mesa.git
6 years agomesa: add OES_EGL_image_external_essl3 support
Ilia Mirkin [Fri, 2 Feb 2018 05:47:54 +0000 (07:47 +0200)]
mesa: add OES_EGL_image_external_essl3 support

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
6 years agor600/fp64: Fix build.
Vinson Lee [Mon, 5 Feb 2018 23:24:45 +0000 (23:24 +0000)]
r600/fp64: Fix build.

  CC       r600_shader.lo
r600_shader.c: In function ‘egcm_int_to_double’:
r600_shader.c:4543:12: error: ‘ctx’ is a pointer; did you mean to use ‘->’?
     if (ctx.bc->chip_class == CAYMAN)
            ^
            ->

Fixes: 35b430157776 ("r600/fp64: fix integer->double conversion")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
6 years agor600/fp64: fix integer->double conversion
Dave Airlie [Mon, 29 Jan 2018 00:55:15 +0000 (10:55 +1000)]
r600/fp64: fix integer->double conversion

Doing a straight uint/int->fp32->fp64 conversion causes
some precision issues, Roland suggested splitting the
integer into two portions and doing two separate
int->fp32->fp64 conversions then adding the results.

This passes the tests in CTS and piglit.

[airlied: fix cypress conversion opcodes]
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
6 years agoac/nir: remove emission of nir_op_fdiv
Samuel Pitoiset [Mon, 5 Feb 2018 20:37:05 +0000 (21:37 +0100)]
ac/nir: remove emission of nir_op_fdiv

RadeonSI and RADV lower fdiv.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agotravis: add macOS meson build
Jon Turney [Tue, 16 Jan 2018 17:51:53 +0000 (17:51 +0000)]
travis: add macOS meson build

v2: Simplify set of options now we have better defaults

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agomeson: osx ld doesn't support --build-id
Jon Turney [Sun, 3 Dec 2017 21:58:12 +0000 (21:58 +0000)]
meson: osx ld doesn't support --build-id

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
6 years agomeson: build src/glx/apple
Jon Turney [Sat, 2 Dec 2017 17:13:39 +0000 (17:13 +0000)]
meson: build src/glx/apple

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agomeson: set apple glx defines
Dylan Baker [Sat, 28 Oct 2017 04:08:07 +0000 (21:08 -0700)]
meson: set apple glx defines

Reviewed-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agomeson: better defaults for osx, windows and cygwin
Jon Turney [Fri, 2 Feb 2018 22:25:48 +0000 (22:25 +0000)]
meson: better defaults for osx, windows and cygwin

set suitable defaults for 'dri-drivers', 'gallium-drivers', 'vulkan-drivers'
and 'platforms' options for osx, windows and cygwin, adding cygwin where
appropriate.

v2: error() for unknown OS

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
6 years agoi965: Move mistakenly placed line
Matt Turner [Wed, 31 Jan 2018 19:09:36 +0000 (11:09 -0800)]
i965: Move mistakenly placed line

Ken called this out in review, but it seems I forgot to make the change.
I noticed that the control flow annotations in the fragment shader
disassembly of tests/shaders/glsl-fs-loop-continue.shader_test were not
correct, and moving this line to the correct place fixes it.

6 years agoglsl/linker: check same name is not used in block and outside
Juan A. Suarez Romero [Mon, 5 Feb 2018 16:38:39 +0000 (17:38 +0100)]
glsl/linker: check same name is not used in block and outside

According with OpenGL GLSL 3.20 spec, section 4.3.9:

  "It is a link-time error if any particular shader interface
   contains:
     - two different blocks, each having no instance name, and each
       having a member of the same name, or
     - a variable outside a block, and a block with no instance name,
       where the variable has the same name as a member in the block."

This fixes a previous commit 9b894c8 ("glsl/linker: link-error using the
same name in unnamed block and outside") that covered this case, but
did not take in account that precision qualifiers are ignored when
comparing blocks with no instance name.

With this commit, the original tests
KHR-GL*.shaders.uniform_block.common.name_matching keep fixed, and also
dEQP-GLES31.functional.shaders.linkage.uniform.block.differing_precision
regression is fixed, which was broken by previous commit.

v2: use helper varibles (Matteo Bruni)

Fixes: 9b894c8 ("glsl/linker: link-error using the same name in unnamed block and outside")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104668
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104777
CC: Mark Janes <mark.a.janes@intel.com>
CC: "18.0" <mesa-stable@lists.freedesktop.org>
Tested-by: Matteo Bruni <matteo.mystral@gmail.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
6 years agomesa: enable ASTC format for CompressedTexSubImage3D
Juan A. Suarez Romero [Mon, 5 Feb 2018 16:00:11 +0000 (17:00 +0100)]
mesa: enable ASTC format for CompressedTexSubImage3D

If extensions GL_KHR_texture_compression_astc_hdr or
GL_KHR_texture_compression_astc_sliced_3d are implemented then ASTC
format are supported in CompressedTex*Îmage3D.

Fixes KHR-GLES2.texture_3d.* with this format.

CC: Eric Anholt <eric@anholt.net>
Reviewed-by: Eric Anholt <eric@anholt.net>
Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
6 years agoutil/build-id: Fix address comparison for binaries with LOAD vaddr > 0
Stephan Gerhold [Wed, 24 Jan 2018 14:13:24 +0000 (15:13 +0100)]
util/build-id: Fix address comparison for binaries with LOAD vaddr > 0

build_id_find_nhdr_for_addr() fails to find the build-id if the first LOAD
segment has a virtual address other than 0x0.

For most shared libraries, the first LOAD segment has vaddr=0x0:

    Type           Offset   VirtAddr   PhysAddr   FileSiz MemSiz  Flg Align
    LOAD           0x000000 0x00000000 0x00000000 0x2d2e26 0x2d2e26 R E 0x1000
    LOAD           0x2d2e54 0x002d3e54 0x002d3e54 0x2e248 0x2f148 RW  0x1000

However, compiling the Intel Vulkan driver as 32-bit binary on Android produces
the following ELF header with vaddr=0x8000 instead:

    Type           Offset   VirtAddr   PhysAddr   FileSiz MemSiz  Flg Align
    PHDR           0x000034 0x00008034 0x00008034 0x00100 0x00100 R   0x4
    LOAD           0x000000 0x00008000 0x00008000 0x224a04 0x224a04 R E 0x1000
    LOAD           0x225710 0x0022e710 0x0022e710 0x25988 0x27364 RW  0x1000

build_id_find_nhdr_callback() compares the address of dli_fbase from dladdr()
and dlpi_addr from dl_iterate_phdr(). With vaddr > 0, these point to a
different memory address, e.g.:

    dli_fbase=0xd8395000 (offset 0x8000)
    dlpi_addr=0xd838d000

At least on glibc and bionic (Android) dli_fbase refers to the address where
the shared object is mapped into the process space, whereas dlpi_addr is just
the base address for the vaddrs declared in the ELF header.

To compare them correctly, we need to calculate the start of the mapping
by adding the vaddr of the first LOAD segment to the base address.

Note: musl users will need the following patch.
https://git.musl-libc.org/cgit/musl/commit/?id=b3ae7beabb9f0c219bb8a8b63567a01c6530c1ac

Cc: Chad Versace <chadversary@chromium.org>
Cc: <mesa-stable@lists.freedesktop.org>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104642
Fixes: 5c98d38 "util: Query build-id by symbol address, not library name"
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
6 years agoradeonsi: enable vcn encode for HEVC main
Boyuan Zhang [Thu, 25 Jan 2018 20:06:35 +0000 (15:06 -0500)]
radeonsi: enable vcn encode for HEVC main

Enable vcn encode for HEVC main profile on Raven.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
6 years agost/va: implement HEVC encode functions
Boyuan Zhang [Thu, 1 Feb 2018 21:25:44 +0000 (16:25 -0500)]
st/va: implement HEVC encode functions

Implement HEVC encode functions based on VAAPI HEVC encode interface.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
6 years agost/va: add HEVC encode functions
Boyuan Zhang [Thu, 1 Feb 2018 21:20:16 +0000 (16:20 -0500)]
st/va: add HEVC encode functions

Add a separate file for HEVC encode functions.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
6 years agost/va: enable dual instances encode only for H264
Boyuan Zhang [Thu, 25 Jan 2018 19:32:04 +0000 (14:32 -0500)]
st/va: enable dual instances encode only for H264

Logics that related to dual instances encode should only be done for
H264, not other codecs.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
6 years agost/va: add entrypoint check for HEVC
Boyuan Zhang [Thu, 25 Jan 2018 19:21:13 +0000 (14:21 -0500)]
st/va: add entrypoint check for HEVC

Add entrypoint check for HEVC to differentiate decode and encode jobs.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
6 years agost/va: add HEVC picture desc
Boyuan Zhang [Thu, 25 Jan 2018 19:18:09 +0000 (14:18 -0500)]
st/va: add HEVC picture desc

Add HEVC picture desc, and add codec check when creating and destroying
context.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
6 years agost/va: move H264 enc functions into separate file
Boyuan Zhang [Thu, 1 Feb 2018 20:47:10 +0000 (15:47 -0500)]
st/va: move H264 enc functions into separate file

Move all H264 encode related functions into separate file. Similar to
VAAPI decode side, there will be separate file for each codec on encode
side as well.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
6 years agoradeon/vcn: add header implementations for HEVC
Boyuan Zhang [Thu, 1 Feb 2018 20:29:30 +0000 (15:29 -0500)]
radeon/vcn: add header implementations for HEVC

Implement encoding of sps, pps, vps, aud, and slice headers for HEVC
based on HEVC specs.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
6 years agoradeon/vcn: add ib implementations for HEVC
Boyuan Zhang [Thu, 25 Jan 2018 16:25:20 +0000 (11:25 -0500)]
radeon/vcn: add ib implementations for HEVC

Implement required ibs for vcn HEVC encode.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
6 years agoradeon/vcn: support picture parameters for HEVC
Boyuan Zhang [Thu, 1 Feb 2018 20:23:49 +0000 (15:23 -0500)]
radeon/vcn: support picture parameters for HEVC

Pass pipe_picture_desc instead of pipe_h264_enc_picture_desc so that
it can be used for different codecs. Add functions to handle picture
parameters that will be used for HEVC encode.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
6 years agoradeon/vcn: add vcn encode interface for HEVC
Boyuan Zhang [Thu, 1 Feb 2018 20:05:17 +0000 (15:05 -0500)]
radeon/vcn: add vcn encode interface for HEVC

Add vcn encode interface for HEVC, and rename radeon_enc_h264_enc_pic
to radeon_enc_pic since radeon_enc_pic is used by both H264 and HEVC.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
6 years agovl: add parameters for HEVC encode
Boyuan Zhang [Thu, 1 Feb 2018 19:57:37 +0000 (14:57 -0500)]
vl: add parameters for HEVC encode

Add HEVC encode interface

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
6 years agobroadcom/vc5: Ignore samplers for finding uniform offsets.
Eric Anholt [Mon, 5 Feb 2018 12:58:06 +0000 (12:58 +0000)]
broadcom/vc5: Ignore samplers for finding uniform offsets.

Fixes:
KHR-GLES3.shaders.struct.uniform.sampler_array_fragment
KHR-GLES3.shaders.struct.uniform.sampler_array_vertex
KHR-GLES3.shaders.struct.uniform.sampler_nested_fragment
KHR-GLES3.shaders.struct.uniform.sampler_nested_vertex

6 years agobroadcom/vc5: Fix non-mipfiltered sampling.
Eric Anholt [Mon, 5 Feb 2018 11:06:59 +0000 (11:06 +0000)]
broadcom/vc5: Fix non-mipfiltered sampling.

We need to clamp the LOD to 0 if mip filtering is disabled.  This is part
of fixing KHR-GLES3.shaders.struct.uniform.sampler_array_fragment.

6 years agobroadcom/vc5: Fix "hardwrae" typo in a field name in XML.
Eric Anholt [Mon, 5 Feb 2018 10:11:30 +0000 (10:11 +0000)]
broadcom/vc5: Fix "hardwrae" typo in a field name in XML.

6 years agoac/nir: fix a crash in load_gs_input() on pre-GFX9 chips
Samuel Pitoiset [Fri, 2 Feb 2018 17:56:39 +0000 (18:56 +0100)]
ac/nir: fix a crash in load_gs_input() on pre-GFX9 chips

Fixes: df1d5174fcc ("ac/nir: replace SI.buffer.load.dword with amdgcn.buffer.load")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agobroadcom/vc5: Try to merge more than 2 QPU instructions together.
Eric Anholt [Sun, 4 Feb 2018 21:37:22 +0000 (21:37 +0000)]
broadcom/vc5: Try to merge more than 2 QPU instructions together.

Obviously it would be good to have an ADD and a MUL and a signal together,
but we can even potentially have multiple signals merged, as well.

total instructions in shared programs: 100423 -> 97874 (-2.54%)
instructions in affected programs:     78812 -> 76263 (-3.23%)

6 years agobroadcom/vc5: Remove no-op MOVs after register allocation.
Eric Anholt [Sun, 4 Feb 2018 21:05:03 +0000 (21:05 +0000)]
broadcom/vc5: Remove no-op MOVs after register allocation.

We emit some MOVs to track lifetimes of payload registers, but we don't
need there to be actual MOV instructions for them.

total instructions in shared programs: 101045 -> 100423 (-0.62%)
instructions in affected programs:     37083 -> 36461 (-1.68%)

6 years agobroadcom/vc5: Add missing shader-db instruction counting.
Eric Anholt [Sun, 4 Feb 2018 21:27:32 +0000 (21:27 +0000)]
broadcom/vc5: Add missing shader-db instruction counting.

I must have misplaced it in the instruction packing rework.

6 years agor600: fix resq for buffer images.
Dave Airlie [Fri, 2 Feb 2018 07:28:15 +0000 (17:28 +1000)]
r600: fix resq for buffer images.

If this is an image buffer, we need to calculate the correct resource
id.

Fixes:
KHR-GL45.shader_image_size.*

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
6 years agor600/eg: fix cube map array buffer images.
Dave Airlie [Fri, 2 Feb 2018 06:56:27 +0000 (16:56 +1000)]
r600/eg: fix cube map array buffer images.

This fixes a crash in:
KHR-GL45.texture_cube_map_array.texture_size_compute_sh.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
6 years agomesa: change ctx->Color.ColorMask into a 32-bit bitmask
Marek Olšák [Wed, 31 Jan 2018 02:03:25 +0000 (03:03 +0100)]
mesa: change ctx->Color.ColorMask into a 32-bit bitmask

4 bits per draw buffer, 8 draw buffers in total --> 32 bits.

This is easier to work with.

Reviewed-by: Eric Anholt <eric@anholt.net>
6 years agoi965: Create new program cache bo when clearing the program cache
Jordan Justen [Fri, 2 Feb 2018 21:03:10 +0000 (13:03 -0800)]
i965: Create new program cache bo when clearing the program cache

When the disk shader cache CI testing was enabled, we started noticing
occasional failures on deqp test runs. (Mainly SNB, rarely HSW)

Before this change, when we cleared the (in memory) program cache we
reused the same bo. Since the disk shader cache quickly restores
programs, it appears that this would lead to overwrites of the older
program binaries in the in memory program cache that apparently were
still executing in some cases. If these programs were still executing,
this could cause a GPU hang.

This issue is probably not disk shader cache specific, but may have
been hidden due to the compiler taking time to recompile programs
after the cache was cleared.

v2:
 * Don't add `copy` param to brw_cache_new_bo (Ken)
 * Call from brw_program_cache_check_size (Ken)

Cc: Kenneth Graunke <kenneth@whitecape.org>
Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
6 years agoaubinator: Multiply count by 4 to compute buffer sizes
Jason Ekstrand [Tue, 30 Jan 2018 04:48:57 +0000 (20:48 -0800)]
aubinator: Multiply count by 4 to compute buffer sizes

The count field is in terms of dwords and not bytes.  In
7d4007d58ab7c0c1796e116b55814f8be4e699a9, I fixed one instance
of this but missed another.

6 years agobroadcom/vc5: Enable UIF XOR on textures.
Eric Anholt [Mon, 22 Jan 2018 01:14:25 +0000 (09:14 +0800)]
broadcom/vc5: Enable UIF XOR on textures.

This should increase performance by reducing SDRAM bank conflicts when
crossing between UIF columns (particularly on power-of-two height
textures).

The uif_xor_disable setup is dropped, since we need to allow XOR on lower
miplevels even when level 0 is XOR.  The level 0 force UIF and level 0 XOR
flags should handle setting XOR properly on imported buffers.

6 years agobroadcom/vc5: Fix alignment of miplevel 1 with UIF.
Eric Anholt [Fri, 2 Feb 2018 16:50:51 +0000 (08:50 -0800)]
broadcom/vc5: Fix alignment of miplevel 1 with UIF.

The alignment here means that we can't get back the padded height from the
size/stride any more, so it's now a field in the slice as well.

Fixes piglit fbo-generatemipmap-formats RGBA16 NPOT.

6 years agobroadcom/vc5: Switch our RGBA4 support to the new gallium format.
Eric Anholt [Sat, 20 Jan 2018 06:35:20 +0000 (22:35 -0800)]
broadcom/vc5: Switch our RGBA4 support to the new gallium format.

Fixes fbo-generatemipmap-formats, fbo-alphatest-formats, etc. tests for
GL_RGBA4, GL_RGB4, GL_RGBA2, etc.

6 years agogallium: Add a new A4B4G4R4 pipe format for Broadcom.
Eric Anholt [Sat, 20 Jan 2018 18:02:07 +0000 (10:02 -0800)]
gallium: Add a new A4B4G4R4 pipe format for Broadcom.

The VC5 HW puts A in the low bits and R in the high bits.  We can't just
swizzle in the shaders because the blending HW can't pick what channel A
is in, so make a new format to match it.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agomesa: Drop incorrect A4B4G4R4 _mesa_format_matches_format_and_type() cases.
Eric Anholt [Thu, 1 Feb 2018 19:12:47 +0000 (11:12 -0800)]
mesa: Drop incorrect A4B4G4R4 _mesa_format_matches_format_and_type() cases.

swapBytes operates on bytes, not 4-bit channels, so you can't just take
non-swapBytes cases and flip the REV flag.

Avoids piglit texture-packed-formats regressions when enabling the
ABGR4444 format.

Fixes: c5a5c9a7db89 ("mesa/formats: add new mesa formats and their pack/unpack functions.")
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agomeson/swr: Updated copyright dates
George Kyriazis [Thu, 1 Feb 2018 16:46:27 +0000 (10:46 -0600)]
meson/swr: Updated copyright dates

cc: mesa-stable@lists.freedesktop.org
cc: dylan@pnwbakers.com

Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
6 years agomeson/swr: re-shuffle generated files
George Kyriazis [Thu, 1 Feb 2018 03:44:54 +0000 (21:44 -0600)]
meson/swr: re-shuffle generated files

Move generated files from codegen/meson.build to other directories, in order
to satisfy generated include file dependencies

Add correct file lists for architecture-specific libraries.

cc: mesa-stable@lists.freedesktop.org
cc: dylan@pnwbakers.com

Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
6 years agoamd: remove support for LLVM 3.9
Marek Olšák [Fri, 2 Feb 2018 18:26:49 +0000 (19:26 +0100)]
amd: remove support for LLVM 3.9

Only these are supported:
- LLVM 4.0
- LLVM 5.0
- LLVM 6.0
- master (7.0)

Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agomeson: Check for actual LLVM required versions
Dylan Baker [Fri, 2 Feb 2018 18:45:12 +0000 (10:45 -0800)]
meson: Check for actual LLVM required versions

Currently we always check for 3.9.0, which is pretty safe since
everything except radv work with >= 3.9 and 3.9 is pretty old at this
point. However, radv actually requires 4.0, and there is a patch for
radeonsi to do the same.

Fixes: 673dda833076 ("meson: build "radv" vulkan driver for radeon hardware")
Signed-off-by: Dylan Baker <dylan.c.baker@intel.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agomeson: Don't confuse the install and search paths for dri drivers
Dylan Baker [Tue, 16 Jan 2018 18:36:28 +0000 (10:36 -0800)]
meson: Don't confuse the install and search paths for dri drivers

Currently there is not a separate option for setting the search path of
DRI drivers in meson, like there is in scons and autotools. This is an
oversight and needs to be fixed. This adds an extra option
`dri-search-path`, which will default to the value of
`dri-drivers-path`, like autotools does.

v2: - Split input list before joining.
v3: - use : instead of ; as the delimiter. The autotools help string
      incorrectly says ; but the code uses :
v4: - Take list in pre : delimited form (Ilia)
    - Ensure that the dri-search-path is absolute when using
      dri_drivers_path

Fixes: db9788420d4bc7b4 ("meson: Add support for configuring dri drivers directory.")
Reported-by: Ilia Mirkin <imirkin@alum.mit.edu>
Signed-off-by: Dylan Baker <dylan.c.baker@intel.com>
Reviewed-by: Eric Anholt <eric@anholt.net> (v2)
Reviewed-by: Eric Engestrom <eric@engestrom.ch> (v3)
6 years agoradeonsi: use pknorm_i16/u16 and pk_i16/u16 LLVM intrinsics
Marek Olšák [Tue, 2 Jan 2018 03:34:53 +0000 (04:34 +0100)]
radeonsi: use pknorm_i16/u16 and pk_i16/u16 LLVM intrinsics

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agotravis: add osx autotools build
Jon Turney [Thu, 18 Jan 2018 13:05:06 +0000 (13:05 +0000)]
travis: add osx autotools build

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agotravis: pip -> pip2
Jon Turney [Thu, 1 Feb 2018 15:23:49 +0000 (15:23 +0000)]
travis: pip -> pip2

On travis, for OSX, python2 from homebrew is pre-installed. per [1]:

 python points to the macOS system Python (with no manual PATH modification)
 python2 points to Homebrew’s Python 2.7.x (if installed)
 python3 points to Homebrew’s Python 3.x (if installed)
 pip doesn't exist
 pip2 points to Homebrew’s Python 2.7.x’s pip (if installed)
 pip3 points to Homebrew’s Python 3.x’s pip (if installed)

We will end up using 'python2' for building mesa.

Just use 'pip2' instead of 'pip', as that seems to work for all platforms on
travis.

[1] https://docs.brew.sh/Homebrew-and-Python.html

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agotravis: conditionalize building of prerequisites on if OS=linux
Jon Turney [Thu, 1 Feb 2018 15:19:08 +0000 (15:19 +0000)]
travis: conditionalize building of prerequisites on if OS=linux

Use a '|' YAML literal block to avoid the convoluted syntax needed to put
the entire conditional on a single line.

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agoglx/test: fix building for osx
Jon Turney [Thu, 25 Jan 2018 17:34:54 +0000 (17:34 +0000)]
glx/test: fix building for osx

An additional stub for applegl_create_context() is needed
Cannot test indirect API as it's not built on osx, currently

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agoi965: check if upload is 0 explicitely, when downsizing a format
Andres Gomez [Thu, 1 Feb 2018 15:15:14 +0000 (17:15 +0200)]
i965: check if upload is 0 explicitely, when downsizing a format

downsize_format_if_needed takes an integer as number of uploads
parameter. Hence, let's do an integer comparation instead of a boolean
check, since that is confusing.

Since we are at it, fix a couple of wrongly tabbed indents.

Cc: Alejandro Piñeiro <apinheiro@igalia.com>
Cc: Kenneth Graunke <kenneth@whitecape.org>
Signed-off-by: Andres Gomez <agomez@igalia.com>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
6 years agomesa: don't flag _NEW_COLOR for KHR adv.blend if prog constant doesn't change
Marek Olšák [Sun, 7 Jan 2018 17:27:40 +0000 (18:27 +0100)]
mesa: don't flag _NEW_COLOR for KHR adv.blend if prog constant doesn't change

This only affects drivers that set DriverFlags.NewBlend.

v2: - fix typo advanded -> advanced
    - return "enum gl_advanced_blend_mode" from
      _mesa_get_advanced_blend_sh_constant
    - don't call FLUSH_VERTICES twice

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
6 years agoac/nir: replace SI.buffer.load.dword with amdgcn.buffer.load
Samuel Pitoiset [Thu, 1 Feb 2018 15:37:15 +0000 (16:37 +0100)]
ac/nir: replace SI.buffer.load.dword with amdgcn.buffer.load

The old one generates useless instructions in there, found while
comparing geometry shaders between RadeonSI and RADV.

This improves all Vulkan demos that use geometry shaders, +4%
for deferredshadows, +9% for viewportarray, +7% for
geometryshader on Polaris10.

This seems to also improve DOW3 a little bit (+1%).

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agor600/eg: add crap indirect compute support.
Dave Airlie [Tue, 30 Jan 2018 02:21:59 +0000 (12:21 +1000)]
r600/eg: add crap indirect compute support.

I think the cp packets can be made work, but I think it might
need a kernel change, so for now just do the worst thing.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
6 years agoi965: Call prepare_external after implicit window-system MSAA resolves
Jason Ekstrand [Thu, 1 Feb 2018 01:31:39 +0000 (17:31 -0800)]
i965: Call prepare_external after implicit window-system MSAA resolves

This fixes some rendering corruption in a couple of Android apps that
use window-system MSAA.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104741
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
6 years agor600: don't do stack workarounds for hemlock
Roland Scheidegger [Tue, 30 Jan 2018 04:48:27 +0000 (05:48 +0100)]
r600: don't do stack workarounds for hemlock

By the looks of it it seems hemlock is treated separately to cypress, but
certainly it won't need the stack workarounds cedar/redwood (and
seemingly every other eg chip except cypress/juniper) need.
(Discovered by accident.)

Acked-by: Alex Deucher <alexander.deucher@amd.com>
6 years agor600: initial attempt at gl_HelperInvocation (v3)
Dave Airlie [Wed, 31 Jan 2018 04:28:26 +0000 (14:28 +1000)]
r600: initial attempt at gl_HelperInvocation (v3)

This passes the CTS and piglit tests.

This also disable sb for helper invocations until it doesn't
mess up the VPM flags.

Thanks to Ilia and Glenn for advice, and Roland for working
out the working evergreen path.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
6 years agoradv: Don't expose VK_KHX_multiview on android.
Bas Nieuwenhuizen [Wed, 31 Jan 2018 11:31:30 +0000 (12:31 +0100)]
radv: Don't expose VK_KHX_multiview on android.

deqp does not allow any KHX extensions, and since deqp is included
in android-cts, android does not allow any khx extensions.

So disable VK_KHX_multiview on android.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
CC: 18.0 <mesa-stable@lists.freedesktop.org>
6 years agovbo: Simplify input array distribution for dlist type draws.
Mathias Fröhlich [Sat, 27 Jan 2018 15:07:22 +0000 (16:07 +0100)]
vbo: Simplify input array distribution for dlist type draws.

Using the newly introduced VAO array maps, we can
simplify vbo_bind_vertex_list.

Signed-off-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Reviewed-by: Brian Paul <brianp@vmware.com>
6 years agovbo: Simplify input array distribution for imm type draws.
Mathias Fröhlich [Sat, 27 Jan 2018 15:07:22 +0000 (16:07 +0100)]
vbo: Simplify input array distribution for imm type draws.

Using the newly introduced VAO array maps, we can
simplify vbo_exec_bind_arrays.

Signed-off-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Reviewed-by: Brian Paul <brianp@vmware.com>
6 years agovbo: Simplify input array distribution for array type draws.
Mathias Fröhlich [Sat, 27 Jan 2018 15:07:22 +0000 (16:07 +0100)]
vbo: Simplify input array distribution for array type draws.

Using the newly introduced VAO state variable, we can
simplify recalculate_input_bindings.

Signed-off-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Reviewed-by: Brian Paul <brianp@vmware.com>
6 years agovbo: Use static const VERT_ATTRIB->VBO_ATTRIB maps.
Mathias Fröhlich [Sat, 27 Jan 2018 15:07:22 +0000 (16:07 +0100)]
vbo: Use static const VERT_ATTRIB->VBO_ATTRIB maps.

Instead of each context having its own map instance for
this purpose, use a global static const map.

v2: s,unsigned char,GLubyte,g
    s,_VP_MODE_MAX,VP_MODE_MAX,g
    Change comment style.

Signed-off-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Reviewed-by: Brian Paul <brianp@vmware.com>
6 years agomesa: Track position/generic0 aliasing in the VAO.
Mathias Fröhlich [Sat, 27 Jan 2018 15:07:22 +0000 (16:07 +0100)]
mesa: Track position/generic0 aliasing in the VAO.

Since the first material attribute no longer aliases with
the generic0 attribute, only aliasing between generic0 and
position is left and entirely dependent on the enabled
state of the VAO. So introduce a gl_attribute_map_mode
in the VAO that is used to track how the position
and the generic 0 attribute alias.
Provide a static const array that can be used to
map from vertex program input indices to VERT_ATTRIB_*
indices. The outer dimension of the array is meant to
be indexed directly by the new VAO member variable.
Also provide methods on the VAO to convert bitmasks of
VERT_BIT's from the VAO numbering to the vertex processing
inputs numbering.

v2: s,unsigned char,GLubyte,g
    s,_ATTRIBUTE_MAP_MODE_MAX,ATTRIBUTE_MAP_MODE_MAX,g
    Change comment style, add comments.

Signed-off-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Reviewed-by: Brian Paul <brianp@vmware.com>
6 years agomesa: Put materials at the end of the generic block.
Mathias Fröhlich [Sat, 27 Jan 2018 15:07:22 +0000 (16:07 +0100)]
mesa: Put materials at the end of the generic block.

The materials are now moved to the end of the
generic attributes block to the range 4-15.

Before, the way the position and generic 0 attribute
is handled was dependent on the presence and kind of
the currently attached vertex program. With this
change the way the position attribute and the generic 0
attribute is treated only depends on the enabled
flag of those two arrays.
This will later help to untangle the update dependencies
between enabled arrays and shader inputs.

v2: s,VERT_ATTRIB_MAT_OFFSET,VERT_ATTRIB_MAT0,g

Signed-off-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Reviewed-by: Brian Paul <brianp@vmware.com>
6 years agomesa: Use defines for the aliased material array attributes.
Mathias Fröhlich [Sat, 27 Jan 2018 15:07:22 +0000 (16:07 +0100)]
mesa: Use defines for the aliased material array attributes.

Instead of just assuming that the material attributes
just overlap with the generic attributes 0-12, give
them symbolic defines so that we can easier move them
to an other range.

Signed-off-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Reviewed-by: Brian Paul <brianp@vmware.com>
6 years agovbo: Correctly handle attribute offsets in dlist draw.
Mathias Fröhlich [Sat, 27 Jan 2018 15:07:22 +0000 (16:07 +0100)]
vbo: Correctly handle attribute offsets in dlist draw.

When executing a display list draw, for the offset
list to be correct, the offset computation needs to
accumulate all attribute size values in order.
Specifically, if we are shuffling around the position
and generic0 attributes, we may violate the order or
if we do not walk the generic vbo attributes we may
skip some of the attributes.
Even if this is an unlikely usecase we can fix this use
case by precomputing the offsets on the full attribute list
and store the full offset list in the display list node.

v2: Formatting fix
v3: Rebase

Signed-off-by: Mathias Fröhlich <Mathias.Froehlich@web.de>
Reviewed-by: Brian Paul <brianp@vmware.com>
6 years agogallivm/llvmpipe: add const qualifiers on sampler variables
Brian Paul [Thu, 1 Feb 2018 20:17:08 +0000 (13:17 -0700)]
gallivm/llvmpipe: add const qualifiers on sampler variables

Once a lp_build_sampler_soa or lp_build_sampler_aos object is created,
it should never be modified.  Found by inspection.

Reviewed-by: Roland Scheidegger <sroland@vmware.com>
6 years agovbo: change an argument in vbo_draw_indirect_prims()
Brian Paul [Wed, 31 Jan 2018 23:19:07 +0000 (16:19 -0700)]
vbo: change an argument in vbo_draw_indirect_prims()

In vbo_draw_indirect_prims() pass the 'indirect_data' argument to
vbo->draw_prims().  All the callers are passing ctx->DrawIndirectBuffer
so this should be no functional change.  Add a (temporary) assertion to
be sure.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Mathias Fröhlich <mathias.froehlich@web.de>
6 years agovbo: add comments on the VBO draw function typedefs
Brian Paul [Wed, 31 Jan 2018 23:15:53 +0000 (16:15 -0700)]
vbo: add comments on the VBO draw function typedefs

And rename indirect_params -> indirect_draw_count_buffer and
indirect_params_offset -> indirect_draw_count_offset to be more
specific.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Mathias Fröhlich <mathias.froehlich@web.de>
6 years agovbo: s/drawcount/drawcount_offset
Brian Paul [Wed, 31 Jan 2018 23:11:12 +0000 (16:11 -0700)]
vbo: s/drawcount/drawcount_offset

This parameter (from the glMultiDrawArraysIndirectCountARB function)
is poorly named.  It's an offset into the buffer which contains the
number of primitives to draw.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Mathias Fröhlich <mathias.froehlich@web.de>
6 years agovbo: use vbo local var for draw call in vbo_save_playback_vertex_list()
Brian Paul [Wed, 31 Jan 2018 21:05:46 +0000 (14:05 -0700)]
vbo: use vbo local var for draw call in vbo_save_playback_vertex_list()

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Reviewed-by: Mathias Fröhlich <mathias.froehlich@web.de>
6 years agosvga: remove unneeded #includes in svga_pipe_draw.c
Brian Paul [Thu, 1 Feb 2018 03:36:35 +0000 (20:36 -0700)]
svga: remove unneeded #includes in svga_pipe_draw.c

Reviewed-by: Neha Bhende <bhenden@vmware.com>
6 years agosvga: whitespace/formatting fixes in svga_pipe_draw.c
Brian Paul [Thu, 1 Feb 2018 03:33:29 +0000 (20:33 -0700)]
svga: whitespace/formatting fixes in svga_pipe_draw.c

Reviewed-by: Neha Bhende <bhenden@vmware.com>
6 years agosvga: clean up retry_draw_range_elements(), retry_draw_arrays()
Brian Paul [Thu, 1 Feb 2018 03:18:52 +0000 (20:18 -0700)]
svga: clean up retry_draw_range_elements(), retry_draw_arrays()

Get rid of a bunch of goto spaghetti.  Remove unneeded do_retry parameter.
No Piglit changes.  Also tested w/ Google Earth and other apps.

Reviewed-by: Neha Bhende <bhenden@vmware.com>
6 years agosvga: remove unused min/max_index params to draw_vgpu10()
Brian Paul [Wed, 31 Jan 2018 23:10:39 +0000 (16:10 -0700)]
svga: remove unused min/max_index params to draw_vgpu10()

Reviewed-by: Neha Bhende <bhenden@vmware.com>
6 years agobroadcom/vc5: Fix image_h setup for both loads and stores.
Eric Anholt [Wed, 24 Jan 2018 01:13:53 +0000 (12:13 +1100)]
broadcom/vc5: Fix image_h setup for both loads and stores.

The image_h for the tiling algorithm needs to be the padded-to-a-uifblock
height of the level, not the unpadded height or the height of level 0.
Fixes some cases of KHR-GLES3.texture_repeat_mode.* and
depthstencil-render-miplevels.

6 years agobroadcom/vc5: Add appropriate height padding for bank conflicts.
Eric Anholt [Sun, 21 Jan 2018 02:25:10 +0000 (10:25 +0800)]
broadcom/vc5: Add appropriate height padding for bank conflicts.

I thought I didn't need this because I was doing level-0-always-UIF and
that the pad there would propagate down, but it turns out that for level 1
the padding ends up being chosen by the HW.  This brings us closer to
being able to turn on UIF XOR for increased performance, as well.

6 years agobroadcom/vc5: Simplify separate stencil surface setup.
Eric Anholt [Sun, 21 Jan 2018 00:44:53 +0000 (16:44 -0800)]
broadcom/vc5: Simplify separate stencil surface setup.

If we just make another gallium surface for the separate stencil, it's a
lot easier to keep track of which set of fields we're using in RCL setup.

This also incidentally fixes a little bug in setting up the surface's
padded height for separate stencil when the UIF-ness changes at different
levels of Z versus stencil.

6 years agobroadcom/vc5: Rename the UIFCFG register in the UAPI.
Eric Anholt [Sat, 20 Jan 2018 23:57:08 +0000 (15:57 -0800)]
broadcom/vc5: Rename the UIFCFG register in the UAPI.

This matches the naming of the other hub regs we get, and I don't know for
sure if UIFCFG will be the same register between the hub and the cores on
all versions.

6 years agobroadcom/vc5: Fix a segfault on mix of booleans.
Eric Anholt [Fri, 19 Jan 2018 17:06:26 +0000 (09:06 -0800)]
broadcom/vc5: Fix a segfault on mix of booleans.

We don't have a src1 to look up if the compare instruction is "i2b".

6 years agobroadcom/vc5: Skip over missing color buffers for a couple of checks.
Eric Anholt [Sat, 20 Jan 2018 18:19:20 +0000 (10:19 -0800)]
broadcom/vc5: Skip over missing color buffers for a couple of checks.

Fixes crashes in piglit alpha-to-coverage-no-draw-buffer-zero 2

6 years agobroadcom/vc5: Add the missing PIPE_CAP_FENCE_SIGNAL.
Eric Anholt [Thu, 1 Feb 2018 01:38:22 +0000 (17:38 -0800)]
broadcom/vc5: Add the missing PIPE_CAP_FENCE_SIGNAL.

6 years agomesa: fix query of GL_TEXTURE_COMPRESSION_HINT_ARB
Baldur Karlsson [Thu, 1 Feb 2018 18:31:56 +0000 (11:31 -0700)]
mesa: fix query of GL_TEXTURE_COMPRESSION_HINT_ARB

Fixes: f96a69f916a ("mesa: replace GLenum with GLenum16 in common
structures (v4)")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=104908
Reviewed-by: Brian Paul <brianp@vmware.com>
6 years agorenderonly: fix dumb BO allocation for non 32bpp formats
Lucas Stach [Tue, 30 Jan 2018 14:11:35 +0000 (15:11 +0100)]
renderonly: fix dumb BO allocation for non 32bpp formats

Take into account the resource format, instead of applying a hardcoded
32bpp. This not only over-allocates 16bpp formats, but also results in
a wrong stride being filled into the handle.

Fixes: 848b49b288f ("gallium: add renderonly library")
CC: <mesa-stable@lists.freedesktop.org>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Reviewed-by: Daniel Stone <daniels@collabora.com>
6 years agointel/decoder: Fix control / evaluation label mixup.
Kenneth Graunke [Thu, 1 Feb 2018 17:43:20 +0000 (09:43 -0800)]
intel/decoder: Fix control / evaluation label mixup.

Trivial.  DS is TES, HS is TCS.

6 years agoi965: Bump official kernel requirement to Linux v3.9.
Kenneth Graunke [Wed, 31 Jan 2018 15:03:17 +0000 (07:03 -0800)]
i965: Bump official kernel requirement to Linux v3.9.

In commit 3f353342a6b6744773c26ed66b12afed42bd57af (present in 17.3.0)
we started unconditionally using I915_EXEC_NO_RELOC, which was
introduced in Linux v3.9.  ChromeOS kernel 3.8 has backported this,
so it should work too.

Running on older kernels would likely result in every single batch
being rejected by the kernel, which is pretty catastrophic.  Yet, it
appears that nobody noticed.  So, let's just bump the official
requirement and move forward ever so slowly.

Fixes: 3f353342a6b ("i965: Use I915_EXEC_NO_RELOC")
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Emil Velikov <emil.velikov@collabora.com>
6 years agomeson: don't install windows headers on non-windows platforms
Marc Dietrich [Thu, 1 Feb 2018 12:27:28 +0000 (13:27 +0100)]
meson: don't install windows headers on non-windows platforms

Only dive into the windows subdir if windows platform is selected.

Signed-off-by: Marc Dietrich <marvin24@gmx.de>
Fixes: 5ef75cb02b2b4db5506b8 "meson: build src/glx/windows"
Reviewed-by: Eric Engestrom <eric.engestrom@imgtec.com>
6 years agoradeonsi: use ac_build_buffer_load_format for image buffer loads
Marek Olšák [Tue, 30 Jan 2018 18:40:43 +0000 (19:40 +0100)]
radeonsi: use ac_build_buffer_load_format for image buffer loads

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agoac/nir: use ac_build_buffer_load_format for image buffer loads
Marek Olšák [Tue, 30 Jan 2018 18:40:43 +0000 (19:40 +0100)]
ac/nir: use ac_build_buffer_load_format for image buffer loads

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agoac: add glc parameter to ac_build_buffer_load_format
Marek Olšák [Tue, 30 Jan 2018 18:24:07 +0000 (19:24 +0100)]
ac: add glc parameter to ac_build_buffer_load_format

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agoradeonsi: load the right number of components for VS inputs and TBOs
Marek Olšák [Tue, 30 Jan 2018 17:34:25 +0000 (18:34 +0100)]
radeonsi: load the right number of components for VS inputs and TBOs

The supported counts are 1, 2, 4. (3=4)

The following snippet loads float, vec2, vec3, and vec4:

Before:
    buffer_load_format_x v9, v4, s[0:3], 0 idxen          ; E0002000 80000904
    buffer_load_format_xyzw v[0:3], v5, s[8:11], 0 idxen  ; E00C2000 80020005
    s_waitcnt vmcnt(0)                                    ; BF8C0F70
    buffer_load_format_xyzw v[2:5], v6, s[12:15], 0 idxen ; E00C2000 80030206
    s_waitcnt vmcnt(0)                                    ; BF8C0F70
    buffer_load_format_xyzw v[5:8], v7, s[4:7], 0 idxen   ; E00C2000 80010507

After:
    buffer_load_format_x v10, v4, s[0:3], 0 idxen         ; E0002000 80000A04
    buffer_load_format_xy v[8:9], v5, s[8:11], 0 idxen    ; E0042000 80020805
    buffer_load_format_xyzw v[0:3], v6, s[12:15], 0 idxen ; E00C2000 80030006
    s_waitcnt vmcnt(0)                                    ; BF8C0F70
    buffer_load_format_xyzw v[3:6], v7, s[4:7], 0 idxen   ; E00C2000 80010307

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agoradeonsi: remove unused si_shader_context members
Marek Olšák [Tue, 30 Jan 2018 16:58:14 +0000 (17:58 +0100)]
radeonsi: remove unused si_shader_context members

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agoglx/apple: locate dispatch table functions to wrap by name
Jon Turney [Tue, 16 Jan 2018 16:26:57 +0000 (16:26 +0000)]
glx/apple: locate dispatch table functions to wrap by name

Avoid reaching into the dispatch table internals (and thus having to deal
with the complexities of remap etc.) by identifying functions to wrap by
name.

See:
https://lists.freedesktop.org/archives/mesa-dev/2015-June/086721.html et seq.
https://bugs.freedesktop.org/show_bug.cgi?id=90311

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agoglx/apple: include util/debug.h for env_var_as_boolean prototype
Jon Turney [Sat, 2 Dec 2017 17:05:43 +0000 (17:05 +0000)]
glx/apple: include util/debug.h for env_var_as_boolean prototype

mesa/src/glx/glxcmds.c:1295:21: error: implicit declaration of function 'env_var_as_boolean' is invalid in C99 [-Werror,-Wimplicit-function-declaration]
mesa/src/glx/apple/apple_visual.c:85:28: error: implicit declaration of function 'env_var_as_boolean' is invalid in C99 [-Werror,-Wimplicit-function-declaration]

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agoosx: ld doesn't support --build-id
Jon Turney [Sun, 3 Dec 2017 21:58:12 +0000 (21:58 +0000)]
osx: ld doesn't support --build-id

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agoconfigure: Default to gbm=no on osx
Jon Turney [Tue, 16 Jan 2018 23:27:43 +0000 (23:27 +0000)]
configure: Default to gbm=no on osx

Signed-off-by: Jon Turney <jon.turney@dronecode.org.uk>
Reviewed-by: Emil Velikov <emil.velikov@collabora.com>
6 years agomesa: remove usage of alloca in externalobjects.c v4
Andres Rodriguez [Wed, 31 Jan 2018 17:22:41 +0000 (12:22 -0500)]
mesa: remove usage of alloca in externalobjects.c v4

Don't want an overly large numBufferBarriers/numTextureBarriers to blow
up the stack.

v2: handle malloc errors
v3: fix patch
v4: initialize texObjs/bufObjs

Suggested-by: Emil Velikov <emil.velikov@collabora.com>
Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
6 years agoradv: do not insert shaders in cache when it's disabled
Samuel Pitoiset [Wed, 31 Jan 2018 14:53:37 +0000 (15:53 +0100)]
radv: do not insert shaders in cache when it's disabled

When the application doesn't provide its own pipeline cache,
the driver uses a in-memory cache but it shouldn't insert any
entries when the cache is explicitely disabled by the user.

Found while running my experimental pipeline-db tool with a
ton of shaders, the memory footprint was just huge, and sometimes
the process was even killed...

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>