mesa.git
11 years agor600g: fix valgrind warning on Cayman
Marek Olšák [Fri, 5 Apr 2013 12:22:42 +0000 (14:22 +0200)]
r600g: fix valgrind warning on Cayman

Warning: "Conditional jump or move depends on uninitialised value(s)".

11 years agogallivm/tgsi: handle untyped moves
Zack Rusin [Tue, 9 Apr 2013 12:11:45 +0000 (05:11 -0700)]
gallivm/tgsi: handle untyped moves

both mov and ucmp can be used to move variables of any type.
correctly note that about ucmp in the tgsi_info and make
sure gallivm can handle that by correctly casting the untyped
moves.

Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
11 years agogallivm: fix loops and conditionals within GS
Zack Rusin [Tue, 9 Apr 2013 12:01:54 +0000 (05:01 -0700)]
gallivm: fix loops and conditionals within GS

We were using simple temporaries, without using alloca or phi
nodes which meant that on every iteration of the loop our
temporaries, which were holding the number of vertices and
primitives which were emitted, were being reset to zero. Now
we're using alloca to allocate those variables to preserve
them across conditionals.

Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
11 years agollvmpipe: implement PIPE_QUERY_SO_STATISTICS
Zack Rusin [Tue, 9 Apr 2013 10:50:32 +0000 (03:50 -0700)]
llvmpipe: implement PIPE_QUERY_SO_STATISTICS

We were missing the implementation of PIPE_QUERY_SO_STATISTICS
query, this change implements it on top of the existing
facilities.

Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
11 years agogallivm: fix unsigned divide and remainder opcodes
Zack Rusin [Tue, 9 Apr 2013 10:28:48 +0000 (03:28 -0700)]
gallivm: fix unsigned divide and remainder opcodes

We want to both make sure we never divide by zero to not generate
sigfpe and that divide by zero is guaranteed to return 0xffffffff.
Based on José idea.

Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
11 years agogallivm: fix breakc
Zack Rusin [Thu, 4 Apr 2013 22:06:14 +0000 (15:06 -0700)]
gallivm: fix breakc

we break when the mask values are 0 not, 1, plus it's bit comparison
not a floating point comparison. This fixes both.

Signed-off-by: Zack Rusin <zackr@vmware.com>
Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
11 years agointel/hsw: Enable hiz (v2)
Chad Versace [Fri, 5 Apr 2013 23:35:47 +0000 (16:35 -0700)]
intel/hsw: Enable hiz (v2)

Enable hiz by setting intel_context::has_hiz.  However, to work around
a hardware bug, we selectively enable hiz for only nicely aligned miptree
slices.

No Piglit regressions on Haswell 0x0d26 rev07 when based atop
mesa-master-4ad3601.

Improves the performance of GLB27_TRex_C24Z16_FixedTimeStep by 18.52%
(hsw-0x0d26-rev07; kernel-3.9.0-rc1; GLBenchmark 2.7.0 Release a68901;
samples=3).

v2: Replace the check for IS_HASWELL(devid) in intel_miptree_slice_has_hiz()
    with a conditional set of has_hiz. [for anholt]

Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
11 years agoi965: Remove brw_context::depthstencil::hiz_mt
Chad Versace [Fri, 5 Apr 2013 21:51:31 +0000 (14:51 -0700)]
i965: Remove brw_context::depthstencil::hiz_mt

After recent refactorings, the field is written but no longer read.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
11 years agointel: Replace checks for hiz_mt with intel_has*hiz()
Chad Versace [Mon, 8 Apr 2013 20:45:49 +0000 (13:45 -0700)]
intel: Replace checks for hiz_mt with intel_has*hiz()

When appropriate, replace each check `hiz_mt != NULL` with either a call
to intel_miptree_slice_has_hiz() or intel_renderbuffer_has_hiz().  No
behavioral change.

This prepares for selectively enabling hiz on individual miptree slices
for Haswell.

This refactoring had several side effects.

  1. To prevent new warnings about discarding the const qualifier,
     I removed 'const' from some variable declarations in
     intel_validate_framebuffer().  The alternative was to add const
     qualifiers to multiple function signatures in the
     intel_renderbuffer_has_hiz call graph. Since the dominant convention
     in the Intel code is to not qualify function parameters as const,
     I chose to remove rather than add const qualifiers.

  2. I changed the signature of brw_emit_depth_stencil_hiz() by replacing
     `struct intel_mipmap_tree *hiz_mt` with `bool hiz`. The function used
     hiz_mt mostly as a boolean indicator of the presence of hiz, so the
     signature change is consistent with the patch's goal.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
11 years agoi965: Change signature of brw_get_depthstencil_tile_masks()
Chad Versace [Fri, 5 Apr 2013 21:29:53 +0000 (14:29 -0700)]
i965: Change signature of brw_get_depthstencil_tile_masks()

Add new parameters `depth_level` and `depth_layer`, which specify depth
miptree's slice of interest.  A following patch will pass the new
parameters through to intel_miptree_slice_has_hiz().

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
11 years agoi965/blorp: Add fields brw_blorp_mip_info::level,layer
Chad Versace [Wed, 20 Mar 2013 00:44:50 +0000 (17:44 -0700)]
i965/blorp: Add fields brw_blorp_mip_info::level,layer

The new fields define the 2D miptree slice to be used. A following patch
will pass the new fields through to intel_miptree_slice_has_hiz().

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
11 years agointel: Add field intel_mipmap_slice::has_hiz
Chad Versace [Thu, 21 Feb 2013 03:18:40 +0000 (19:18 -0800)]
intel: Add field intel_mipmap_slice::has_hiz

On Haswell, HiZ will selectively be enabled on individual miptree slices
to workaround a hardware bug. The new field 'has_hiz' indicates if HiZ is
enabled for a given slice.

Also add two new accessor functions for this field.
  intel_miptree_slice_has_hiz
  intel_renderbuffer_has_hiz

The new field and accessor functions are not yet used. Also, this patch
introduces no behavioral change because, in this patch,
intel_miptree_alloc_hiz() sets has_hiz for all slices.

Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
11 years agoi965/blorp: Align rectangle primitive for hiz ops
Chad Versace [Tue, 12 Mar 2013 02:21:46 +0000 (19:21 -0700)]
i965/blorp: Align rectangle primitive for hiz ops

The hardware docs and the simulator require that the rectangle primitive
emitted during fast depth clears and hiz resolves must be aligned to 8x4
pixels.

Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
11 years agoi965/vs: Use GRFs for pull constant offsets on gen7.
Eric Anholt [Thu, 4 Apr 2013 21:10:18 +0000 (14:10 -0700)]
i965/vs: Use GRFs for pull constant offsets on gen7.

This allows the computation of the offset to get written directly into the
message source.

shader-db results:
total instructions in shared programs: 3308390 -> 3283025 (-0.77%)
instructions in affected programs:     442998 -> 417633 (-5.73%)

No difference in GLB2.7 low res (n=9).

Reviewed-by: Matt Turner <mattst88@gmail.com>
11 years agoi965/vs: When asked to make a dst_reg for a src.xxxx, just write to src.x.
Eric Anholt [Thu, 4 Apr 2013 23:30:49 +0000 (16:30 -0700)]
i965/vs: When asked to make a dst_reg for a src.xxxx, just write to src.x.

We have several places in our pull constant handling where we make a
temporary src_reg for an int, and then turn it into a dst.  In doing so,
we were writing to the dst.xyzw, so we never register coalesced it with a
later mov from dst.x to real_dst.x.

These extra channels written would be removed if we had channel-wise DCE
in the backend, but we don't.  Fix it for now by just not writing these
extra channels that won't get used.

Reviewed-by: Matt Turner <mattst88@gmail.com>
11 years agoi965/gen6: Reduce updates of transform feedback offsets with HW contexts.
Eric Anholt [Tue, 9 Apr 2013 20:34:08 +0000 (13:34 -0700)]
i965/gen6: Reduce updates of transform feedback offsets with HW contexts.

The software-tracked transform feedback offsets (svbi_0_starting_index)
are incorrect in the presence of primitive restart, so we were actually
updating it with a bogus value if the batch wrapped and we emitted the
packet again during a single transform feedback.  By reducing state
emission, we avoid the bug.

Fixes piglit OpenGL 3.1/primitive-restart-xfb flush
Reviewed-by: Paul Berry <stereotype441@gmail.com>
NOTE: This is a candidate for the 9.1 branch.

11 years agoi965/gen7: Skip resetting SOL offsets at batch start with HW contexts.
Eric Anholt [Mon, 8 Apr 2013 17:56:23 +0000 (10:56 -0700)]
i965/gen7: Skip resetting SOL offsets at batch start with HW contexts.

The software-tracked transform feedback offsets (svbi_0_starting_index)
are incorrect in the presence of primitive restart, so we can't reliably
compute offsets for our buffer pointers after a batch flush.  Thanks to HW
contexts, our transform feedback offsets are now saved, so we can just
keep using the ones from before the batch wrap.

Fixes piglit OpenGL 3.1/primitive-restart-xfb flush
Reviewed-by: Paul Berry <stereotype441@gmail.com>
NOTE: This is a candidate for the 9.1 branch.

11 years agoradeonsi: remove sampler writemask v3
Christian König [Tue, 26 Mar 2013 14:09:27 +0000 (15:09 +0100)]
radeonsi: remove sampler writemask v3

v2: fix instrinsic name as well
v3: LLVM revision incremented as well

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com>
11 years agopipe-loader: Fix out of source build
Niels Ole Salscheider [Sun, 24 Feb 2013 22:00:04 +0000 (23:00 +0100)]
pipe-loader: Fix out of source build

Signed-off-by: Niels Ole Salscheider <niels_ole@salscheider-online.de>
11 years agost/mesa: remove #if FEATURE_GL/ES tests
Brian Paul [Wed, 10 Apr 2013 00:43:40 +0000 (18:43 -0600)]
st/mesa: remove #if FEATURE_GL/ES tests

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
11 years agomesa: remove old comment about FEATURE_GL
Brian Paul [Wed, 10 Apr 2013 00:43:40 +0000 (18:43 -0600)]
mesa: remove old comment about FEATURE_GL

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
11 years agomesa: remove #ifdef FEATURE_ES2, add some comments instead
Brian Paul [Wed, 10 Apr 2013 00:43:40 +0000 (18:43 -0600)]
mesa: remove #ifdef FEATURE_ES2, add some comments instead

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
11 years agost/mesa: remove #include mfeatures.h
Brian Paul [Wed, 10 Apr 2013 00:43:40 +0000 (18:43 -0600)]
st/mesa: remove #include mfeatures.h

None of these were needed.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
11 years agodocs: initial 9.2 release notes file
Brian Paul [Tue, 9 Apr 2013 14:52:30 +0000 (08:52 -0600)]
docs: initial 9.2 release notes file

11 years agost/osmesa: re-use buffers in OSMesaMakeCurrent()
Brian Paul [Tue, 9 Apr 2013 14:47:00 +0000 (08:47 -0600)]
st/osmesa: re-use buffers in OSMesaMakeCurrent()

Rather than creating a new buffer each time.  Fixes problems found
with vtk.

Tested-by: Kevin H. Hobbs <hobbsk@ohio.edu>
11 years agomesa: update derived framebuffer state in GetMultisamplefv
Marek Olšák [Sat, 6 Apr 2013 22:22:32 +0000 (00:22 +0200)]
mesa: update derived framebuffer state in GetMultisamplefv

This makes sure that ctx->DrawBuffer->Visual.samples is up-to-date.

Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agomesa: fix glGet queries depending on derived framebuffer state (v2)
Marek Olšák [Sat, 6 Apr 2013 22:19:06 +0000 (00:19 +0200)]
mesa: fix glGet queries depending on derived framebuffer state (v2)

"ctx->DrawBuffer->Visual" might be invalid if (NewState &_NEW_BUFFERS) != 0.

v2: also fix:
    - RGBA_INTEGER_MODE_EXT
    - RGBA_FLOAT_MODE_ARB (also check API support)
    - FRAMEBUFFER_SRGB_CAPABLE_EXT

NOTE: This is a candidate for stable branches.

Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965/gen7.5: Allow HW primitive restart for all primitive types.
Paul Berry [Sat, 6 Apr 2013 03:15:39 +0000 (20:15 -0700)]
i965/gen7.5: Allow HW primitive restart for all primitive types.

Gen7.5 (Haswell) hardware supports primitive restart for all primitive
types.  It also handles all possible primitive restart indices.
Rather than specialize both can_cut_index_handle_restart_index() and
the switch statement in can_cut_index_handle_prims() for Haswell, just
return early if the hardware is Haswell because we know it can handle
everything.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Only use brw_draw.c's trim() function when necessary.
Paul Berry [Fri, 5 Apr 2013 21:56:21 +0000 (14:56 -0700)]
i965: Only use brw_draw.c's trim() function when necessary.

brw_draw.c contains a trim() function which modifies the vertex count
for quads and quad strips in order to discard dangling vertices.  In
principle this shouldn't be necessary, since hardware since Gen4 is
capable of discarding dangling vertices by itself.  However, it's
necessary because as a hack to speed up rendering on Gen 4-5, we
sometimes convert quads to trifans and quad strips to tristrips.  The
trim() function isn't necessary on Gen6 and up.

This patch documents why and when the trim() function is necessary,
and avoids calling it when it's not needed.

This will avoid creating problems when we enable hardware support for
primitive restart of quads and quad strips on Haswell.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965/vs: Fix DEBUG_SHADER_TIME when VS terminates with 2 URB writes.
Paul Berry [Sun, 7 Apr 2013 13:29:46 +0000 (06:29 -0700)]
i965/vs: Fix DEBUG_SHADER_TIME when VS terminates with 2 URB writes.

The call to emit_shader_time_end() before the second URB write was
conditioned with "if (eot)", but eot is always false in this code
path, so emit_shader_time_end() was never being called for vertex
shaders that performed 2 URB writes.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
11 years agost/vdpau: fix subtitle related bug v2
Christian König [Tue, 9 Apr 2013 16:36:22 +0000 (18:36 +0200)]
st/vdpau: fix subtitle related bug v2

Drawing subtitles didn't increased the dirty area of the surface.

Reported and tested by freeedrich on irc.

v2: don't clear the surface

Signed-off-by: Christian König <christian.koenig@amd.com>
11 years agoglsl/linker: Reduce scope of non-flat integer varying fix.
Paul Berry [Tue, 9 Apr 2013 17:37:16 +0000 (10:37 -0700)]
glsl/linker: Reduce scope of non-flat integer varying fix.

In the mailing list discussion of "glsl/linker: fix varying packing
for non-flat integer varyings." (commit 7862bde), we concluded that
since the bug only applies to integral variables, it is safer to just
apply the bug fix to integer varyings.  I forgot to make the change
before pushing the patch upstream.  (Note: we aren't aware of any bugs
in commit 7862bde; it just seems wise to be on the safe side).

This patch makes the change.  Assuming commit 7862bde gets
cherry-picked back to 9.1, this commit should be cherry-picked too.

NOTE: This is a candidate for the 9.1 release branch.

11 years agoglsl/linker: Adapt flat varying handling in preparation for geometry shaders.
Paul Berry [Sat, 6 Apr 2013 17:50:46 +0000 (10:50 -0700)]
glsl/linker: Adapt flat varying handling in preparation for geometry shaders.

When a varying is consumed by transform feedback, but is not used by
the fragment shader, assign_varying_locations() sets its interpolation
type to "flat" in order to ensure that lower_packed_varyings never has
to deal with non-flat integral varyings (the GLSL spec doesn't require
integral vertex outputs to be flat if they aren't consumed by the
fragment shader).

A similar situation will arise when geometry shader support is added,
since the GLSL spec only requires integral vertex shader outputs to be
flat when they are consumed by the fragment shader.  This patch
modifies the linker to handle this situation too.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
11 years agoglsl: Document lower_packed_varyings' "flat" requirement with an assert.
Paul Berry [Sat, 6 Apr 2013 17:33:25 +0000 (10:33 -0700)]
glsl: Document lower_packed_varyings' "flat" requirement with an assert.

To minimize the variety of type conversions that lower_packed_varyings
needs to perform, it assumes that integral varyings are always
qualified as "flat".  link_varyings.cpp takes care of ensuring that
this is the case (even in the circumstances where GLSL doesn't require
it).

This patch documents the assumption with an assertion, for ease in
future debugging.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
11 years agoglsl/linker: fix varying packing for non-flat integer varyings.
Paul Berry [Sat, 6 Apr 2013 16:36:06 +0000 (09:36 -0700)]
glsl/linker: fix varying packing for non-flat integer varyings.

Commit dfb57e7 (glsl: Fix error checking on "flat" keyword to match
GLSL ES 3.00, GLSL 1.50) relaxed the rules for integral varyings: they
only need to be declared as "flat" if they are a fragment shader
inputs.  This allowed for the possibility of a vertex shader output
being a non-flat integer, provided that it was not matched to a
fragment shader input.  A non-contrived situation where this might
arise is if a vertex shader generates some integral outputs which are
consumed by tranform feedback, but not by the fragment shader.

Unfortunately, lower_packed_varyings assumes that *all* integral
varyings are flat, regardless of whether they are consumed by the
fragment shader.  As a result, attempting to create a non-flat
integral vertex output of a size that required packing (i.e. a size
other than ivec4 or uvec4) would cause an assertion failure in
lower_packed_varyings.

This patch prevents the assertion failure by forcing vertex shader
outputs to be "flat" whenever they are not consumed by the fragment
shader.  This should have no effect on rendering since the "flat"
keyword only affects the behaviour of fragment shader inputs.

Fixes piglit test "spec/EXT_transform_feedback/nonflat-integral".

NOTE: This is a candidate for the 9.1 release branch.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
11 years agoglsl: Check the size of ir_print_visitor's mode[] array with STATIC_ASSERT.
Paul Berry [Tue, 9 Apr 2013 17:03:11 +0000 (10:03 -0700)]
glsl: Check the size of ir_print_visitor's mode[] array with STATIC_ASSERT.

ir_print_visitor::visit(ir_variable *)'s mode[] array needs to match
the declaration of the enum ir_variable_mode.  It's hard to verify
that at compile time, but at least we can use a STATIC_ASSERT to make
sure it's the right size.

This required adding ir_var_mode_count to the enum.

11 years agoglsl: Fix ir_print_visitor's handling of interpolation qualifiers.
Paul Berry [Sun, 7 Apr 2013 02:16:58 +0000 (19:16 -0700)]
glsl: Fix ir_print_visitor's handling of interpolation qualifiers.

This patch updates the interp[] array to match the enum
glsl_interp_qualifier.

Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
v2: Add a STATIC_ASSERT to make sure the array is the correct size.
This required adding INTERP_QUALIFIER_COUNT to the enum.

11 years agoautotools: Better describe which cases OProfileJIT is required.
Johannes Obermayr [Tue, 9 Apr 2013 16:38:42 +0000 (17:38 +0100)]
autotools: Better describe which cases OProfileJIT is required.

Signed-off-by: José Fonseca <jfonseca@vmware.com>
11 years agosoftpipe: misc updates to image dumping in softpipe_flush()
Brian Paul [Tue, 9 Apr 2013 01:13:29 +0000 (19:13 -0600)]
softpipe: misc updates to image dumping in softpipe_flush()

11 years agotgsi: Ensure struct tgsi_ind_register field Index is initialized.
Vinson Lee [Sat, 6 Apr 2013 03:46:30 +0000 (20:46 -0700)]
tgsi: Ensure struct tgsi_ind_register field Index is initialized.

Fixes uninitialized scalar variable defect reported by Coverity.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Brian Paul <brianp@vmware.com>
11 years agor600g: Fix UMAD on Cayman
Martin Andersson [Tue, 2 Apr 2013 20:43:33 +0000 (22:43 +0200)]
r600g: Fix UMAD on Cayman

The multiplication part of tgsi_umad did not work on Cayman, because it did
not populate the correct vector slots.

This fixed hardlocks in the EXT_transform_feedback/order tests.

NOTE: This is a candidate for the stable branches.
(might not be easy to cherry-pick though)

Signed-off-by: Marek Olšák <maraeo@gmail.com>
11 years agointel: Remove the texture_tiling driconf option.
Kenneth Graunke [Tue, 9 Apr 2013 02:39:20 +0000 (19:39 -0700)]
intel: Remove the texture_tiling driconf option.

This option can force textures to be untiled.  However, on Gen6+, depth
buffers must be Y-tiled.  MSAA buffers also must be Y-tiled.  So setting
this option on even a trivial application like glxgears causes assertion
failures in a debug build, and likely GPU hangs in a release build.

It's just giving users a license to shoot themselves in the foot.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Prefer Y-tiling on Gen6+.
Kenneth Graunke [Tue, 9 Apr 2013 02:27:38 +0000 (19:27 -0700)]
i965: Prefer Y-tiling on Gen6+.

In the past, we preferred X-tiling for color buffers because our BLT
code couldn't handle Y-tiling.  However, the BLT paths have been largely
replaced by BLORP on Gen6+, which can handle any kind of tiling.

We hadn't measured any performance improvement in the past, but that's
probably because compressed textures were all untiled anyway.

Improves performance in GLB27_TRex_C24Z16_FixedTime by 7.69231%.

v2: Rebase on top of Eric's untiled-for-larger-than-aperture changes.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Use tiling even for compressed textures.
Kenneth Graunke [Tue, 9 Apr 2013 02:27:37 +0000 (19:27 -0700)]
i965: Use tiling even for compressed textures.

The code has no rationale for why we would force compressed textures to
be untiled, and it appears to work fine.  Git archeology indicates that
it's been that way dating back to when we first started tiling.

Improves performance in GLB27_TRex_C24Z16_FixedTimeStep at 1280x720 by
10.0529% +/- 0.573075% (n=12).  Improves performance in Xonotic by
4.56409% +/- 0.27965% (n=3).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agointel: Refactor selection of miptree tiling
Chad Versace [Tue, 9 Apr 2013 02:27:36 +0000 (19:27 -0700)]
intel: Refactor selection of miptree tiling

This patch (1) extracts from intel_miptree_create() the spaghetti logic
that selects the tiling format, (2) rewrites that spaghetti into a lucid
form, and (3) moves it to a new function, intel_miptree_choose_tiling().
No behavioral change.

As a bonus, it is now evident that the force_y_tiling parameter to
intel_miptree_create() does not really force Y tiling.

v2 (Ken): Rebase on top of Eric's untiled-for-larger-than-aperture
changes.  This required passing in the miptree.

Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agointel: Allocate hiz in intel_renderbuffer_move_to_temp()
Chad Versace [Fri, 5 Apr 2013 22:18:00 +0000 (15:18 -0700)]
intel: Allocate hiz in intel_renderbuffer_move_to_temp()

When moving the renderbuffer to a new miptree, we neglected to allocate
the hiz buffer for the new miptree. Oops.

Fixes all Piglit depthstencil-render-miplevels tests from crash to pass on
Sandybridge.

Note: This is a candidate for the 9.1 branch.
Reviewed-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
Signed-off-by: Chad Versace <chad.versace@linux.intel.com>
11 years agost/mesa: fix levels in initial texture creation
Dave Airlie [Sun, 7 Apr 2013 04:29:59 +0000 (14:29 +1000)]
st/mesa: fix levels in initial texture creation

calim pointed out we were getting mipmap levels for array multisamples,
this didn't make sense. So then I noticed this function takes last_level
so we are passing in a too high value here.

I think this should fix the case he was seeing.

Reviewed-by: Brian Paul <brianp@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
11 years agoglsl: Don't early-out for error-type inputs
Ian Romanick [Fri, 15 Mar 2013 22:23:19 +0000 (15:23 -0700)]
glsl: Don't early-out for error-type inputs

Check the type of the array operand and the index operand before doing
other checks.  This simplifies the code a bit now (eliminating the
error_emitted parameter), and enables some later functional changes.

The shader

uniform float x[6];
uniform sampler2D s;
void main() { gl_Position.x = xx[s + 1]; }

still generates (only) the two expected errors:

0:3(33): error: `xx' undeclared
0:3(39): error: Operands to arithmetic operators must be numeric

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
11 years agoglsl: Don't emit spurious errors for constant indexes of the wrong type
Ian Romanick [Fri, 15 Mar 2013 22:14:18 +0000 (15:14 -0700)]
glsl: Don't emit spurious errors for constant indexes of the wrong type

Previously the shader

uniform float x[6];
void main() { gl_Position.x = x[1.0]; }

would have generated the errors

0:2(33): error: array index must be integer type
0:2(36): error: array index must be < 6

Now only

0:2(33): error: array index must be integer type

will be generated.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
11 years agoglsl: Collect all of the non-constant index error checks together
Ian Romanick [Fri, 15 Mar 2013 22:10:35 +0000 (15:10 -0700)]
glsl: Collect all of the non-constant index error checks together

This puts all of the checks togeher for easier reading.  It also means
that all the checks are blocked on array->type->is_array.  Shortly this
will allow elimination of some is_error check work-arounds in this
function.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
11 years agoglsl: Minor code compaction in _mesa_ast_array_index_to_hir
Ian Romanick [Fri, 15 Mar 2013 22:09:48 +0000 (15:09 -0700)]
glsl: Minor code compaction in _mesa_ast_array_index_to_hir

Also, document the reason for not checking for type->is_array in some of
the bound-checking cases.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
11 years agoglsl: Don't return a value from check_builtin_array_max_size
Ian Romanick [Fri, 15 Mar 2013 21:33:01 +0000 (14:33 -0700)]
glsl: Don't return a value from check_builtin_array_max_size

That last consumer of the return value was changed to not use it by the
previous commit.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
11 years agoglsl: Remove some unnecessary uses of error_emitted
Ian Romanick [Fri, 15 Mar 2013 21:27:22 +0000 (14:27 -0700)]
glsl: Remove some unnecessary uses of error_emitted

The error_emitted flag is used in semantic checking to prevent spurious
cascading errors.  For example,

void foo(sampler2D s, float a)
{
    float x = a + (1.2 + s);

    ...
}

should only generate a single error.  Without the error_emitted flag for
the first error, "a + ..." would also generate an error.

However, a bunch of cases in _mesa_ast_array_index_to_hir that were
setting error_emitted would mask legitimate errors.  For example,

    vec4 a[7];
    float b = a[3.14];

should generate two error (float index and type mismatch in assignment).
The uses of error_emitted would cause only the first to be emitted.

This patch removes most of the places in _mesa_ast_array_index_to_hir
that would set the error_emitted flag.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
11 years agoglsl: Refactor handling of ast_array_index to a separate function
Ian Romanick [Fri, 15 Mar 2013 21:10:12 +0000 (14:10 -0700)]
glsl: Refactor handling of ast_array_index to a separate function

I love 800+ line switch-statements as much as the next guy... Future
commits will make changes to this part of the AST-to-HIR conversion, and
extracting this code will make that a bit easier.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
11 years agoglsl: Make check_build_array_max_size externally visible
Ian Romanick [Fri, 15 Mar 2013 21:09:00 +0000 (14:09 -0700)]
glsl: Make check_build_array_max_size externally visible

A future commit will try to use this function in a different file.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
11 years agointel: Avoid making tiled miptrees we won't be able to blit.
Eric Anholt [Wed, 3 Apr 2013 00:28:41 +0000 (17:28 -0700)]
intel: Avoid making tiled miptrees we won't be able to blit.

Doing so was breaking miptree mapping, which we really need to be able to
handle.  With this change, intel_miptree_map_direct() falls through to
doing a CPU mapping on the buffer like we need.

With the previous 2 patches, all of these should be fixed:
piglit max-texture-size (all 3 patches required!)
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=37871
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=44958
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=53494

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
11 years agointel: Do temporary CPU maps of textures that are too big to GTT map.
Eric Anholt [Wed, 3 Apr 2013 00:21:25 +0000 (17:21 -0700)]
intel: Do temporary CPU maps of textures that are too big to GTT map.

This still fails, since 8192*4bpp == 32768, which is too big to use the
blitter on.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
11 years agointel: Add support for writing to our linear-temporary-CPU-map case.
Eric Anholt [Wed, 3 Apr 2013 00:19:55 +0000 (17:19 -0700)]
intel: Add support for writing to our linear-temporary-CPU-map case.

This will be used for handling updates of large textures.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chad Versace <chad.versace@linux.intel.com>.
11 years agointel: Remove check for kernel 2.6.29.
Kenneth Graunke [Sat, 6 Apr 2013 07:08:37 +0000 (00:08 -0700)]
intel: Remove check for kernel 2.6.29.

Now that we require 2.6.39, there's no need to also check for 2.6.29.
Calling drm_intel_bufmgr_gem_enable_fenced_relocs() without checking
should be safe, as it simply sets a flag.

This does remove the check for zero fences available, but that doesn't
seem worth checking.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agointel: Require kernel 2.6.39 for relaxed relocation support.
Kenneth Graunke [Sat, 6 Apr 2013 06:59:52 +0000 (23:59 -0700)]
intel: Require kernel 2.6.39 for relaxed relocation support.

Chris Wilson's relaxed relocation patch landed in March 2011.  Anyone
running pre-3.0 kernels probably isn't going to get the latest Mesa
anyway.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Remove a few BRW_STATE_... enum values.
Kenneth Graunke [Sat, 6 Apr 2013 06:31:57 +0000 (23:31 -0700)]
i965: Remove a few BRW_STATE_... enum values.

These were likely used for BRW_NEW_... dirty bit flags at one point, but
they're unused now.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Remove brw->vb.info and struct brw_vertex_info.
Kenneth Graunke [Sat, 6 Apr 2013 05:58:39 +0000 (22:58 -0700)]
i965: Remove brw->vb.info and struct brw_vertex_info.

Nobody uses this value, so there's no need to set it.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Remove the BRW_NEW_INPUT_DIMENSIONS flag.
Kenneth Graunke [Sat, 6 Apr 2013 05:54:17 +0000 (22:54 -0700)]
i965: Remove the BRW_NEW_INPUT_DIMENSIONS flag.

When I removed the proj_attrib_mask optimization, I also removed the
last consumer of this bit without realizing it.

Since nobody uses it, there's no point in flagging it.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoregister_allocate: Fix the type of best_benefit.
Matt Turner [Tue, 2 Apr 2013 20:38:07 +0000 (13:38 -0700)]
register_allocate: Fix the type of best_benefit.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
11 years agoradeon/llvm: Bump minimum LLVM version to 3.3
Tom Stellard [Mon, 8 Apr 2013 14:43:34 +0000 (07:43 -0700)]
radeon/llvm: Bump minimum LLVM version to 3.3

11 years agoclover: Fix linkage of libOpenCL
Niels Ole Salscheider [Thu, 4 Apr 2013 21:26:45 +0000 (23:26 +0200)]
clover: Fix linkage of libOpenCL

Clover needs the irreader component of llvm

v2: Check for irreader component
irreader is only available with LLVM 3.3 >= 177971

Reviewed-by: Tom Stellard <thomas.stellard@amd.com>
Signed-off-by: Niels Ole Salscheider <niels_ole@salscheider-online.de>
11 years agor600g/llvm: Add support for native isa for pre EG
Vincent Lejeune [Sat, 6 Apr 2013 16:12:26 +0000 (18:12 +0200)]
r600g/llvm: Add support for native isa for pre EG

This fixes bug 62756 :
https://bugs.freedesktop.org/show_bug.cgi?id=62756#c12

11 years agogallium/util: add const to a parameter of util_max_layer
Marek Olšák [Fri, 5 Apr 2013 12:18:22 +0000 (14:18 +0200)]
gallium/util: add const to a parameter of util_max_layer

11 years agost/mesa: don't expose ARB_color_buffer_float without driver support in GL core
Marek Olšák [Thu, 28 Mar 2013 02:16:25 +0000 (03:16 +0100)]
st/mesa: don't expose ARB_color_buffer_float without driver support in GL core

Reviewed-by: Brian Paul <brianp@vmware.com>
11 years agomesa: allow drivers not to expose ARB_color_buffer_float in GL core profile
Marek Olšák [Thu, 28 Mar 2013 02:02:14 +0000 (03:02 +0100)]
mesa: allow drivers not to expose ARB_color_buffer_float in GL core profile

Reviewed-by: Brian Paul <brianp@vmware.com>
11 years agomesa: move updating clamp control derived state out of mesa_update_state_locked
Marek Olšák [Thu, 28 Mar 2013 01:48:17 +0000 (02:48 +0100)]
mesa: move updating clamp control derived state out of mesa_update_state_locked

It has 2 dependencies: glClampColor and the framebuffer, we might just as well
do the update where those two are changed.

v2: cosmetic changes from Brian's email

Reviewed-by: Brian Paul <brianp@vmware.com>
11 years agomesa: don't set _ClampFragmentColor to TRUE if it has no effect
Marek Olšák [Thu, 28 Mar 2013 00:56:01 +0000 (01:56 +0100)]
mesa: don't set _ClampFragmentColor to TRUE if it has no effect

This should reduce shader recompilations with drivers that emulate fragment
color clamping, because we want the clamping to be enabled only if there is
a signed normalized or floating-point colorbuffer.

Reviewed-by: Brian Paul <brianp@vmware.com>
11 years agomesa: refactor clamping controls, get rid of _ClampReadColor
Marek Olšák [Thu, 28 Mar 2013 00:50:21 +0000 (01:50 +0100)]
mesa: refactor clamping controls, get rid of _ClampReadColor

v2: cosmetic changes from Brian's email

Reviewed-by: Brian Paul <brianp@vmware.com>
11 years agomesa: don't memcmp() off the end of a cache key.
Chris Forbes [Sun, 31 Mar 2013 23:51:59 +0000 (12:51 +1300)]
mesa: don't memcmp() off the end of a cache key.

Reported-by: `per` in #intel-gfx
The size of the cache key varies, so store the actual size as well as
the key blob itself, rather than just assuming it's the same as the size
passed in.

NOTE: This is a candidate for stable branches.

V2: Don't leave silly holes in structure; use unsigned instead of GLuint.
V3: Fix missing case for `last` match.

Signed-off-by: Chris Forbes <chrisf@ijw.co.nz>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoradeonsi: Add compute support v3
Tom Stellard [Thu, 25 Oct 2012 17:50:10 +0000 (13:50 -0400)]
radeonsi: Add compute support v3

v2:
  - Only dump shaders when env variable is set.

v3:
  - Don't emit VGT registers

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com
11 years agoradeonsi: Set TCL1_ACTION_ENA when invalidating the texture cache
Tom Stellard [Wed, 13 Mar 2013 16:59:33 +0000 (12:59 -0400)]
radeonsi: Set TCL1_ACTION_ENA when invalidating the texture cache

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com
11 years agoradeonsi: Remove si_pm4_inval_vertex_cache()
Tom Stellard [Wed, 13 Mar 2013 17:01:32 +0000 (13:01 -0400)]
radeonsi: Remove si_pm4_inval_vertex_cache()

This function is a holdover from r600g and is identical to
si_pm4_inval_texture_cache(), so it is not needed.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Michel Dänzer <michel.daenzer@amd.com
11 years agogallium: PIPE_COMPUTE_CAP_IR_TARGET - allow drivers to specify a processor v2
Tom Stellard [Thu, 7 Mar 2013 15:51:25 +0000 (10:51 -0500)]
gallium: PIPE_COMPUTE_CAP_IR_TARGET - allow drivers to specify a processor v2

This target string now contains four values instead of three.  The old
processor field (which was really being interpreted as arch) has been split
into two fields: processor and arch.  This allows drivers to pass a
more a more detailed description of the hardware to compiler frontends.

v2:
  - Adapt to libclc changes

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
11 years agoutil: add ETC as compressed format
Wladimir [Fri, 5 Apr 2013 17:49:26 +0000 (19:49 +0200)]
util: add ETC as compressed format

Add UTIL_FORMAT_LAYOUT_ETC to util_format_is_compressed. It was missing.

Signed-off-by: Wladimir J. van der Laan <laanwj@gmail.com>
Reviewed-by: Brian Paul <brianp@vmware.com>
11 years agogallium/u_blitter: fix is_blit_generic_supported() stencil checking
Brian Paul [Fri, 5 Apr 2013 17:21:09 +0000 (11:21 -0600)]
gallium/u_blitter: fix is_blit_generic_supported() stencil checking

Don't check if there's sampler support for stencil if we're not
going to actually blit/copy stencil values.  Fixes the case where
we mistakenly said we can't support a blit of depth values from
S8Z24 to X8Z24.

Also, rename the is_stencil variable to dst_has_stencil to improve
readability.

NOTE: This is a candidate for the stable branches.

Reviewed-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: José Fonseca <jfonseca@vmware.com>
11 years agoHonor GLX_DONT_CARE in MATCH_MASK
Alexander Monakov [Mon, 1 Apr 2013 21:38:27 +0000 (01:38 +0400)]
Honor GLX_DONT_CARE in MATCH_MASK

NOTE: This is a candidate for stable branches.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=47478
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=62999
Bugzilla: http://bugs.winehq.org/show_bug.cgi?id=26763

11 years agofreedreno: use autogenerated register defs
Rob Clark [Fri, 5 Apr 2013 16:54:37 +0000 (12:54 -0400)]
freedreno: use autogenerated register defs

Switch to use the envytools generated headers for register/bitfield
definitions.  This is the first step in preparing to add a3xx support,
since it avoids having conflicting names for a3xx and a2xx registers.
And since I'm using envytools for a3xx it is simpler to just use it for
everything.

This shouldn't cause any functional change, it is really just a lot of
renaming.

Signed-off-by: Rob Clark <robdclark@gmail.com>
11 years agost/wgl: Install our windows message hook to threads created before the ICD is loaded.
José Fonseca [Thu, 4 Apr 2013 19:27:39 +0000 (20:27 +0100)]
st/wgl: Install our windows message hook to threads created before the ICD is loaded.

Otherwise we will not receive destroy windows events, causing framebuffers
to leak.

This happens particularly with java and jogl.

Tested with java + jogl, MATLAB.

VMware Internal Bug Number: 1013086.

Reviewed-by: Brian Paul <brianp@vmware.com>
11 years agollvmpipe: Work without sse2 if llvm is new enough
Adam Jackson [Thu, 4 Apr 2013 21:16:22 +0000 (17:16 -0400)]
llvmpipe: Work without sse2 if llvm is new enough

At least on llvm 3.2 this appears to work fine.  Tested on an Athlon XP
2600+, which has sse and 3dnow but not sse2.

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
Signed-off-by: Adam Jackson <ajax@redhat.com>
11 years agowinsys/radeon: add command stream replay dump for faulty lockup v3
Jerome Glisse [Wed, 27 Mar 2013 15:04:29 +0000 (11:04 -0400)]
winsys/radeon: add command stream replay dump for faulty lockup v3

Build time option, set RADEON_CS_DUMP_ON_LOCKUP to 1 in radeon_drm_cs.h to
enable it.

When enabled after each cs submission the code will try to detect lockup by
waiting on one of the buffer of the cs to become idle, after a timeout it
will consider that the cs triggered a lockup and will write a radeon_lockup.c
file in current directory that have all information for replaying the cs.

To build this file :
gcc -O0 -g radeon_lockup.c -ldrm -o radeon_lockup -I/usr/include/libdrm

v2: Add radeon_ctx.h file to mesa git tree
v3: Slightly improve dumped file for easier editing, only dump first faulty cs

Signed-off-by: Jerome Glisse <jglisse@redhat.com>
11 years agost/xlib: add HUD support for xlib/GLX
Brian Paul [Thu, 4 Apr 2013 20:06:51 +0000 (14:06 -0600)]
st/xlib: add HUD support for xlib/GLX

For the softpipe and llvmpipe drivers.

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
11 years agogallium/hud: add GALLIUM_HUD_PERIOD env var
Brian Paul [Thu, 4 Apr 2013 22:37:56 +0000 (16:37 -0600)]
gallium/hud: add GALLIUM_HUD_PERIOD env var

To set the graph update rate, in seconds.  The default update rate
has also been changed to 1/2 second.

Reviewed-by: Marek Olšák <maraeo@gmail.com>
11 years agogallium/hud: initialize sampler state
Brian Paul [Thu, 4 Apr 2013 22:24:40 +0000 (16:24 -0600)]
gallium/hud: initialize sampler state

The default wrap mode (PIPE_TEX_WRAP_REPEAT) is incompatible with
unnormalized texcoords (at least for softpipe).

v2: use PIPE_TEX_WRAP_CLAMP_TO_EDGE

Reviewed-by: Marek Olšák <maraeo@gmail.com>
11 years agoglsl: Add an optimization pass to flatten simple nested if blocks.
Kenneth Graunke [Thu, 4 Apr 2013 06:56:57 +0000 (23:56 -0700)]
glsl: Add an optimization pass to flatten simple nested if blocks.

GLBenchmark 2.7's shaders contain conditional blocks like:

if (x) {
    if (y) {
        ...
    }
}

where the outer conditional's then clause contains exactly one statement
(the nested if) and there are no else clauses.  This can easily be
optimized into:

if (x && y) {
    ...
}

This saves a few instructions in GLBenchmark 2.7:

    total instructions in shared programs: 11833 -> 11649 (-1.55%)
    instructions in affected programs:     8234 -> 8050 (-2.23%)

It also helps CS:GO slightly (-0.05%/-0.22%).  More importantly,
however, it simplifies the control flow graph, which could enable other
optimizations.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Use a variable for the push constant size in kB.
Kenneth Graunke [Wed, 3 Apr 2013 04:11:51 +0000 (21:11 -0700)]
i965: Use a variable for the push constant size in kB.

This clarifies that the offset of 2 is actually 16 kB / 8kB units.
It also keys both computations off of a single variable, which should
make it easier to change in the future.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965: Turn brw->urb.vs_size and gs_size into local variables.
Kenneth Graunke [Wed, 3 Apr 2013 04:11:50 +0000 (21:11 -0700)]
i965: Turn brw->urb.vs_size and gs_size into local variables.

These variables are only used within a single function, so we may as
well make them local variables.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965: Remove BRW_NEW_WM_INPUT_DIMENSIONS dirty bit.
Kenneth Graunke [Wed, 13 Mar 2013 05:16:37 +0000 (22:16 -0700)]
i965: Remove BRW_NEW_WM_INPUT_DIMENSIONS dirty bit.

This was only produced by the brw_wm_input_dimensions atom, which was
removed in the previous commit.  So there's no need for the dirty bit.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Delete brw_vs_constval.c and the brw_wm_input_sizes atom.
Kenneth Graunke [Wed, 13 Mar 2013 04:12:08 +0000 (21:12 -0700)]
i965: Delete brw_vs_constval.c and the brw_wm_input_sizes atom.

This was only used to compute proj_attrib_mask, which was removed by the
previous commit.  That makes this dead code.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Remove now dead brw_wm_prog_key::proj_attrib_mask field.
Kenneth Graunke [Wed, 13 Mar 2013 04:09:35 +0000 (21:09 -0700)]
i965: Remove now dead brw_wm_prog_key::proj_attrib_mask field.

The previous commit removed the last user of this field, so there's no
longer any point in setting it.  Removing this should eliminate
state-dependent recompiles, and make the precompile more reliable.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Remove fixed-function texture projection avoidance optimization.
Kenneth Graunke [Wed, 13 Mar 2013 04:09:19 +0000 (21:09 -0700)]
i965: Remove fixed-function texture projection avoidance optimization.

This optimization attempts to avoid extra attribute interpolation
instructions for texture coordinates where the W-component is 1.0.

Unfortunately, it requires a lot of complexity: the brw_wm_input_sizes
state atom (all the brw_vs_constval.c code) needs to run on each draw.
It computes the input_size_masks array, then uses that to compute
proj_attrib_mask.  Differences in proj_attrib_mask can cause
state-dependent fragment shader recompiles.  We also often fail to guess
proj_attrib_mask for the fragment shader precompile, causing us to
needlessly compile it twice.

Furthermore, this optimization only applies to fixed-function programs;
it does not help modern GLSL-based programs at all.  Generally, older
fixed-function programs run fine on modern hardware anyway.

The optimization has existed in some form since the initial commit.  When
we rewrote the fragment shader backend, we dropped it for a while.  Eric
readded it in commit eb30820f268608cf451da32de69723036dddbc62 as part of
an attempt to cure a ~1% performance regression caused by converting the
fixed-function fragment shader generation code from Mesa IR to GLSL IR.
However, no performance data was included in the commit message, so it's
unclear whether or not it was successful.

Time has passed, so I decided to re-measure this.  Surprisingly,
Eric's OpenArena timedemo actually runs /faster/ after removing this and
the brw_wm_input_sizes atom.  On Ivybridge at 1024x768, I measured a
1.39532% +/- 0.91833% increase in FPS (n = 55).  On Ironlake, there was
no statistically significant difference (n = 37).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Eric Anholt <eric@anholt.net>
11 years agoi965: Use ctx->Stencil._WriteEnabled in DEPTH_STENCIL_STATE.
Kenneth Graunke [Tue, 2 Apr 2013 17:28:07 +0000 (10:28 -0700)]
i965: Use ctx->Stencil._WriteEnabled in DEPTH_STENCIL_STATE.

This is the same computation as the _WriteEnabled flag, so we may as
well use it.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agoi965: Fix stencil write enable flag in 3DSTATE_DEPTH_BUFFER on Gen7+.
Kenneth Graunke [Tue, 2 Apr 2013 17:29:37 +0000 (10:29 -0700)]
i965: Fix stencil write enable flag in 3DSTATE_DEPTH_BUFFER on Gen7+.

ctx->Stencil.WriteMask is a statically sized array of 3 elements.
Checking it against 0 actually is a NULL check, and can never fail,
which meant that we always said stencil writes were enabled.

Use the new core Mesa derived state flag to fix this.

NOTE: This is a candidate for stable branches.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agomesa: Add new ctx->Stencil._WriteEnabled derived state flag.
Kenneth Graunke [Tue, 2 Apr 2013 17:22:18 +0000 (10:22 -0700)]
mesa: Add new ctx->Stencil._WriteEnabled derived state flag.

i965 needs to know whether stencil writes are enabled in several places,
and gets the test wrong sometimes.  While we could create a function to
compute this, it seems generally useful enough to warrant a new piece of
derived state.  Also, all the plumbing is already in place.

NOTE: This is a candidate for stable branches.
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Paul Berry <stereotype441@gmail.com>
11 years agogallivm: some minor cube map cleanup
Roland Scheidegger [Thu, 4 Apr 2013 21:20:49 +0000 (23:20 +0200)]
gallivm: some minor cube map cleanup

The ar_ge_as_at variable was just very very confusing since the condition
was actually the other way around (as_at_ge_ar). So change the condition
(and the selects depending on it) to match the variable name.
And also change the chosen major axis in case the coord values are the
same. OpenGL doesn't care one bit which one is chosen in this case but
it looks like dx10 would require z chosen over y, and y chosen over x
(previously did x chosen over y, y chosen over z). Since it's all the
same effort just honor dx10's wishes. (Though actually, for some prefered
orderings, we could save one (or two with derivatives) selects since the
tnewx and tnewz (and the corresponding dmax values) are the same.)

Reviewed-by: Jose Fonseca <jfonseca@vmware.com>
11 years agoi965: Ask the register allocator to round-robin through registers.
Eric Anholt [Sat, 1 Dec 2012 00:34:09 +0000 (16:34 -0800)]
i965: Ask the register allocator to round-robin through registers.

The way we were allocating registers before, packing into low register
numbers for Ironlake, resulted in an overly-constrained dependency graph
for instruction scheduling.  Improves GLBenchmark 2.1 performance by
4.5% +/- 0.7% (n=26).  No difference on my old GLSL demo (n=20).  No
difference on nexuiz (n=15).

v2: Fix off-by-one bug that made the change only work for 16-wide on i965.
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>