mesa.git
8 years agoi965/blorp/gen8: Stop multiplying depth by 6 for cubes
Jason Ekstrand [Sat, 23 Jul 2016 02:53:59 +0000 (19:53 -0700)]
i965/blorp/gen8: Stop multiplying depth by 6 for cubes

intel_mipmap_tree::logical_depth0 is now in 2-D slices so there is no need
for us to multiply by 6 when we go to fill out a blorp surface state.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
8 years agonvc0: use nvc0_m2mf_push_linear() to reduce code duplication
Samuel Pitoiset [Mon, 25 Jul 2016 22:46:47 +0000 (00:46 +0200)]
nvc0: use nvc0_m2mf_push_linear() to reduce code duplication

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
8 years agonvc0: use nve4_p2mf_push_linear() to reduce code duplication
Samuel Pitoiset [Mon, 25 Jul 2016 23:38:49 +0000 (01:38 +0200)]
nvc0: use nve4_p2mf_push_linear() to reduce code duplication

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
8 years agobuild: Remove unused AX_CHECK_COMPILE_FLAG macro
Andreas Boll [Wed, 20 Jul 2016 08:37:45 +0000 (10:37 +0200)]
build: Remove unused AX_CHECK_COMPILE_FLAG macro

Unused since 1a6ae840413d7fb6d2e83f6a83081d5246c7ac9e

Signed-off-by: Andreas Boll <andreas.boll.dev@gmail.com>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
8 years agomain: memcpy larger chunks in _mesa_propagate_uniforms_to_driver_storage
Nils Wallménius [Fri, 22 Jul 2016 11:10:03 +0000 (13:10 +0200)]
main: memcpy larger chunks in _mesa_propagate_uniforms_to_driver_storage

When possible, do the memcpy on larger blocks. This reduces cycles
spent in _mesa_propagate_uniforms_to_driver_storage from
1.51 % to 0.62% according to perf during the Unigine Heaven benchmark.
It did not affect the framerate of the benchmark. The system used for
testing was an i5 6600K with a Radeon R9 380.

Piglit hangs randomly on this system both with and without the patch
so i could not make a comparison.

v2: fixed whitespace

Signed-off-by: Nils Wallménius <nils.wallmenius@gmail.com>
Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
8 years agost/va: enable h264 VAAPI encode
Boyuan Zhang [Thu, 21 Jul 2016 23:40:22 +0000 (19:40 -0400)]
st/va: enable h264 VAAPI encode

Enable H.264 VAAPI encoding through config. Currently only H.264 baseline is supported. Encode entrypoint is not accepted by driver.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
8 years agost/va: add function to handle misc param type frame rate
Boyuan Zhang [Thu, 21 Jul 2016 23:40:21 +0000 (19:40 -0400)]
st/va: add function to handle misc param type frame rate

Frame rate can be passed to driver either through VAEncSequenceParameterBufferType or VAEncMiscParameterTypeFrameRate. Previous code only implement the former one, which is used by Gstreamer-Vaapi. Now adding implementation for VAEncMiscParameterTypeFrameRate. Also adding default frame rate as 30 just in case application never provides frame rate information to driver.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
8 years agost/va: add enviromental variable to disable interlace
Boyuan Zhang [Thu, 21 Jul 2016 23:40:20 +0000 (19:40 -0400)]
st/va: add enviromental variable to disable interlace

Add environmental variable to disable interlace mode. At VAAPI decoding stage, driver can not distinguish b/w pure decoding case and transcoding case. And since interlace encoding is not supported, we have to disable interlace for transcoding case. The temporary solution is to use enviromental variable to disable interlace mode.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
8 years agost/va: add preset values for VAAPI encode
Boyuan Zhang [Thu, 21 Jul 2016 23:40:19 +0000 (19:40 -0400)]
st/va: add preset values for VAAPI encode

Add some hardcoded values hardware needs mainly for rate control purpose. With previously hardcoded values for OMX, the rate control result is not correct. This change fixed the rate control result by setting correct values for Vaapi.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
8 years agost/va: add functions for VAAPI encode
Boyuan Zhang [Thu, 21 Jul 2016 23:40:18 +0000 (19:40 -0400)]
st/va: add functions for VAAPI encode

Add necessary functions/changes for VAAPI encoding to buffer and picture. These changes will allow driver to handle all Vaapi encode related operations. This patch doesn't change the Vaapi decode behaviour.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
8 years agost/va: get rate control method from configattrib v2
Boyuan Zhang [Thu, 21 Jul 2016 23:40:17 +0000 (19:40 -0400)]
st/va: get rate control method from configattrib v2

Rate control method is passed from app to driver through config attrib list.
That is why we need to store this rate control method to config. And later
on, we will pass this value to context->desc.h264enc.rate_ctrl.rate_ctrl_method.

v2 (chk): fix broken build and commit message

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
8 years agost/va: add conversion for yv12 to nv12in putimage v2
Boyuan Zhang [Thu, 21 Jul 2016 23:40:16 +0000 (19:40 -0400)]
st/va: add conversion for yv12 to nv12in putimage v2

For putimage call, if image format is yv12 (or IYUV with U V field swap) and
surface format is nv12, then we need to convert yv12 to nv12 and then copy
the converted data from image to surface. We can't use the existing logic
where surface is destroyed and re-created with yv12 format.

v2 (chk): fix some compiler warnings and commit message

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
8 years agovl/util: add copy func for yv12image to nv12surface v2
Boyuan Zhang [Thu, 21 Jul 2016 23:40:15 +0000 (19:40 -0400)]
vl/util: add copy func for yv12image to nv12surface v2

Add function to copy from yv12 image to nv12 surface for VAAPI putimage call.
We need this function in VaPutImage call where copying from yv12 image to nv12
surface for encoding. Existing function can't be used because it only work for
copying from yv12 surface to nv12 image in Vaapi.

v2: cleanup variable types and commit message

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
8 years agost/va: add encode entrypoint v2
Boyuan Zhang [Thu, 21 Jul 2016 23:40:14 +0000 (19:40 -0400)]
st/va: add encode entrypoint v2

VAAPI passes PIPE_VIDEO_ENTRYPOINT_ENCODE as entry point for encoding case. We
will save this encode entry point in config. config_id was used as profile
previously. Now, config has both profile and entrypoint field, and config_id is
used to get the config object. Later on, we pass this entrypoint to
context->templat.entrypoint instead of always hardcoded to
PIPE_VIDEO_ENTRYPOINT_BITSTREAM for decoding case previously. Encode entrypoint
is not accepted by driver until we enable Vaapi encode in later patch.

v2 (chk): fix commit message to match 80 chars, use switch instead of ifs,
  fix memory leaks in the error path, implement vlVaQueryConfigEntrypoints
  as well, drop VAEntrypointEncPicture (only used for JPEG).

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
8 years agonvc0: upload sample locations on GM20x
Samuel Pitoiset [Sun, 24 Jul 2016 22:02:12 +0000 (00:02 +0200)]
nvc0: upload sample locations on GM20x

This fixes a bunch of multisample piglit tests on GM206, like
bin/arb_texture_multisample-texelfetch 2 -auto -fbo

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
8 years agofreedreno/a4xx: time-elapsed query should be active for clears
Rob Clark [Sun, 24 Jul 2016 13:32:09 +0000 (09:32 -0400)]
freedreno/a4xx: time-elapsed query should be active for clears

Signed-off-by: Rob Clark <robdclark@gmail.com>
8 years agonvc0/ir: fix up an assertion in emitUADD()
Samuel Pitoiset [Sat, 23 Jul 2016 01:08:34 +0000 (03:08 +0200)]
nvc0/ir: fix up an assertion in emitUADD()

It's illegal to have neg modifiers on both sources for OP_ADD,
and it's illegal to have OP_SUB with just src0 neg.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
8 years agonvc0: fix wrong indentation in nvc0_validate_fb()
Samuel Pitoiset [Sat, 23 Jul 2016 22:28:11 +0000 (00:28 +0200)]
nvc0: fix wrong indentation in nvc0_validate_fb()

Trivial.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
8 years agoglsl: reuse main extension table to appropriately restrict extensions
Ilia Mirkin [Sun, 12 Jun 2016 22:56:43 +0000 (18:56 -0400)]
glsl: reuse main extension table to appropriately restrict extensions

Previously we were only restricting based on ES/non-ES-ness and whether
the overall enable bit had been flipped on. However we have been adding
more fine-grained restrictions, such as based on compat profiles, as
well as specific ES versions. Most of the time this doesn't matter, but
it can create awkward situations and duplication of logic.

Here we separate the main extension table into a separate object file,
linked to the glsl compiler, which makes use of it with a custom
function which takes the ES-ness of the shader into account (thus
allowing desktop shaders to properly use ES extensions that would
otherwise have been disallowed.) We can also now use this logic to
generate #define's for all supported extensions automatically, removing
the duplicate (and often inaccurate) list in glcpp.

The effect of this change should be nil in most cases. However in some
situations, extensions like GL_ARB_gpu_shader5 which were formerly
available in compat contexts on the GLSL side of things will now become
inaccessible.

This regresses two ES CTS tests:

  ES3-CTS.shaders.shader_integer_mix.define
  ES31-CTS.shader_integer_mix.define

however that is due to them using #version 100 instead of 300 es. As the
extension is only defined for ES3, I believe this is the correct
behavior.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com> (v2)
v2 -> v3: integrate glcpp defines into the same mechanism

8 years agofreedreno/a4xx: timestamp queries
Rob Clark [Sun, 14 Feb 2016 16:14:06 +0000 (11:14 -0500)]
freedreno/a4xx: timestamp queries

Signed-off-by: Rob Clark <robdclark@gmail.com>
8 years agofreedreno: hw timestamp support
Rob Clark [Tue, 23 Feb 2016 17:03:43 +0000 (12:03 -0500)]
freedreno: hw timestamp support

If the kernel supports it, use hw counter for timestamps.

Signed-off-by: Rob Clark <robdclark@gmail.com>
8 years agofreedreno: prep work for timestamp queries
Rob Clark [Tue, 1 Mar 2016 15:45:32 +0000 (10:45 -0500)]
freedreno: prep work for timestamp queries

We need "NULL" state to be a valid bit in the bitmask, because timestamp
queries are not restricted to draw/etc stages (ie. the only commands to
submit may just be to read the timestamp).  And just because there are
no draws, isn't a reason to skip the flush and return zero.

Signed-off-by: Rob Clark <robdclark@gmail.com>
8 years agoradeonsi: ensure sample locations are set for line and polygon smoothing
Nicolai Hähnle [Sat, 16 Jul 2016 18:41:18 +0000 (20:41 +0200)]
radeonsi: ensure sample locations are set for line and polygon smoothing

Since commit d938b8c, the sample locations are no longer set unconditionally,
so we need to set the atom to dirty on all chips, not just Polaris.

Cc: 12.0 <mesa-stable@lists.freedesktop.org>
8 years agoradeonsi: fix Polaris MSAA regression
Nicolai Hähnle [Sat, 16 Jul 2016 18:37:29 +0000 (20:37 +0200)]
radeonsi: fix Polaris MSAA regression

The regression was introduced by commit d938b8c. The problem here is that in
order to use the small primitive filter, we need to explicitly set the sample
locations to 0. But the DB doesn't properly process the change of sample
locations without a flush, and so we can end up with incorrect Z values.

Instead of doing a flush, just disable the small primitive filter when MSAA
is force-disabled.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96908
Cc: 12.0 <mesa-stable@lists.freedesktop.org>
8 years agofreedreno/ir3: Add missing braces in initializer
francians@gmail.com [Wed, 20 Jul 2016 07:49:34 +0000 (09:49 +0200)]
freedreno/ir3: Add missing braces in initializer

Signed-off-by: Rob Clark <robdclark@gmail.com>
8 years agofreedreno/a2xx: silence missing case 'SHADER_COMPUTE' warning (v2)
francians@gmail.com [Fri, 22 Jul 2016 06:28:46 +0000 (08:28 +0200)]
freedreno/a2xx: silence missing case 'SHADER_COMPUTE' warning (v2)

v2: no need for break after an unreachable (Matt Turner)

Signed-off-by: Francesco Ansanelli <francians@gmail.com>
Signed-off-by: Rob Clark <robdclark@gmail.com>
8 years agoradeonsi: implement buffer_subdata without indirect calls
Marek Olšák [Sat, 16 Jul 2016 19:52:20 +0000 (21:52 +0200)]
radeonsi: implement buffer_subdata without indirect calls

There is less noise in CPU profile data now.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
8 years agogallium/util: don't modify usage in pipe_buffer_write
Marek Olšák [Sun, 17 Jul 2016 12:34:50 +0000 (14:34 +0200)]
gallium/util: don't modify usage in pipe_buffer_write

All drivers were already doing it except virgl.

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
8 years agogallium: split transfer_inline_write into buffer and texture callbacks
Marek Olšák [Sat, 16 Jul 2016 19:19:48 +0000 (21:19 +0200)]
gallium: split transfer_inline_write into buffer and texture callbacks

to reduce the call indirections with u_resource_vtbl.

The worst call tree you could get was:
  - u_transfer_inline_write_vtbl
    - u_default_transfer_inline_write
      - u_transfer_map_vtbl
        - driver_transfer_map
      - u_transfer_unmap_vtbl
        - driver_transfer_unmap

That's 6 indirect calls. Some drivers only had 5. The goal is to have
1 indirect call for drivers that care. The resource type can be determined
statically at most call sites.

The new interface is:
  pipe_context::buffer_subdata(ctx, resource, usage, offset, size, data)
  pipe_context::texture_subdata(ctx, resource, level, usage, box, data,
                                stride, layer_stride)

v2: fix whitespace, correct ilo's behavior

Reviewed-by: Nicolai Hähnle <nicolai.haehnle@amd.com>
Acked-by: Roland Scheidegger <sroland@vmware.com>
8 years agonir: Lower interp_var_at_* like a normal load_var for flat inputs.
Kenneth Graunke [Fri, 22 Jul 2016 00:42:01 +0000 (17:42 -0700)]
nir: Lower interp_var_at_* like a normal load_var for flat inputs.

"flat centroid" and "flat sample" both just mean "flat", so we should
ignore interpolateAtCentroid/Sample and just return the flat value.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97032
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
8 years agomesa: Don't call GenerateMipmap if Width or Height == 0.
Kenneth Graunke [Fri, 22 Jul 2016 05:13:38 +0000 (22:13 -0700)]
mesa: Don't call GenerateMipmap if Width or Height == 0.

One of the WebGL 2.0 conformance tests is trying to call
glGenerateMipmaps with a width and height of 0.  With the meta
implementation, this generates a "framebuffer attachment incomplete"
status, and falls back to the CPU path, calling MapTextureImage.

Except that there's no actual texture to map, and we assert fail.

There's no work to do in this case.  The test expects it to succeed,
so just return early with no error and avoid hassling the driver.

Cc: mesa-stable@lists.freedesktop.org
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=96911
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Anuj Phogat <anuj.phogat@gmail.com>
8 years agoanv/pipeline: Set up point coord enables
Jason Ekstrand [Fri, 22 Jul 2016 22:59:22 +0000 (15:59 -0700)]
anv/pipeline: Set up point coord enables

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Tested-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: "12.0" <mesa-stable@lists.freedesktop.org>
8 years agospirv/nir: Add support for ImageQuerySamples
Jason Ekstrand [Thu, 21 Jul 2016 20:15:59 +0000 (13:15 -0700)]
spirv/nir: Add support for ImageQuerySamples

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agospirv/nir: Handle texture projectors
Jason Ekstrand [Thu, 21 Jul 2016 19:01:46 +0000 (12:01 -0700)]
spirv/nir: Handle texture projectors

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agonir/spirv: Refactor coordinate handling in handle_texture
Jason Ekstrand [Thu, 21 Jul 2016 19:01:35 +0000 (12:01 -0700)]
nir/spirv: Refactor coordinate handling in handle_texture

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agospirv/nir: Refactor type handling in handle_texture
Jason Ekstrand [Thu, 21 Jul 2016 18:56:09 +0000 (11:56 -0700)]
spirv/nir: Refactor type handling in handle_texture

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agospirv/nir: Move opcode selection higher up in handle_texture
Jason Ekstrand [Thu, 21 Jul 2016 18:47:44 +0000 (11:47 -0700)]
spirv/nir: Move opcode selection higher up in handle_texture

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agoanv/image: Assert that the image format is actually supported
Jason Ekstrand [Thu, 21 Jul 2016 04:10:20 +0000 (21:10 -0700)]
anv/image: Assert that the image format is actually supported

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agospirv/nir: Don't increment coord_components for array lod queries
Jason Ekstrand [Thu, 21 Jul 2016 03:57:37 +0000 (20:57 -0700)]
spirv/nir: Don't increment coord_components for array lod queries

For lod query instructions, we really don't care whether or not the sampler
is an array type because that doesn't factor into the LOD.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agoi965: Get rid of the do_lower_unnormalized_offsets pass
Jason Ekstrand [Thu, 21 Jul 2016 20:07:17 +0000 (13:07 -0700)]
i965: Get rid of the do_lower_unnormalized_offsets pass

We can do this in NIR now.  No need to keep a GLSL pass lying around for
it.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agoi965/nir: Enable NIR lowering of txf and rect offsets
Jason Ekstrand [Thu, 21 Jul 2016 19:55:21 +0000 (12:55 -0700)]
i965/nir: Enable NIR lowering of txf and rect offsets

This fixes the following piglit tests on gen6+:

tex-miplevel-selection textureProjGradOffset 2DRect
tex-miplevel-selection textureGradOffset 2DRect
tex-miplevel-selection textureGradOffset 2DRectShadow
tex-miplevel-selection textureProjGradOffset 2DRect_ProjVec4
tex-miplevel-selection textureProjGradOffset 2DRectShadow

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agonir/lower_tex: Add support for lowering coordinate offsets
Jason Ekstrand [Thu, 21 Jul 2016 03:32:31 +0000 (20:32 -0700)]
nir/lower_tex: Add support for lowering coordinate offsets

On i965, we can't support coordinate offsets for texelFetch or rectangle
textures.  Previously, we were doing this with a GLSL pass but we need to
do it in NIR if we want those workarounds for SPIR-V.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agonir/lower_tex: Add some helpers for working with tex sources
Jason Ekstrand [Thu, 21 Jul 2016 03:40:43 +0000 (20:40 -0700)]
nir/lower_tex: Add some helpers for working with tex sources

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agonir: Add a helper for determining the type of a texture source
Jason Ekstrand [Wed, 4 May 2016 03:18:50 +0000 (20:18 -0700)]
nir: Add a helper for determining the type of a texture source

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agoanv/pipeline: Set binding_table.gather_texture_start
Jason Ekstrand [Thu, 21 Jul 2016 03:06:09 +0000 (20:06 -0700)]
anv/pipeline: Set binding_table.gather_texture_start

This should get texture gather working on gen8+ and mostly working on gen7.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agospirv/nir: Properly handle gather components
Jason Ekstrand [Thu, 21 Jul 2016 02:56:45 +0000 (19:56 -0700)]
spirv/nir: Properly handle gather components

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agospirv/nir: Add support for shadow samplers that return vec4
Jason Ekstrand [Thu, 21 Jul 2016 02:55:51 +0000 (19:55 -0700)]
spirv/nir: Add support for shadow samplers that return vec4

While SPIR-V technically doesn't support "old style" shadow, the
shadow-compare gather instruction does return a vec4 so we need to be able
to set the old_style_shadow bit in NIR.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agospirv/nir: Fix some texture opcode asserts
Jason Ekstrand [Thu, 21 Jul 2016 02:23:17 +0000 (19:23 -0700)]
spirv/nir: Fix some texture opcode asserts

We can't get an lod with txf_ms and SPIR-V considers textureGrad to be an
explicit-LOD texturing instruction.

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-dev@lists.freedesktop.org>
8 years agonv50/ir: allow to swap sources for OP_SUB
Samuel Pitoiset [Wed, 20 Jul 2016 18:07:35 +0000 (20:07 +0200)]
nv50/ir: allow to swap sources for OP_SUB

This allows the load-propagation pass to swap the sources in presence
of immediate values.

Maxwell (GM107):

total instructions in shared programs :1928187 -> 1927634 (-0.03%)
total gprs used in shared programs    :330741 -> 330154 (-0.18%)
total local used in shared programs   :28032 -> 28032 (0.00%)

                local        gpr       inst      bytes
    helped           0         271         425         425
      hurt           0           0         194         194

Fermi (GF114):

total instructions in shared programs :2334474 -> 2333829 (-0.03%)
total gprs used in shared programs    :380934 -> 380215 (-0.19%)
total local used in shared programs   :33304 -> 33264 (-0.12%)

                local        gpr       inst      bytes
    helped           5         314         521         521
      hurt           0           4         195         195

No regressions on GM107 and GF114 with full piglit.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
8 years agogallium/radeon: make deferred flushes asynchronous
Marek Olšák [Fri, 15 Jul 2016 13:51:34 +0000 (15:51 +0200)]
gallium/radeon: make deferred flushes asynchronous

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
8 years agogallium: add PIPE_FLUSH_DEFERRED
Marek Olšák [Fri, 15 Jul 2016 13:44:29 +0000 (15:44 +0200)]
gallium: add PIPE_FLUSH_DEFERRED

There are 2 uses:
- Asynchronous flushing for multithreaded drivers.
- Return a fence without flushing (mid-command-buffer fence). The driver
  can defer flushing until fence_finish is called.

This is required to make Bioshock Infinite faster, which creates
1000 fences (flushes) per frame.

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
Reviewed-by: Rob Clark <robdclark@gmail.com>
8 years agogallium/os: use CLOCK_MONOTONIC for sleeps (v2)
Marek Olšák [Sat, 16 Jul 2016 16:38:21 +0000 (18:38 +0200)]
gallium/os: use CLOCK_MONOTONIC for sleeps (v2)

v2: handle EINTR, remove backslashes

Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
8 years agomapi: fix typo in macro name
Eric Engestrom [Fri, 22 Jul 2016 09:24:03 +0000 (10:24 +0100)]
mapi: fix typo in macro name

Fixes: 5ec140c17b54c2592009 ("mapi: Massage code to allow clang to compile.")
Reported-by: Alexandre Demers <alexandre.f.demers@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
Signed-off-by: Eric Engestrom <eric.engestrom@imgtec.com>
8 years agodocs: Put swr back on the GL_ARB_texture_buffer_object_rgb32 list.
Kenneth Graunke [Fri, 22 Jul 2016 16:56:53 +0000 (09:56 -0700)]
docs: Put swr back on the GL_ARB_texture_buffer_object_rgb32 list.

Looks like this was lost when resolving merge conflicts in
commit d1fbd4cdb1bdb8041362a8e5f05833c43a39c9a6.

8 years agoglsl: subroutine types cannot be compared
Andres Gomez [Mon, 18 Jul 2016 13:39:43 +0000 (16:39 +0300)]
glsl: subroutine types cannot be compared

subroutine variables are to be used just in the way functions are
called. Although the spec doesn't say it explicitely, this means that
these variables are not to be used in any other way than those left
for function calls. Therefore, a comparison between 2 subroutine
variables should also cause a compilation error.

From The OpenGL® Shading Language 4.40, page 117:

  "  To use subroutines, a subroutine type is declared, one or more
     functions are associated with that subroutine type, and a
     subroutine variable of that type is declared. The function
     currently assigned to the variable function is then called by
     using function calling syntax replacing a function name with the
     name of the subroutine variable. Subroutine variables are
     uniforms, and are assigned to specific functions only through
     commands (UniformSubroutinesuiv) in the OpenGL API."

From The OpenGL® Shading Language 4.40, page 118:

  "  Subroutine uniform variables are called the same way functions
     are called. When a subroutine variable (or an element of a
     subroutine variable array) is associated with a particular
     function, all function calls through that variable will call that
     particular function."

Fixes GL44-CTS.shader_subroutine.subroutines_cannot_be_assigned_float_int_values_or_be_compared

Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
8 years agoi965: fix varying output setup
Timothy Arceri [Fri, 22 Jul 2016 13:19:09 +0000 (23:19 +1000)]
i965: fix varying output setup

Since 7f53fead5c we treat every location as using all
four components so we only need special handling for
doubles when they cross multiple locations.

This fixes a crash in GL45-CTS.enhanced_layouts.varying_locations
where the outputs array would overflow when a dmat2 was stored at
the max varying location i.e 30.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
8 years agonvc0/mme: fix offsets used for indirect draws
Samuel Pitoiset [Thu, 21 Jul 2016 20:15:53 +0000 (22:15 +0200)]
nvc0/mme: fix offsets used for indirect draws

This fixes a regression introduced in
1da704a94c57aa0b0cf8faaa3236fe47dfb8f88c because the offset has moved
from 0x180 to 0x1a0, and the macros have to be re-compiled.

Fixes: 1da704a ("nvc0: increase the tex handles area size in the driver")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
8 years agonvc0: fix offsets of MP perf counters input parameters
Samuel Pitoiset [Thu, 21 Jul 2016 20:15:52 +0000 (22:15 +0200)]
nvc0: fix offsets of MP perf counters input parameters

This fixes a regression introduced in
1da704a94c57aa0b0cf8faaa3236fe47dfb8f88c because the offset has moved
from 0x600 to 0x620, and the kernels used for reading MP perf counters
have to be re-assembled.

This also fixes amd_performance_monitor_measure piglit.

Fixes: 1da704a ("nvc0: increase the tex handles area size in the driver")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
8 years agomesa: Add GL_BGRA_EXT to the list of GenerateMipmap internal formats.
Kenneth Graunke [Thu, 21 Jul 2016 23:48:05 +0000 (16:48 -0700)]
mesa: Add GL_BGRA_EXT to the list of GenerateMipmap internal formats.

The GL_EXT_texture_format_BGRA8888 extension specification defines a
GL_BGRA_EXT unsized internal format (which is a little odd - usually
BGRA is a pixel transfer format).  The extension is written against
the ES 1.0 specification, so it's a little hard to map, but I believe
it's effectively adding it to the table used here, so we should allow
it here as well.

Note that GL_EXT_texture_format_BGRA8888 is always enabled (dummy_true),
so we don't need to check if it's enabled here.

This fixes mipmap generation in Skia and ChromeOS.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
References: https://bugs.chromium.org/p/chromium/issues/detail?id=630371
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reported-by: Stéphane Marchesin <marcheu@chromium.org>
Cc: mesa-stable@lists.freedesktop.org
8 years agoi965: Fix "operation operation" in comment.
Kenneth Graunke [Fri, 22 Jul 2016 04:28:15 +0000 (21:28 -0700)]
i965: Fix "operation operation" in comment.

From the redundant redundant department.

Reported-by: Michael Schellenberger Costa <mschellenbergercosta@googlemail.com>
8 years agoi965: Fix shared atomic intrinsics to pay attention to base.
Kenneth Graunke [Mon, 18 Jul 2016 21:55:07 +0000 (14:55 -0700)]
i965: Fix shared atomic intrinsics to pay attention to base.

Cc: "12.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
8 years agonir: Add a base const_index to shared atomic intrinsics.
Kenneth Graunke [Mon, 18 Jul 2016 21:47:42 +0000 (14:47 -0700)]
nir: Add a base const_index to shared atomic intrinsics.

Commit 52e75dcb8c04c0dde989970c4c587cbe8313f7cf made nir_lower_io
start using nir_intrinsic_set_base instead of writing const_index[0]
directly.  However, those intrinsics apparently don't /have/ a base,
so this caused assert failures.

However, the old code was happily setting non-existent const_index
fields, so it was pretty bogus too.

Jason pointed out that load_shared and store_shared have a base,
and that the i965 driver uses that field.  So presumably atomics
should have one as well, so that loads/stores/atomics all refer
to variables with consistent addressing.

Cc: "12.0" <mesa-stable@lists.freedesktop.org>
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Timothy Arceri <timothy.arceri@collabora.com>
8 years agoglsl: re-enable varying packing in GL4.4+
Timothy Arceri [Thu, 26 May 2016 04:50:43 +0000 (14:50 +1000)]
glsl: re-enable varying packing in GL4.4+

We can still do packing we just need to get the packing type from the consumer
rather than the producer.

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97033

8 years agoi965: Include VUE handles for GS with invocations > 1.
Kenneth Graunke [Wed, 20 Jul 2016 23:11:18 +0000 (16:11 -0700)]
i965: Include VUE handles for GS with invocations > 1.

We always resort to the pull model for instanced GS inputs.  So, we'd
better include the VUE handles, or else we can't actually pull anything.

Ian reports that on his branch with OES_geometry_shader enabled,
this fixes a bunch of dEQP-GLES31.functional.geometry_shading tests::

- instanced.draw_2_instances_geometry_2_invocations
- instanced.draw_2_instances_geometry_8_invocations
- instanced.draw_4_instances_geometry_2_invocations
- instanced.draw_4_instances_geometry_8_invocations
- instanced.draw_8_instances_geometry_2_invocations
- instanced.draw_8_instances_geometry_8_invocations
- instanced.geometry_2_invocations
- instanced.geometry_32_invocations
- instanced.geometry_8_invocations
- instanced.geometry_max_invocations
- instanced.geometry_output_different_2_invocations
- instanced.geometry_output_different_32_invocations
- instanced.geometry_output_different_8_invocations
- instanced.geometry_output_different_max_invocations
- instanced.invocation_output_vary_by_attribute
- instanced.invocation_output_vary_by_texture
- instanced.invocation_output_vary_by_uniform
- query.primitives_generated_instanced

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Samuel Iglesias Gonsálvez <siglesias@igalia.com>
Tested-by: Ian Romanick <ian.d.romanick@intel.com>
8 years agomesa: Add -fno-math-errno -fno-trapping-math to CXXFLAGS.
Matt Turner [Thu, 30 Jun 2016 22:02:49 +0000 (15:02 -0700)]
mesa: Add -fno-math-errno -fno-trapping-math to CXXFLAGS.

Not sure why I forgot to add them to CXXFLAGS in commit f55c408067 or
commit 875458b778. Cuts about 1k of .text.

   text     data      bss      dec      hex  filename
5806354   287816    29384  6123554   5d7022  i965_dri.so before
5805497   287744    29384  6122625   5d6c81  i965_dri.so after

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
8 years agomesa: Drop -fno-builtin-memcmp.
Matt Turner [Thu, 30 Jun 2016 21:16:16 +0000 (14:16 -0700)]
mesa: Drop -fno-builtin-memcmp.

According to the referenced bug report, gcc-4.5 and newer do not inline
memcmp(). I see no difference in performance of ipers with llvmpipe on a
Sandybridge (which does not have "Enhanced REP MOVSB/STOSB") by removing
this flag.

I attempted to confirm the problem with gcc-4.4, but it fails to compile
for quite a few different reasons.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
8 years agomapi: Massage code to allow clang to compile.
Matt Turner [Mon, 11 Jul 2016 17:44:25 +0000 (10:44 -0700)]
mapi: Massage code to allow clang to compile.

According to https://llvm.org/bugs/show_bug.cgi?id=19778#c3 this code
was violating the spec, resulting in it failing to compile.

Cc: mesa-stable@lists.freedesktop.org
Co-authored-by: Tomasz Paweł Gajc <tpgxyz@gmail.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89599
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
8 years agodocs: Add extensions not part of any GL or GL ES version
Ian Romanick [Tue, 31 May 2016 23:04:00 +0000 (16:04 -0700)]
docs: Add extensions not part of any GL or GL ES version

Based loosely on patches submitted ages ago by Thomas Helland.

v2: Add lots of missing data provided by Ilia.  Fix sort order of
GL_ARB_sparse_texture extensions suggested by Ilia.

v3: Note that Dave Airlie has started work on GL_ARB_bindless_texture.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
8 years agodocs: Update GL3.txt for OpenGL 4.0 on i965-ish hardware
Ian Romanick [Mon, 16 May 2016 18:30:19 +0000 (11:30 -0700)]
docs: Update GL3.txt for OpenGL 4.0 on i965-ish hardware

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
8 years agodocs: Update GL3.txt for OpenGL ES on i965-ish hardware
Ian Romanick [Mon, 16 May 2016 18:26:15 +0000 (11:26 -0700)]
docs: Update GL3.txt for OpenGL ES on i965-ish hardware

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
8 years agoi965: print error messages if gs fails to compile
Timothy Arceri [Thu, 21 Jul 2016 04:07:01 +0000 (14:07 +1000)]
i965: print error messages if gs fails to compile

We do this for all other stages.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
8 years agoi965: enable GL4.4 for Gen8+
Timothy Arceri [Thu, 21 Jul 2016 01:53:14 +0000 (11:53 +1000)]
i965: enable GL4.4 for Gen8+

Acked-by: Kenneth Graunke <kenneth@whitecape.org>
8 years agoi965: enable ARB_enhanced_layouts for gen6+
Timothy Arceri [Fri, 20 May 2016 03:37:45 +0000 (13:37 +1000)]
i965: enable ARB_enhanced_layouts for gen6+

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
8 years agoi965/vec4: add packing support for tcs load outputs
Timothy Arceri [Fri, 24 Jun 2016 02:31:56 +0000 (12:31 +1000)]
i965/vec4: add packing support for tcs load outputs

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
8 years agoi965/vec4: add support for packing tes inputs
Timothy Arceri [Fri, 24 Jun 2016 02:14:30 +0000 (12:14 +1000)]
i965/vec4: add support for packing tes inputs

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
8 years agoi965/vec4: add support for packing tcs outputs
Timothy Arceri [Fri, 24 Jun 2016 02:10:12 +0000 (12:10 +1000)]
i965/vec4: add support for packing tcs outputs

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
8 years agoi965/vec4: support packing tcs inputs
Timothy Arceri [Thu, 23 Jun 2016 04:26:16 +0000 (14:26 +1000)]
i965/vec4: support packing tcs inputs

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
8 years agoi965/vec4: add component packing for gs
Timothy Arceri [Thu, 23 Jun 2016 03:03:47 +0000 (13:03 +1000)]
i965/vec4: add component packing for gs

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
8 years agoi965/vec4: add support for packing vs/gs/tes outputs
Timothy Arceri [Thu, 23 Jun 2016 02:49:53 +0000 (12:49 +1000)]
i965/vec4: add support for packing vs/gs/tes outputs

Here we create a new output_generic_reg array with the ability to
store the dst_reg for each component of user defined varyings.
This is needed as the previous code only stored the dst_reg based
on the varying location which meant packed varyings would overwrite
each other.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
8 years agoi965/vec4: add support for packing inputs
Timothy Arceri [Wed, 22 Jun 2016 23:37:03 +0000 (09:37 +1000)]
i965/vec4: add support for packing inputs

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
8 years agoi965: add helper for creating packing writemask
Timothy Arceri [Thu, 23 Jun 2016 02:52:05 +0000 (12:52 +1000)]
i965: add helper for creating packing writemask

For example where n=3 first_component=1 this will give us
0xE (WRITEMASK_YZW).

V2:
Add assert to check first component is <= 4 (Suggested by Ken)

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
8 years agoi965: add helpers for creating component layout swizzle
Timothy Arceri [Wed, 22 Jun 2016 23:32:39 +0000 (09:32 +1000)]
i965: add helpers for creating component layout swizzle

This will be used to swizzle components to the beginning or end
of the vector based on the component layout qualifier and whether
we are doing a load or store.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
8 years agovc4: Return V3D version details in the GL renderer info.
Eric Anholt [Tue, 12 Jul 2016 18:57:13 +0000 (11:57 -0700)]
vc4: Return V3D version details in the GL renderer info.

This is as close as we get to a name for the 3D blocks.

8 years agovc4: Check the V3D version reported by the kernel.
Eric Anholt [Tue, 12 Jul 2016 00:55:33 +0000 (17:55 -0700)]
vc4: Check the V3D version reported by the kernel.

We don't want to bring up an old userspace driver on a kernel for
newer hardware.  We'll also want to look at the other ident fields in
the future.

8 years agovc4: Detect and report kernel support for branching.
Eric Anholt [Thu, 10 Mar 2016 23:18:31 +0000 (15:18 -0800)]
vc4: Detect and report kernel support for branching.

8 years agovc4: Switch to using the libdrm-provided vc4_drm.h.
Eric Anholt [Sat, 2 Jul 2016 21:35:34 +0000 (14:35 -0700)]
vc4: Switch to using the libdrm-provided vc4_drm.h.

The required version is set to .69 for the getparam ioctl that will be
used in the next commit.

8 years agodocs: mark ARB_enhanced_layouts as DONE for i965
Timothy Arceri [Wed, 16 Dec 2015 02:55:50 +0000 (13:55 +1100)]
docs: mark ARB_enhanced_layouts as DONE for i965

Reviewed-by: Edward O'Callaghan <funfunctor@folklore1984.net>
8 years agoi965: enable ARB_enhanced_layouts for gen8+
Timothy Arceri [Fri, 20 May 2016 03:37:45 +0000 (13:37 +1000)]
i965: enable ARB_enhanced_layouts for gen8+

Acked-by: Edward O'Callaghan <funfunctor@folklore1984.net>
8 years agonir: add doubles component packing support
Timothy Arceri [Tue, 19 Jul 2016 05:40:14 +0000 (15:40 +1000)]
nir: add doubles component packing support

This makes sure we give the correct driver location
for doubles when using component packing. Specifically
it handles packing a dvec3 with a double which is the
only packing scenario allowed which spans across two
locations.

Acked-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
8 years agoi965: add component packing support for load_output intrinsics
Timothy Arceri [Wed, 15 Jun 2016 02:35:49 +0000 (12:35 +1000)]
i965: add component packing support for load_output intrinsics

Here we use the component qualifier (which is the first component)
as an offset when loading output varyings.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
8 years agoi965: enable component packing for vs and fs
Timothy Arceri [Mon, 23 May 2016 06:48:05 +0000 (16:48 +1000)]
i965: enable component packing for vs and fs

Rather than trying to work out the total number of components
used at a location we simply treat all outputs as vec4s. This
removes the need for complex code looping over varyings to match
packed locations and the need for storing the total number of
components used at each location.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
8 years agoi965: bring back type_size_vec4_times_4()
Timothy Arceri [Tue, 19 Jul 2016 05:13:51 +0000 (15:13 +1000)]
i965: bring back type_size_vec4_times_4()

We will use this for output varyings. To make component
packing simpler we will just treat all varyings as vec4s.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
8 years agonir/inline: Constant-initialize local variables in the callee if needed
Jason Ekstrand [Fri, 15 Jul 2016 22:44:32 +0000 (15:44 -0700)]
nir/inline: Constant-initialize local variables in the callee if needed

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-stable@lists.freedesktop.org>
8 years agonir: Add a nir_deref_foreach_leaf helper
Jason Ekstrand [Fri, 15 Jul 2016 22:21:19 +0000 (15:21 -0700)]
nir: Add a nir_deref_foreach_leaf helper

Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Cc: "12.0" <mesa-stable@lists.freedesktop.org>
8 years agoclover: Re-order includes in invocation.cpp to fix build
Tom Stellard [Tue, 19 Jul 2016 15:07:23 +0000 (15:07 +0000)]
clover: Re-order includes in invocation.cpp to fix build

The build was failing because the official CL headers have a few defines,
like: # define cl_khr_gl_sharing 1

Which have the same name as some class members of clang's OpenCLOptions class.
If we include the cl headers first, this breaks the build because the member
names of this class are replaced by the literal 1.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Vedran Miletić <vedran@miletic.net>
8 years agoclover: Add missing include v2
Tom Stellard [Tue, 19 Jul 2016 15:05:34 +0000 (15:05 +0000)]
clover: Add missing include v2

clang commit r275822 removed unnecessary includes from header files,
so we now need to explicitly include clang/Lex/PreprocessorOptions.h

v2:
  - Use <> instead of "" for the include path.

Reviewed-by: Francisco Jerez <currojerez@riseup.net>
Reviewed-by: Vedran Miletić <vedran@miletic.net>
8 years agoi965: Move VS load_input handling to nir_emit_vs_intrinsic().
Kenneth Graunke [Wed, 20 Jul 2016 02:00:19 +0000 (19:00 -0700)]
i965: Move VS load_input handling to nir_emit_vs_intrinsic().

TCS/TES/GS and now FS all handle these in stage-specific functions.
CS don't have inputs, so VS was the only one left using this code.

Move it to the VS-specific function for clarity.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
8 years agoi965: Delete the FS_OPCODE_INTERPOLATE_AT_CENTROID virtual opcode.
Kenneth Graunke [Sat, 16 Jul 2016 18:52:36 +0000 (11:52 -0700)]
i965: Delete the FS_OPCODE_INTERPOLATE_AT_CENTROID virtual opcode.

We no longer use this message.  As far as I can tell, it's fairly
useless - the equivalent information is provided in the payload.

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
8 years agoi965: Rewrite FS input handling to use the new NIR intrinsics.
Kenneth Graunke [Tue, 12 Jul 2016 10:57:25 +0000 (03:57 -0700)]
i965: Rewrite FS input handling to use the new NIR intrinsics.

This eliminates the need to walk the list of input variables, recurse
into their types (via logic largely redundant with nir_lower_io), and
interpolate all possible inputs up front.  The backend no longer has
to care about variables at all, which eliminates complications from
trying to pack multiple variables into the same location.  Instead,
each intrinsic specifies exactly what's needed.

This should unblock Timothy's work on GL_ARB_enhanced_layouts.

Each load_interpolated_input intrinsic corresponds to PLN instructions,
while load_barycentric_at_* intrinsics correspond to pixel interpolator
messages.  The pixel/centroid/sample barycentric intrinsics simply refer
to payload fields (delta_xy[]), and don't actually generate any code.

Because we use a single intrinsic for both centroid-qualified variables
and interpolateAtCentroid(), they become indistinguishable.  We stop
sending pixel interpolator messages for those, and instead use the
payload provided data, which should be considerably faster.

On Broadwell:

total instructions in shared programs: 9067751 -> 9067570 (-0.00%)
instructions in affected programs: 145902 -> 145721 (-0.12%)
helped: 422
HURT: 209

total spills in shared programs: 2849 -> 2899 (1.76%)
spills in affected programs: 760 -> 810 (6.58%)
helped: 0
HURT: 10

total fills in shared programs: 3910 -> 3950 (1.02%)
fills in affected programs: 617 -> 657 (6.48%)
helped: 0
HURT: 10

LOST:   3
GAINED: 3

The differences mostly appear to be slight changes in MOVs.

v2: Use nir_shader_compiler_options::use_interpolated_input_intrinsics
    flag rather than passing it directly to nir_lower_io.  Use the
    unreachable() macro rather than assert in one place.  (Review
    feedback from Chris Forbes.)

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisforbes@google.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
8 years agoi965: Move load_interpolated_input/barycentric_* intrinsics to the top.
Kenneth Graunke [Mon, 18 Jul 2016 01:44:58 +0000 (18:44 -0700)]
i965: Move load_interpolated_input/barycentric_* intrinsics to the top.

Currently, i965 interpolates all FS inputs at the top of the program.
This has advantages and disadvantages, but I'd like to keep that policy
while reworking this code.  We can consider changing it independently.

The next patch will make the compiler generate PLN instructions "on the
fly", when it encounters an input load intrinsic, rather than doing it
for all inputs at the start of the program.

To emulate this behavior, we introduce an ugly pass to move all NIR
load_interpolated_input and payload-based (not interpolator message)
load_barycentric_* intrinsics to the shader's start block.

This helps avoid regressions in shader-db for cases such as:

   if (...) {
      ...load some input...
   } else {
      ...load that same input...
   }

which CSE can't handle, because there's no dominance relationship
between the two loads.  Because the start block dominates all others,
we can CSE all inputs and emit PLNs exactly once, as we did before.

Ideally, global value numbering would eliminate these redundant loads,
while not forcing them all the way to the start block.  When that lands,
we should consider dropping this hacky pass.

Again, this pass currently does nothing, as i965 doesn't generate these
intrinsics yet.  But it will shortly, and I figured I'd separate this
code as it's relatively self-contained.

v2: Dramatically simplify pass - instead of creating new instructions,
    just remove/re-insert their list nodes (suggested by Jason Ekstrand).

Signed-off-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Chris Forbes <chrisforbes@google.com> [v1]
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>