platform/upstream/mesa.git
6 years agoradv: do not support blitting surfaces for R32G32B32 formats
Samuel Pitoiset [Fri, 12 Oct 2018 12:04:39 +0000 (14:04 +0200)]
radv: do not support blitting surfaces for R32G32B32 formats

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108113
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoscons: Allow building with custom MSVC_USE_SCRIPT script.
Jose Fonseca [Tue, 9 Oct 2018 15:46:21 +0000 (16:46 +0100)]
scons: Allow building with custom MSVC_USE_SCRIPT script.

SCons MSVC support relies on vcvarsall.bat to extract the PATH, CPP
includes, library paths, etc.

And SCons also has an build env var named MSVC_USE_SCRIPT which one can
use to point to alternative vcvarsall.bat script.

This change exposes this MSVC_USE_SCRIPT build env variable as a SCons
command line variable.  This will enable using MSVC outside Program
Files (e.g, network shares, etc.)

This change also links advapi32 library, necessary for the Windows
Registry API used by WGL state tracker, avoiding missing symbols.

Reviewed-by: Brian Paul <brianp@vmware.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
6 years agoradv: emit the GLC bit for SSBO loads/stores when needed
Samuel Pitoiset [Wed, 3 Oct 2018 21:06:34 +0000 (23:06 +0200)]
radv: emit the GLC bit for SSBO loads/stores when needed

This fixes some new memory model tests:
dEQP-VK.memory_model.message_passing.core11.u32.coherent.fence_fence.atomicwrite.device.*

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108112
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agospirv/nir: handle memory access qualifiers for SSBO loads/stores
Samuel Pitoiset [Wed, 10 Oct 2018 08:42:19 +0000 (10:42 +0200)]
spirv/nir: handle memory access qualifiers for SSBO loads/stores

v2: - change how the access qualifiers are accumulated
v3: - duplicate members in struct_member_decoration_cb()
    - handle access qualifiers on variables
    - remove access qualifiers handling in _vtn_variable_load_store()
    - fix setting access qualifiers on type->array_element

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net
6 years agoanv/android: we need git_sha1.h in include paths
Tapani Pälli [Wed, 3 Oct 2018 10:46:42 +0000 (13:46 +0300)]
anv/android: we need git_sha1.h in include paths

Fixes: e4538b9 "anv: Implement VK_KHR_driver_properties"
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Acked-by: Eric Engestrom <eric.engestrom@intel.com>
6 years agoanv: Clear WM_HZ_OP overrides in init_device_state
Nanley Chery [Thu, 11 Oct 2018 23:31:08 +0000 (16:31 -0700)]
anv: Clear WM_HZ_OP overrides in init_device_state

This is basically a port of commit,
3ade766684933ac84e41634429fb693f85353c11
("i965: Disable 3DSTATE_WM_HZ_OP fields.")

The BDW+ docs describe how to use the 3DSTATE_WM_HZ_OP instruction in
the section titled, "Optimized Depth Buffer Clear and/or Stencil Buffer
Clear." It mentions that the packet overrides GPU state for the clear
operation and needs to be reset to 0s to clear the overrides. Depending
on the kernel, we may not get a context with the GPU state for this
packet zeroed. Do it ourselves just in case.

Prevents a number of GPU hangs when running crucible on ICL. I tried to
get the exact number of hangs that occurs without this patch, but was
unsuccessful. The test machine became unresponsive before completing the
full run.

Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
6 years agoi965/gen10+: Initialize new fields in STATE_BASE_ADDRESS
Jordan Justen [Wed, 10 Oct 2018 09:31:00 +0000 (02:31 -0700)]
i965/gen10+: Initialize new fields in STATE_BASE_ADDRESS

Ref: 263b584d5e4 "i965/skl: Emit extra zeros in STATE_BASE_ADDRESS on Skylake."
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
6 years agoanv/gen9+: Initialize new fields in STATE_BASE_ADDRESS
Jordan Justen [Wed, 28 Mar 2018 08:29:18 +0000 (01:29 -0700)]
anv/gen9+: Initialize new fields in STATE_BASE_ADDRESS

Ref: 263b584d5e4 "i965/skl: Emit extra zeros in STATE_BASE_ADDRESS on Skylake."
Signed-off-by: Jordan Justen <jordan.l.justen@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
6 years agonir: Add a bunch of b2[if] optimizations
Jason Ekstrand [Thu, 11 Oct 2018 03:36:52 +0000 (22:36 -0500)]
nir: Add a bunch of b2[if] optimizations

The b2f and b2i conversions always produce zero or one which are both
representable in every type and size.  Since b2i and b2f support all bit
sizes, we can just get rid of the conversion opcode.

total instructions in shared programs: 15089335 -> 15084368 (-0.03%)
instructions in affected programs: 212564 -> 207597 (-2.34%)
helped: 896
HURT: 0

total cycles in shared programs: 369831123 -> 369826267 (<.01%)
cycles in affected programs: 2008647 -> 2003791 (-0.24%)
helped: 693
HURT: 216

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
6 years agointel/vec4: Fix nir_op_b2[fi] with 64-bit result
Jason Ekstrand [Thu, 11 Oct 2018 03:04:17 +0000 (22:04 -0500)]
intel/vec4: Fix nir_op_b2[fi] with 64-bit result

This is valid NIR but you can't actually hit this case today.  GLSL IR
doesn't have a bool to double opcode; it does f2d(b2f(x)).  In SPIR-V we
don't have any to/from bool conversion opcodes at all.  However, the
next commit will make us start generating it so we should be ready.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
6 years agointel/fs: Fix nir_op_b2[fi] with 64-bit result on Gen8 LP and Gen9 LP
Jason Ekstrand [Wed, 10 Oct 2018 22:17:11 +0000 (15:17 -0700)]
intel/fs: Fix nir_op_b2[fi] with 64-bit result on Gen8 LP and Gen9 LP

Several of the Atom GPUs have additional restrictions on alignment when
moving < 64-bit source to a 64-bit destination.  All of the nir_op_*2*64
code generation paths respected this, but nir_op_b2[fi] did not.

Previous to commit a68dd47b911 it was not possible to generate such an
instruction from the GLSL path.  It may have been possible from SPIR-V,
but it's not clear.  The aforementioned patch converts a 64-bit
nir_op_fsign into a sequence of operations including a nir_op_b2f with a
64-bit result.  This "just works" everywhere except these Atom parts.

This problem was not detected during normal CI testing because the Atom
parts are not included in developer builds.

v2 (idr): Make the patch compile, and make some cosmetic changes.  Add a
commit message.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108319
Fixes: a68dd47b911 "nir/algebraic: Simplify fsat of fsign"
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
6 years agoegl: Use correct shared libraries suffix on macOS.
Vinson Lee [Wed, 3 Oct 2018 21:56:26 +0000 (14:56 -0700)]
egl: Use correct shared libraries suffix on macOS.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
6 years agomesa: Fix pack_uint_Z_FLOAT32()
Illia Iorin [Thu, 11 Oct 2018 15:06:18 +0000 (18:06 +0300)]
mesa: Fix pack_uint_Z_FLOAT32()

Fixed pack_uint_Z_FLOAT32 by casting row data to float instead uint.
Remove code duplicate function pack_uint_Z_FLOAT32_X24S8.
Edited case in "_mesa_get_pack_uint_z_func".
Now it looks like "_mesa_get_pack_float_z_func".
Remove _mesa_problem call, which was added for debuging this issue.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=91433
Signed-off-by: Illia Iorin <illia.iorin@globallogic.com>
Reviewed-by: Nanley Chery <nanley.g.chery@intel.com>
6 years agointel: Introducing Whiskey Lake platform
Rodrigo Vivi [Thu, 30 Aug 2018 21:39:57 +0000 (14:39 -0700)]
intel: Introducing Whiskey Lake platform

Whiskey Lake uses the same gen graphics as Coffe Lake, including some
ids that were previously marked as reserved on Coffe Lake, but that
now are moved to WHL page.

This follows the ids and approach used on kernel's commit
b9be78531d27 ("drm/i915/whl: Introducing Whiskey Lake platform")
and commit c1c8f6fa731b ("drm/i915: Redefine some Whiskey Lake SKUs")

v2: Lionel noticed that GT{1,2,3} on kernel wasn't following
spec when looking to number of EUs, so kernel has been updated.

Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: José Roberto de Souza <jose.souza@intel.com>
Cc: Anuj Phogat <anuj.phogat@gmail.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agost/va: use provided sizes and coords for vlVaGetImage
Boyuan Zhang [Wed, 10 Oct 2018 19:08:44 +0000 (15:08 -0400)]
st/va: use provided sizes and coords for vlVaGetImage

vlVaGetImage should respect the width, height, and coordinates x and y that
passed in. Therefore, pipe_box should be created with the passed in values
instead of surface width/height.

v2: add input size check, return error when size out of bounds
v3: fix the size check for vaimage
v4: add size adjustment for x and y coordinates

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Cc: "18.2" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
Acked-by: Christian König <christian.koenig@amd.com>
6 years agoradv: implement clear operations for R32G32B32
Samuel Pitoiset [Tue, 9 Oct 2018 10:26:42 +0000 (12:26 +0200)]
radv: implement clear operations for R32G32B32

This fixes crashes for some CTS:
dEQP-VK.api.copy_and_blit.core.blit_image.all_formats.color.*.linear_*_*
dEQP-VK.api.copy_and_blit.core.blit_image.all_formats.color.*.*_linear_*

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108113
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoradv: disallow 3D images and mipmaps/layers for R32G32B32 linear formats
Samuel Pitoiset [Mon, 8 Oct 2018 12:40:17 +0000 (14:40 +0200)]
radv: disallow 3D images and mipmaps/layers for R32G32B32 linear formats

R32G32B32 are weird formats and we are only going to support
some basic operations for now.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoradv: add a workaround for a VGT hang with prim restart and strips
Samuel Pitoiset [Wed, 10 Oct 2018 12:04:42 +0000 (14:04 +0200)]
radv: add a workaround for a VGT hang with prim restart and strips

Otherwise, Yakuza and The Evil Within hang the GPU with DXVK.
This apparently only works on Polaris.

Suggested by Marek.

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoglsl: remove redundant es_shader checks
Timothy Arceri [Thu, 11 Oct 2018 00:25:08 +0000 (11:25 +1100)]
glsl: remove redundant es_shader checks

The es check is already covered by the is_version() check.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
6 years agost/glsl_to_tgsi: initialise need_uarl in contructor
Dave Airlie [Fri, 5 Oct 2018 00:52:51 +0000 (10:52 +1000)]
st/glsl_to_tgsi: initialise need_uarl in contructor

Found by coverity

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agoglspirv: drop pointless assert (size_t is unsigned)
Dave Airlie [Thu, 4 Oct 2018 23:20:09 +0000 (09:20 +1000)]
glspirv: drop pointless assert (size_t is unsigned)

Found by coverity

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
6 years agoradv: remove unsigned comparison against 0
Dave Airlie [Thu, 4 Oct 2018 23:30:44 +0000 (09:30 +1000)]
radv: remove unsigned comparison against 0

The value is always >= 0 here.

Found by coverity

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agoradv: remove dead code for master_fd close
Dave Airlie [Thu, 4 Oct 2018 23:24:31 +0000 (09:24 +1000)]
radv: remove dead code for master_fd close

We have never opened master_Fd at this point, so remove code to
close it.

Found by coverity.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agoradv: don't pass shader key by copy
Dave Airlie [Thu, 4 Oct 2018 23:17:45 +0000 (09:17 +1000)]
radv: don't pass shader key by copy

Coverity pointed out we were copying 168 bytes here unnecessarily.

Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
6 years agoanv: add missing unlock in error path.
Dave Airlie [Thu, 4 Oct 2018 23:56:19 +0000 (09:56 +1000)]
anv: add missing unlock in error path.

Not going to matter, but be consistent.

Found by coverity

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Fixes: caf41c78c (anv/allocator: Support softpin in the BO cache)

6 years agointel: Don't propagate conditional modifiers if a UD source is negated
Jason Ekstrand [Mon, 8 Oct 2018 17:22:35 +0000 (12:22 -0500)]
intel: Don't propagate conditional modifiers if a UD source is negated

This fixes a bug uncovered by my NIR integer division by constant
optimization series.

Fixes: 19f9cb72c8b "i965/fs: Add pass to propagate conditional..."
Fixes: 627f94b72e0 "i965/vec4: adding vec4_cmod_propagation..."
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
6 years agoutil: Add tests for fast integer division by constants
Jason Ekstrand [Thu, 13 Sep 2018 14:56:20 +0000 (09:56 -0500)]
util: Add tests for fast integer division by constants

While I generally trust rediculousfish to have done his homework, we've
made some adjustments to suit the needs of mesa and it'd be good to
test those.  Also, there's no better place than unit tests to clearly
document the different edge cases of the different methods.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agoutil: Add power-of-two divisor support to compute_fast_udiv_info
Marek Olšák [Sat, 6 Oct 2018 01:42:16 +0000 (20:42 -0500)]
util: Add power-of-two divisor support to compute_fast_udiv_info

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agoutil: Generalize fast integer division to be variable bit-width
Jason Ekstrand [Sat, 6 Oct 2018 01:29:31 +0000 (20:29 -0500)]
util: Generalize fast integer division to be variable bit-width

There's nothing inherently fixed-width in the code.  All that's required
to generalize it is to make everything internally 64-bit and pass
UINT_BITS in as a parameter to util_compute_fast_[us]div_info.  With
that, it can now handle 8, 16, 32, and 64-bit integer division by a
constant.

We also add support for division by 1 and by other powers of 2.  This is
useful if you want to divide by a uniform value in a shader where you
have the opportunity to adjust the uniform on the CPU before passing it
in.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agoutil: Add fast division helpers
Marek Olšák [Sat, 6 Oct 2018 01:28:40 +0000 (20:28 -0500)]
util: Add fast division helpers

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
6 years agoutil: import public domain code for integer division by a constant
Marek Olšák [Sun, 23 Sep 2018 16:57:51 +0000 (12:57 -0400)]
util: import public domain code for integer division by a constant

Compilers can use this to generate optimal code for integer division
by a constant.

Additionally, an unsigned division by a uniform that is constant but not
known at compile time can still be optimized by passing 2-4 division
factors to the shader as uniforms and executing one of the fast_udiv*
variants. The signed division algorithm doesn't have this capability.

Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agoutil: Add a simple big math library
Jason Ekstrand [Fri, 29 Dec 2017 21:30:32 +0000 (13:30 -0800)]
util: Add a simple big math library

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
6 years agomeson: Don't allow building EGL on Windows or MacOS
Dylan Baker [Mon, 1 Oct 2018 19:40:34 +0000 (12:40 -0700)]
meson: Don't allow building EGL on Windows or MacOS

Currently mesa only supports EGL on Unix like systems, cygwin, and
haiku. Meson should actually enforce this. This fixes the default build
on MacOS.

v2: - invert the condition, mark darwin and windows as not supported
      instead of trying to mark what is supported.
v3: - add missing )
v3: - Update comment to reflect condition change in v2

CC: 18.2 <mesa-stable@lists.freedesktop.org>
Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
6 years agoglsl: ignore trailing whitespace when define redefined
Timothy Arceri [Wed, 10 Oct 2018 00:03:47 +0000 (11:03 +1100)]
glsl: ignore trailing whitespace when define redefined

The Nvidia/AMD binary drivers allow this, as does GCC.

This fixes shader compilation issues in the latest update of
No Mans Sky.

Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
6 years agointel/compiler: Don't handle fsign.sat
Ian Romanick [Sun, 9 Sep 2018 21:06:43 +0000 (14:06 -0700)]
intel/compiler: Don't handle fsign.sat

No shader-db or CI changes on any Intel platform.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
6 years agonir/algebraic: Simplify fsat of fsign
Ian Romanick [Sun, 9 Sep 2018 20:40:02 +0000 (13:40 -0700)]
nir/algebraic: Simplify fsat of fsign

These allows us to not support fsign.sat in the Intel compiler backend,
and that will simplify some later changes.

No shader-db changes on any Intel platform.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
6 years agonir/algebraic: sign(x)*x*x is abs(x)*x
Ian Romanick [Mon, 10 Sep 2018 17:39:42 +0000 (10:39 -0700)]
nir/algebraic: sign(x)*x*x is abs(x)*x

shader-db results:

All Gen7+ platforms had similar results. (Skylake shown)
total instructions in shared programs: 15106023 -> 15105981 (<.01%)
instructions in affected programs: 300 -> 258 (-14.00%)
helped: 6
HURT: 0
helped stats (abs) min: 7 max: 7 x̄: 7.00 x̃: 7
helped stats (rel) min: 14.00% max: 14.00% x̄: 14.00% x̃: 14.00%
95% mean confidence interval for instructions value: -7.00 -7.00
95% mean confidence interval for instructions %-change: -14.00% -14.00%
Instructions are helped.

total cycles in shared programs: 566050327 -> 566050075 (<.01%)
cycles in affected programs: 2826 -> 2574 (-8.92%)
helped: 6
HURT: 0
helped stats (abs) min: 40 max: 44 x̄: 42.00 x̃: 42
helped stats (rel) min: 8.89% max: 8.94% x̄: 8.92% x̃: 8.92%
95% mean confidence interval for cycles value: -44.30 -39.70
95% mean confidence interval for cycles %-change: -8.95% -8.88%
Cycles are helped.

No changes on Gen6 or earlier.

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
6 years agonir: Add helper functions to get the instruction that generated a nir_src
Ian Romanick [Wed, 6 Jun 2018 02:19:39 +0000 (19:19 -0700)]
nir: Add helper functions to get the instruction that generated a nir_src

Signed-off-by: Ian Romanick <ian.d.romanick@intel.com>
Reviewed-by: Thomas Helland <thomashelland90@gmail.com>
6 years agosvga: change svga_destroy_shader_variant() to return void
Brian Paul [Tue, 9 Oct 2018 14:55:46 +0000 (08:55 -0600)]
svga: change svga_destroy_shader_variant() to return void

svga_destroy_shader_variant() itself flushes and retries the command
if there's a failure.  So no need for the callers to do it.  Other
callers of the function were already ignoring the return value.

This also fixes a corner-case double-free reported by Coverity
(and reported by Dave Airlie).

Tested with various OpenGL apps.

Reviewed-by: Charmaine Lee <charmainel@vmware.com>
6 years agomeson: Don't build glsl compiler tests unless OpenGL is enabled
Dylan Baker [Fri, 5 Oct 2018 16:37:32 +0000 (09:37 -0700)]
meson: Don't build glsl compiler tests unless OpenGL is enabled

Since there are no other users of the glsl compiler.

Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
6 years agomeson: Only build gallium state tracker tests with shared_glapi
Dylan Baker [Fri, 5 Oct 2018 16:36:05 +0000 (09:36 -0700)]
meson: Only build gallium state tracker tests with shared_glapi

This has always been a requirement, it's just somehow been missed in the
meson build.

Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
6 years agomeson: only build clapi tests when OpenGL is being built
Dylan Baker [Fri, 5 Oct 2018 16:33:59 +0000 (09:33 -0700)]
meson: only build clapi tests when OpenGL is being built

Otherwise building just vulkan (among other things) will build these
tests, pull in a bunch of stuff they shouldn't, and potentially fail to
compile.

Reviewed-by: Eric Engestrom <eric.engestrom@intel.com>
6 years agonvc0: fix blitting red to srgb8_alpha
Ilia Mirkin [Sun, 7 Oct 2018 21:45:07 +0000 (17:45 -0400)]
nvc0: fix blitting red to srgb8_alpha

For some reason the 2d engine can't handle this. Red formats get special
treatment there, so perhaps related.

Fixes dEQP-GLES3 tests of the form:

  dEQP-GLES3.functional.fbo.blit.conversion.r{8,16f,32f}_to_srgb8_alpha8

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Cc: mesa-stable@lists.freedesktop.org
6 years agonv50,nvc0: guard against zero-size blits
Ilia Mirkin [Sun, 7 Oct 2018 21:43:57 +0000 (17:43 -0400)]
nv50,nvc0: guard against zero-size blits

The current state tracker can generate these sometimes. Fixing this is
more involved, and due to some integer math we can generate
divisions-by-zero.

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Cc: mesa-stable@lists.freedesktop.org
6 years agonv50,nvc0: mark RGBX_UINT formats as renderable
Ilia Mirkin [Sun, 7 Oct 2018 18:57:17 +0000 (14:57 -0400)]
nv50,nvc0: mark RGBX_UINT formats as renderable

This helps st/mesa avoid some (apparently) buggy fallbacks. Specifically
the CopyTexSubImage fallback tries to read texture A as RGBA_FLOAT and
write back that data into the target format, which fails for integer
formats which have no appropriate logic to do the conversion.

Since integer formats don't blend, there's no harm in the fact that the
"A" component gets written anyways.

Fixes, among others:
  https://www.khronos.org/registry/webgl/sdk/tests/conformance2/textures/canvas/tex-2d-rgb8ui-rgb_integer-unsigned_byte.html

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Cc: mesa-stable@lists.freedesktop.org
6 years agoradv: add missing meson c++ visibility arguments
Eric Engestrom [Mon, 8 Oct 2018 15:23:25 +0000 (16:23 +0100)]
radv: add missing meson c++ visibility arguments

Fixes: 6f3aee40f90d725653b6 "radv: using tls to store llvm related info
                             and speed up compiles (v10)"
Cc: Dave Airlie <airlied@redhat.com>
Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
6 years agogbm: Add GBM_FORMAT_ARGB1555 support
Michel Dänzer [Fri, 5 Oct 2018 10:21:37 +0000 (12:21 +0200)]
gbm: Add GBM_FORMAT_ARGB1555 support

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agost/dri: Handle BGRA5551 format
Michel Dänzer [Fri, 5 Oct 2018 10:17:47 +0000 (12:17 +0200)]
st/dri: Handle BGRA5551 format

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
6 years agofreedreno/a5xx+a6xx: fix LRZ pitch alignment
Rob Clark [Thu, 4 Oct 2018 19:52:06 +0000 (15:52 -0400)]
freedreno/a5xx+a6xx: fix LRZ pitch alignment

Both RB_2D_DST_SIZE.PITCH (a6xx) and RB_MRT[n].PITCH (a5xx) need
alignment to 64.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: add LRZ support
Rob Clark [Thu, 4 Oct 2018 19:06:12 +0000 (15:06 -0400)]
freedreno/a6xx: add LRZ support

As with a5xx, hidden behind FD_MESA_DEBUG=lrz due to being paranoid
about z-fighting issues with some games (in particular, this was
observed with 0ad on a5xx.. but I think the proper solution to enable
this by default is to figure out how to do driver specific driconf
options).

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno: update generated headers
Rob Clark [Thu, 4 Oct 2018 19:05:39 +0000 (15:05 -0400)]
freedreno: update generated headers

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: add helper for various CP_EVENT_WRITE
Rob Clark [Fri, 5 Oct 2018 13:19:43 +0000 (09:19 -0400)]
freedreno/a6xx: add helper for various CP_EVENT_WRITE

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: remove unused fxns
Rob Clark [Fri, 5 Oct 2018 13:14:59 +0000 (09:14 -0400)]
freedreno/a6xx: remove unused fxns

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agofreedreno/a6xx: remove fd6_shader_stateobj
Rob Clark [Tue, 2 Oct 2018 18:31:20 +0000 (14:31 -0400)]
freedreno/a6xx: remove fd6_shader_stateobj

Earlier gen's already got this cleanup, but a6xx was still off on a
branch then.

Signed-off-by: Rob Clark <robdclark@gmail.com>
6 years agoglsl: fix array assignments of a swizzled vector
Ilia Mirkin [Fri, 5 Oct 2018 06:18:20 +0000 (02:18 -0400)]
glsl: fix array assignments of a swizzled vector

This happens in situations where we might do

  vec.wzyx[i] = ...

The swizzle would get effectively ignored because of the interaction
between how ir_assignment->set_lhs works and overwriting the write_mask.
There are two cases, one where i is a constant, and another where i is
variable. We have to be extra-careful in both cases.

Fixes the following WebGL test:

  https://www.khronos.org/registry/webgl/sdk/tests/conformance2/glsl3/vector-dynamic-indexing-swizzled-lvalue.html

And the new piglit tests:

  swizzled-writemask-indexing-nonconst.shader_test
  swizzled-writemask-indexing.shader_test

Signed-off-by: Ilia Mirkin <imirkin@alum.mit.edu>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Cc: mesa-stable@lists.freedesktop.org
6 years agoradv: tidy up radv_pipeline_init_multisample_state()
Samuel Pitoiset [Wed, 3 Oct 2018 14:09:25 +0000 (16:09 +0200)]
radv: tidy up radv_pipeline_init_multisample_state()

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoradv: always set PA_SC_MODE_CNTL_1.OUT_OF_ORDER_WATER_MARK
Samuel Pitoiset [Wed, 3 Oct 2018 14:09:24 +0000 (16:09 +0200)]
radv: always set PA_SC_MODE_CNTL_1.OUT_OF_ORDER_WATER_MARK

It has probably no effect without out of order rasterization
anyway.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoradv: set DB_EQAA.INCOHERENT_EQAA_READS
Samuel Pitoiset [Wed, 3 Oct 2018 14:09:23 +0000 (16:09 +0200)]
radv: set DB_EQAA.INCOHERENT_EQAA_READS

My attempt was to set this field instead of duplicating one.

Fixes: 6cfa321c39 ("radv: add potential missing fields for DB_EQAA")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
6 years agoi965: fallback RGBX to RGBA in glEGLImageTargetRenderbufferStorageOES
Chystiakov, Dmytro [Wed, 3 Oct 2018 09:52:52 +0000 (12:52 +0300)]
i965: fallback RGBX to RGBA in glEGLImageTargetRenderbufferStorageOES

In the same fashion as is done for glEGLImageTextureTarget2D.

v2: share the fallback which sets baseformat and internalformat correctly
    which makes both of the tests pass (Tapani)

Fixes android.hardware.nativehardware.cts.AHardwareBufferNativeTests:

   #SingleLayer_ColorTest_GpuColorOutputCpuRead_R8G8B8X8_UNORM
   #SingleLayer_ColorTest_GpuColorOutputIsRenderable_R8G8B8X8_UNORM

Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
6 years agoglsl: do not attempt assignment if operand type not parsed correctly
Tapani Pälli [Tue, 25 Sep 2018 14:04:40 +0000 (17:04 +0300)]
glsl: do not attempt assignment if operand type not parsed correctly

v2: check types of both operands (Ian)

Cc: mesa-stable@lists.freedesktop.org
Signed-off-by: Tapani Pälli <tapani.palli@intel.com>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108012

6 years agoutil/u_queue: add UTIL_QUEUE_INIT_SET_FULL_THREAD_AFFINITY
Marek Olšák [Mon, 1 Oct 2018 19:51:06 +0000 (15:51 -0400)]
util/u_queue: add UTIL_QUEUE_INIT_SET_FULL_THREAD_AFFINITY

Initial version discussed with Rob Clark under a different patch name.
This approach leaves his driver unaffected.

6 years agoradeonsi: fix a typo at CS_PARTIAL_FLUSH
Marek Olšák [Fri, 21 Sep 2018 07:37:16 +0000 (03:37 -0400)]
radeonsi: fix a typo at CS_PARTIAL_FLUSH

harmless

6 years agoac: add ac_build_round
Marek Olšák [Sat, 22 Sep 2018 01:30:09 +0000 (21:30 -0400)]
ac: add ac_build_round

6 years agoac: correct PKT3_COPY_DATA definitions
Marek Olšák [Fri, 21 Sep 2018 07:30:18 +0000 (03:30 -0400)]
ac: correct PKT3_COPY_DATA definitions

6 years agoac: simplify LLVM alloca helpers
Marek Olšák [Fri, 21 Sep 2018 07:27:06 +0000 (03:27 -0400)]
ac: simplify LLVM alloca helpers

6 years agoac: define all address spaces properly
Marek Olšák [Fri, 7 Sep 2018 22:44:54 +0000 (18:44 -0400)]
ac: define all address spaces properly

6 years agogallivm: Make it possible to disable some optimization shortcuts in release builds
Gert Wollny [Fri, 5 Oct 2018 13:08:51 +0000 (15:08 +0200)]
gallivm: Make it possible to disable some optimization shortcuts in release builds

For testing it is of interest that all tests of dEQP pass, e.g. to test
virglrenderer on a host only providing software rendering like in a CI.
Hence make it possible to disable certain optimizations that make tests fail.

While we are there also add some documentation to the flags to make it clear
that this is opt-out.

Setting the environment variable "GALLIVM_PERF=no_filter_hacks" can be used to make
the following tests pass in release mode:

  dEQP-GLES2.functional.texture.mipmap.2d.affine.*_linear_*
  dEQP-GLES2.functional.texture.mipmap.cube.generate.*
  dEQP-GLES2.functional.texture.vertex.2d.filtering.*_mipmap_linear_*
  dEQP-GLES2.functional.texture.vertex.2d.wrap.*

Related:
  https://bugs.freedesktop.org/show_bug.cgi?id=94957

v2: rename optimization disabling flag to 'safemath' and also move the
    nopt flag to the perf flags.

v3: rename flag "safemath" to "no_filter_hacks" since safemath is usually
    associated with floating point operations (Roland)

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
6 years agovirgl: Pass resource size and transfer offsets
Tomeu Vizoso [Thu, 4 Oct 2018 14:40:08 +0000 (16:40 +0200)]
virgl: Pass resource size and transfer offsets

Pass the size of a resource when creating it so a backing can be kept in
the other side.

Also pass the required offset to transfer commands.

This moves vtest closer to how virtio-gpu works, making it more useful
for testing.

v2: - Use new messages for creation and transfers, as changing the
      behavior of the existing messages would be messy given that we don't
      want to break compatibility with older servers.

v3: - Use correct strides: The resource corresponding to the output display
      might have a differnt line stride then the IOVs, so when reading back
      to this resource take the resource stride and the the IOV stride
      into account.

v4: Fix transfer size calculation (Andrey Simiklit)

v5: Add comment about transfer size value in the PUT commend (Gurchetan).
    Add a comment about the size correction for transfers for reading and
    writing the resource. Fixing this by correctly evaluating the size
    upfront will need some work also  on the virglrenderer side.

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com> (v2)
Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
6 years agovirgl, vtest: Correct the transfer size calculation
Gert Wollny [Thu, 4 Oct 2018 14:40:07 +0000 (16:40 +0200)]
virgl, vtest: Correct the transfer size calculation

The transfer size used in virglrenderer refers to uint32_t, so one
must add 3 and then divide by 4 instead of adding 3/4 which is a no-op
with integers.

Fixes: b3b82fe8ea virgl/vtest: add vtest driver

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
6 years agoutil: Make xmlconfig.c build on Solaris without d_type in dirent (v2)
Alan Coopersmith [Fri, 5 Oct 2018 23:34:35 +0000 (16:34 -0700)]
util: Make xmlconfig.c build on Solaris without d_type in dirent (v2)

v2: check for lstat() failing

Fixes: 04bdbbcab3c "xmlconfig: read more config files from drirc.d/"
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Reviewed-by: Roland Mainz <roland.mainz@nrubsig.org>
Reviewed-by: Ian Romanick <ian.d.romanick@intel.com>
6 years agoradeonsi:optimizing SET_CONTEXT_REG for shaders vgt_vertex_reuse
Sonny Jiang [Wed, 3 Oct 2018 15:53:14 +0000 (11:53 -0400)]
radeonsi:optimizing SET_CONTEXT_REG for shaders vgt_vertex_reuse

Signed-off-by: Sonny Jiang <sonny.jiang@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
6 years agoradeonsi:optimizing SET_CONTEXT_REG for shaders Tessellation
Sonny Jiang [Wed, 3 Oct 2018 15:53:13 +0000 (11:53 -0400)]
radeonsi:optimizing SET_CONTEXT_REG for shaders Tessellation

Signed-off-by: Sonny Jiang <sonny.jiang@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
6 years agoradeonsi:optimizing SET_CONTEXT_REG for shaders PS
Sonny Jiang [Wed, 3 Oct 2018 15:53:12 +0000 (11:53 -0400)]
radeonsi:optimizing SET_CONTEXT_REG for shaders PS

Signed-off-by: Sonny Jiang <sonny.jiang@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
6 years agoradeonsi:optimizing SET_CONTEXT_REG for shaders VS
Sonny Jiang [Wed, 3 Oct 2018 15:53:11 +0000 (11:53 -0400)]
radeonsi:optimizing SET_CONTEXT_REG for shaders VS

Signed-off-by: Sonny Jiang <sonny.jiang@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
6 years agoradeonsi:optimizing SET_CONTEXT_REG for shaders GS
Sonny Jiang [Wed, 3 Oct 2018 15:53:10 +0000 (11:53 -0400)]
radeonsi:optimizing SET_CONTEXT_REG for shaders GS

Signed-off-by: Sonny Jiang <sonny.jiang@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
6 years agoradeonsi: optimize and allow reg > 31 in radeon_opt_set_context_reg functions
Marek Olšák [Fri, 5 Oct 2018 22:09:37 +0000 (18:09 -0400)]
radeonsi: optimize and allow reg > 31 in radeon_opt_set_context_reg functions

reg_saved will have 64 bits, and (1 << reg) where reg > 31 has undefined
behavior. (1ull << reg) would be correct for 64 bits.

This commit shifts the other way in order to merge the conditions.

6 years agoradeonsi: optimizing SET_CONTEXT_REG for shaders ES
Sonny Jiang [Wed, 3 Oct 2018 15:53:09 +0000 (11:53 -0400)]
radeonsi: optimizing SET_CONTEXT_REG for shaders ES

Signed-off-by: Sonny Jiang <sonny.jiang@amd.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
6 years agospirv: mark variables decorated with XfbBuffer as always active
Samuel Pitoiset [Fri, 5 Oct 2018 12:39:01 +0000 (14:39 +0200)]
spirv: mark variables decorated with XfbBuffer as always active

Otherwise, they are removed during NIR linking or in some
lowering passes.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
6 years agodocs: update calendar, add news and link release notes to 18.2.2
Juan A. Suarez Romero [Fri, 5 Oct 2018 10:51:34 +0000 (12:51 +0200)]
docs: update calendar, add news and link release notes to 18.2.2

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
6 years agodocs: add sha256 checksums for 18.2.2
Juan A. Suarez Romero [Fri, 5 Oct 2018 10:45:35 +0000 (12:45 +0200)]
docs: add sha256 checksums for 18.2.2

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
(cherry picked from commit cb63a4e1144d9cd8feda3799c68a32a769417b5f)

6 years agodocs: add release notes for 18.2.2
Juan A. Suarez Romero [Fri, 5 Oct 2018 10:13:33 +0000 (12:13 +0200)]
docs: add release notes for 18.2.2

Signed-off-by: Juan A. Suarez Romero <jasuarez@igalia.com>
(cherry picked from commit abaeb79eb2c16d7abad06719f24d1e59ad775aa6)

6 years agonir/alu_to_scalar: Use ssa_for_alu_src in hand-rolled expansions
Jason Ekstrand [Wed, 3 Oct 2018 17:14:20 +0000 (12:14 -0500)]
nir/alu_to_scalar: Use ssa_for_alu_src in hand-rolled expansions

The ssa_for_alu_src helper will correctly handle swizzles and other
source modifiers for you.  The expansions for unpack_half_2x16,
pack_uvec2_to_uint, and pack_uvec4_to_uint were all broken with regards
to swizzles.  The brokenness of unpack_half_2x16 was causing rendering
errors in Rise of the Tomb Raider on Intel ever since c11833ab24dcba26
which added an extra copy propagation to the optimization pipeline and
caused us to start seeing swizzles where we hadn't seen any before.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107926
Fixes: 9ce901058f3d "nir: Add lowering of nir_op_unpack_half_2x16."
Fixes: 9b8786eba955 "nir: Add lowering support for packing opcodes."
Tested-by: Alex Smith <asmith@feralinteractive.com>
Tested-by: Józef Kucia <joseph.kucia@gmail.com>
Reviewed-by: Matt Turner <mattst88@gmail.com>
6 years agoglsl/linker: Check the subroutine associated functions names
Vadym Shovkoplias [Wed, 3 Oct 2018 08:39:04 +0000 (11:39 +0300)]
glsl/linker: Check the subroutine associated functions names

>From Section 6.1.2 (Subroutines) of the GLSL 4.00 specification

    "A program will fail to compile or link if any shader
     or stage contains two or more functions with the same
     name if the name is associated with a subroutine type."

v2:
  - error out earlier (Tapani)
  - style fixes (Iago)

Fixes:
    * no-overloads.vert

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108109
Signed-off-by: Vadym Shovkoplias <vadym.shovkoplias@globallogic.com>
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Reviewed-by: Tapani Pälli <tapani.palli@intel.com>
6 years agovirgl: Negotiate version with vtest server
Tomeu Vizoso [Tue, 2 Oct 2018 07:07:31 +0000 (09:07 +0200)]
virgl: Negotiate version with vtest server

Check if server supports version negotation by sending a PING_PROTOCOL_VERSION
message right before a dummy RESOURCE_BUSY_WAIT. If we don't get a reply
for the first, we know the server doesn't support it.

If it does support it, we can query the max protocol version supported
by the server and fall back if needed.

v2: - Send a new message to negotiate the protocol version, checking if
      the server supports this message by immediately sending a busy wait
      message. (Dave Airlie)

v3: - Send a zero-arg command PING_PROTOCOL_VERSION so we actually keep
      compatibility with older servers. (Code by Dave Airlie)

Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Gurchetan Singh <gurchetansingh@chromium.org>
6 years agointel: aubinator: Fix memory leaks
Sagar Ghuge [Wed, 5 Sep 2018 17:19:47 +0000 (10:19 -0700)]
intel: aubinator: Fix memory leaks

Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agointel/decoder: construct correct xml filename
Sagar Ghuge [Thu, 6 Sep 2018 04:14:23 +0000 (21:14 -0700)]
intel/decoder: construct correct xml filename

construct correct gen xml filename when we try to load hardware xml
description from a given path

v2: remove temporary variable (Francesco Ansanelli)

Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agointel/decoder: Avoid freeing invalid pointer
Sagar Ghuge [Thu, 6 Sep 2018 19:37:28 +0000 (12:37 -0700)]
intel/decoder: Avoid freeing invalid pointer

v2: Free ctx.spec if error while reading genxml (Lionel Landwerlin)

v3: Handle case where genxml is empty (Lionel Landwerlin)

Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agointel/decoder: add gen_spec_init method
Sagar Ghuge [Thu, 6 Sep 2018 04:05:21 +0000 (21:05 -0700)]
intel/decoder: add gen_spec_init method

Initialize gen_spec instance properly when loading hardware xml
description from specifc directory to avoid segmentation fault.

v2: correct function definition (Lionel Landwerlin)

Signed-off-by: Sagar Ghuge <sagar.ghuge@intel.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agoradv: fix resetting the pool for timestamp queries
Samuel Pitoiset [Thu, 4 Oct 2018 08:37:09 +0000 (10:37 +0200)]
radv: fix resetting the pool for timestamp queries

Since the driver no longer uses the availability bit for
timestamp queries it shouldn't reset it. Instead, it should
reset the query values to UINT32_MAX. This fixes VM faults.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=108164
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Tested-by: Józef Kucia <joseph.kucia@gmail.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
6 years agoetnaviv: Use write combine instead of unached mappings for shader bo
Guido Günther [Mon, 1 Oct 2018 16:37:28 +0000 (18:37 +0200)]
etnaviv: Use write combine instead of unached mappings for shader bo

The later are sensitive to unaligned accesses on arm64[1] and we don't
need an uncached mapping here.

[1]: https://lists.freedesktop.org/archives/etnaviv/2018-September/001956.html

Signed-off-by: Guido Günther <guido.gunther@puri.sm>
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
6 years agodrirc: add a workaround for ARMA 3
Marek Olšák [Thu, 4 Oct 2018 04:55:52 +0000 (00:55 -0400)]
drirc: add a workaround for ARMA 3

Cc: 18.2 <mesa-stable@lists.freedesktop.org>
6 years agoanv/batch_chain: Don't start a new BO just for BATCH_BUFFER_START
Jason Ekstrand [Tue, 2 Oct 2018 22:19:32 +0000 (17:19 -0500)]
anv/batch_chain: Don't start a new BO just for BATCH_BUFFER_START

Previously, we just went ahead and emitted MI_BATCH_BUFFER_START as
normal.  If we are near enough to the end, this can cause us to start a
new BO just for the MI_BATCH_BUFFER_START which messes up chaining.  We
always reserve enough space at the end for an MI_BATCH_BUFFER_START so
we can just increment cmd_buffer->batch.end prior to emitting the
command.

Fixes: a0b133286a3 "anv/batch_chain: Simplify secondary batch return..."
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=107926
Tested-by: Alex Smith <asmith@feralinteractive.com>
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agoanv: Use separate MOCS settings for external BOs
Jason Ekstrand [Mon, 9 Jul 2018 21:21:33 +0000 (14:21 -0700)]
anv: Use separate MOCS settings for external BOs

On Broadwell and above, we have to use different MOCS settings to allow
the kernel to take over and disable caching when needed for external
buffers.  On Broadwell, this is especially important because the kernel
can't disable eLLC so we have to do it in userspace.  We very badly
don't want to do that on everything so we need separate MOCS for
external and internal BOs.

In order to do this, we add an anv-specific BO flag for "external" and
use that to distinguish between buffers which may be shared with other
processes and/or display and those which are entirely internal.  That,
together with an anv_mocs_for_bo helper lets us choose the right MOCS
settings for each BO use.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99507
Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
6 years agomeson: remove invalid "opencl" llvm component
Emil Velikov [Fri, 7 Sep 2018 13:58:03 +0000 (14:58 +0100)]
meson: remove invalid "opencl" llvm component

Seeming copy/paste mistake from configure.ac which uses $2 for the
component and $3 for the fancy name printing.

Cc: Dylan Baker <dylan@pnwbakers.com>
Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
6 years agoRevert "mesa: remove unnecessary 'sort by year' for the GL extensions"
Emil Velikov [Mon, 24 Sep 2018 15:01:38 +0000 (16:01 +0100)]
Revert "mesa: remove unnecessary 'sort by year' for the GL extensions"

This reverts commit 3d81e11b49366b5636b8524ba0f8c7076e3fdf34.

As reported by Federico, some games require the 'sort by year' since
they truncate the extensions which do not fit the fixed size string
array.

Seemingly I did not consider that, as the documentation (both Mesa and
Nvidia) mentions about program crashes ... which are worked around by
setting the env. variable.

This commit reinstates the workaround and enhances the documentation.

Cc: Marek Olšák <maraeo@gmail.com>
Cc: Ian Romanick <idr@freedesktop.org>
Reported-by: Federico Dossena <info@fdossena.com>
Fixes: 3d81e11b493 ("mesa: remove unnecessary 'sort by year' for the GL
extensions")
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Tested-by: Federico Dossena <info@fdossena.com>
6 years agomesa: reorder and document the tokens in glheader.h
Emil Velikov [Wed, 5 Sep 2018 16:35:17 +0000 (17:35 +0100)]
mesa: reorder and document the tokens in glheader.h

Split into different sections, document each one as well as strange
cases like GL_ATI_texture_compression_3dc.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agomesa: remove duplicate declarations from glheader.h
Emil Velikov [Wed, 5 Sep 2018 16:35:16 +0000 (17:35 +0100)]
mesa: remove duplicate declarations from glheader.h

Remove all the desktop GL and GLX entries from the list.
Former are pulled by the gl.h and glext.h includes at the top while the
latter are no longer needed.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agoi965: reference __DRI_ATTRIB_SWAP_COPY token over the GLX one
Emil Velikov [Wed, 5 Sep 2018 16:35:15 +0000 (17:35 +0100)]
i965: reference __DRI_ATTRIB_SWAP_COPY token over the GLX one

Earlier commit updated the code to use the DRI tokens, yet forgot to
update the comment.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agoi915: reference __DRI_ATTRIB_SWAP_COPY token over the GLX one
Emil Velikov [Wed, 5 Sep 2018 16:35:14 +0000 (17:35 +0100)]
i915: reference __DRI_ATTRIB_SWAP_COPY token over the GLX one

Earlier commit updated the code to use the DRI tokens, yet forgot to
update the comment.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
6 years agodri/common: move the required GLX_* token definitions locally
Emil Velikov [Wed, 5 Sep 2018 16:35:13 +0000 (17:35 +0100)]
dri/common: move the required GLX_* token definitions locally

Will allow us to remove even bigger hack elsewhere. But more
importantly, we should not be using _any_ GLX tokens in DRI.

Document the gory details about the current side-effects.

Signed-off-by: Emil Velikov <emil.velikov@collabora.com>
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>