence.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agovbo: fix MSVC double->float conversion warnings
Brian Paul [Tue, 29 Oct 2013 20:11:28 +0000 (14:11 -0600)]
vbo: fix MSVC double->float conversion warnings

11 years agoswrast: fix MSVC double->float conversion warnings
Brian Paul [Tue, 29 Oct 2013 20:11:18 +0000 (14:11 -0600)]
swrast: fix MSVC double->float conversion warnings

11 years agomesa: fix some MSVC signed/unsigned compiler warnings
Brian Paul [Tue, 29 Oct 2013 20:11:01 +0000 (14:11 -0600)]
mesa: fix some MSVC signed/unsigned compiler warnings

11 years agometa: fix assorted MSVC int/float conversion warnings
Brian Paul [Tue, 29 Oct 2013 20:10:11 +0000 (14:10 -0600)]
meta: fix assorted MSVC int/float conversion warnings

11 years agoglsl: fix MSVC int->bool conversion warning
Brian Paul [Tue, 29 Oct 2013 20:09:46 +0000 (14:09 -0600)]
glsl: fix MSVC int->bool conversion warning

11 years agost/draw: silence Mingw warning in pointer_to_offset()
Brian Paul [Tue, 29 Oct 2013 15:48:32 +0000 (09:48 -0600)]
st/draw: silence Mingw warning in pointer_to_offset()

Fixes "warning: cast from pointer to integer of different size" for
64-bit builds.

11 years agoi965/fs: Perform CSE on CMP(N) instructions.
Matt Turner [Sun, 20 Oct 2013 18:38:17 +0000 (11:38 -0700)]
i965/fs: Perform CSE on CMP(N) instructions.

Optimizes

      cmp.ge.f0(8)  null     g45<8,8,1>F  0F
(+f0) sel(8)        g50<1>F  g40<8,8,1>F  g10<8,8,1>F
      cmp.ge.f0(8)  null     g45<8,8,1>F  0F
(+f0) sel(8)        g51<1>F  g41<8,8,1>F  g11<8,8,1>F
      cmp.ge.f0(8)  null     g45<8,8,1>F  0F
(+f0) sel(8)        g52<1>F  g42<8,8,1>F  g12<8,8,1>F
      cmp.ge.f0(8)  null     g45<8,8,1>F  0F
(+f0) sel(8)        g53<1>F  g43<8,8,1>F  g13<8,8,1>F

into

      cmp.ge.f0(8)  null     g45<8,8,1>F  0F
(+f0) sel(8)        g50<1>F  g40<8,8,1>F  g10<8,8,1>F
(+f0) sel(8)        g51<1>F  g41<8,8,1>F  g11<8,8,1>F
(+f0) sel(8)        g52<1>F  g42<8,8,1>F  g12<8,8,1>F
(+f0) sel(8)        g53<1>F  g43<8,8,1>F  g13<8,8,1>F

total instructions in shared programs: 1644938 -> 1638181 (-0.41%)
instructions in affected programs:     574955 -> 568198 (-1.18%)

Two more 16-wide programs (in L4D2). Some large (-9%) decreases in
instruction count in some of Valve's Source Engine games. No
regressions.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/fs: Don't emit null MOVs in CSE.
Matt Turner [Tue, 22 Oct 2013 22:40:08 +0000 (15:40 -0700)]
i965/fs: Don't emit null MOVs in CSE.

We'd like to CSE some instructions, like CMP, that often have null
destinations. Instead of replacing them with MOVs to null, just don't
emit the MOV.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/fs: Use reads_flag and writes_flag methods in the scheduler.
Matt Turner [Tue, 22 Oct 2013 23:23:27 +0000 (16:23 -0700)]
i965/fs: Use reads_flag and writes_flag methods in the scheduler.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/fs: Add reads_flag() and writes_flag() to fs_inst.
Matt Turner [Sun, 20 Oct 2013 18:32:01 +0000 (11:32 -0700)]
i965/fs: Add reads_flag() and writes_flag() to fs_inst.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/fs: Add is_null() method to fs_reg.
Matt Turner [Tue, 22 Oct 2013 19:32:23 +0000 (12:32 -0700)]
i965/fs: Add is_null() method to fs_reg.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/fs: Use the gen7 scratch read opcode when possible.
Eric Anholt [Wed, 16 Oct 2013 18:51:22 +0000 (11:51 -0700)]
i965/fs: Use the gen7 scratch read opcode when possible.

This avoids a lot of message setup we had to do otherwise.  Improves
GLB2.7 performance with register spilling force enabled by 1.6442% +/-
0.553218% (n=4).

v2: Use BRW_PREDICATE_NONE, improve a comment (by Paul).

Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965: Merge together opcodes for SHADER_OPCODE_GEN4_SCRATCH_READ/WRITE
Eric Anholt [Wed, 16 Oct 2013 18:45:06 +0000 (11:45 -0700)]
i965: Merge together opcodes for SHADER_OPCODE_GEN4_SCRATCH_READ/WRITE

I'm going to be introducing gen7 variants, and the previous naming was
going to get confusing.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/fs: Fix register unspills from a reg_offset.
Eric Anholt [Wed, 16 Oct 2013 19:39:07 +0000 (12:39 -0700)]
i965/fs: Fix register unspills from a reg_offset.

We were clearing the reg_offset before trying to use it.  Oops.  Fixes
glsl-fs-texture2drect with the reg spilling debug enabled.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/fs: Fix register spilling for 16-wide.
Eric Anholt [Wed, 16 Oct 2013 19:16:51 +0000 (12:16 -0700)]
i965/fs: Fix register spilling for 16-wide.

Things blew up when I enabled the debug register spill code without
disabling 16-wide, so I decided to just fix 16-wide spilling.

We still don't generate 16-wide when register spilling happens as part of
allocation (since we expect it to be slower), but now we can experiment
with allowing it in some cases in the future.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/fs: Exit the compile if spilling would overwrite in-use MRFs.
Eric Anholt [Tue, 29 Oct 2013 19:46:18 +0000 (12:46 -0700)]
i965/fs: Exit the compile if spilling would overwrite in-use MRFs.

I believe this will never happen in SIMD8 mode, but it could for SIMD16
when we fix it.

v2: Fix off-by-one in my register counting comment (caught by Paul).

Reviewed-by: Paul Berry <stereotype441@gmail.com> (v1)
11 years agoi965/fs: Fix broken register spilling debug code.
Eric Anholt [Wed, 16 Oct 2013 19:02:41 +0000 (12:02 -0700)]
i965/fs: Fix broken register spilling debug code.

Now that reg spilling generates new vgrfs, we were looping forever if you
ever turned it on.

Instead, move the debug code into the register allocator right near where
we'd be doing spilling anyway, which should more accurately reflect how
register spilling occurs in the wild.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/fs: Split "find what MRFs were used" to a helper function.
Eric Anholt [Tue, 29 Oct 2013 19:18:10 +0000 (12:18 -0700)]
i965/fs: Split "find what MRFs were used" to a helper function.

I'm going to need to reuse this for fixing register spilling on SIMD16.
Note that BRW_MAX_MRF is 16, which is the same as BRW_MAX_GRF -
GEN7_MRF_HACK_START.

Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965/fs: Update an ancient, wrong comment about reg_offset.
Eric Anholt [Tue, 29 Oct 2013 08:06:09 +0000 (01:06 -0700)]
i965/fs: Update an ancient, wrong comment about reg_offset.

This hasn't been true since SIMD16 mode was added.

Reviewed-by: Paul Berry <stereotype441@gmail.com>