mesa.git
5 years agonir/algebraic: Mark other comparison exact when removing a == a
Ian Romanick [Thu, 24 Oct 2019 21:41:20 +0000 (14:41 -0700)]
nir/algebraic: Mark other comparison exact when removing a == a

This prevents some additional optimizations that would change the
original result.  This includes things like (b < a && b < c) => b <
min(a, c) and !(a < b) => b >= a.  Both of these optimizations were
specifically observed in the piglit tests added in piglit!160.

This was discovered while investigating
https://gitlab.freedesktop.org/mesa/mesa/issues/1958.  However, the
problem in that issue was Chrome or Angle is replacing calls to isnan()
with some stuff that we (correctly) optimize to false.  If they had left
the calls to isnan() alone, everything would have just worked.

No shader-db changes on any Intel platform.

I also tried marking the comparison generated by the isnan() function
precise.  The precise marker "infects" every computation involved in
calculating the parameter to the isnan() function, and this severely
hurt all of the (few) shaders in shader-db that use isnan().

I also considered adding a new ir_unop_isnan opcode that would implement
the functionality.  During GLSL IR-to-NIR translation, the resulting
comparison operation would be marked exact (and the samething would need
to happen in SPIR-V translation).

This approach taken by this patch seemed easier, but we may want to do
the ir_unop_isnan thing anyway.

Fixes: d55835b8bdf ("nir/algebraic: Add optimizations for "a == a && a CMP b"")
Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
5 years agonir/algebraic: Add the ability to mark a replacement as exact
Ian Romanick [Thu, 24 Oct 2019 20:41:59 +0000 (13:41 -0700)]
nir/algebraic: Add the ability to mark a replacement as exact

Reviewed-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
5 years agocompiler: make variable::data::binding unsigned
Marek Olšák [Fri, 25 Oct 2019 20:36:15 +0000 (16:36 -0400)]
compiler: make variable::data::binding unsigned

Nothing seems to set a negative value.

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
5 years agost/mesa: call nir_lower_flrp only once per shader
Marek Olšák [Fri, 25 Oct 2019 04:15:37 +0000 (00:15 -0400)]
st/mesa: call nir_lower_flrp only once per shader

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
5 years agost/mesa: call nir_opt_access only once
Marek Olšák [Fri, 25 Oct 2019 04:06:34 +0000 (00:06 -0400)]
st/mesa: call nir_opt_access only once

Reviewed-by: Connor Abbott <cwabbott0@gmail.com>
5 years agoac: add missing Arcturus to the info of pc lines
Leo Liu [Mon, 4 Nov 2019 16:18:46 +0000 (11:18 -0500)]
ac: add missing Arcturus to the info of pc lines

Signed-off-by: Leo Liu <leo.liu@amd.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Cc: Marek Olšák <marek.olsak@amd.com>
5 years agopanfrost/ci: Update T760 expectations
Alyssa Rosenzweig [Mon, 4 Nov 2019 20:35:23 +0000 (15:35 -0500)]
panfrost/ci: Update T760 expectations

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopan/midgard: Extend default_phys_reg to !32-bit
Alyssa Rosenzweig [Thu, 31 Oct 2019 18:57:54 +0000 (14:57 -0400)]
pan/midgard: Extend default_phys_reg to !32-bit

We can pass through a size.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopan/midgard: Extend swizzle packing for vec4/16-bit
Alyssa Rosenzweig [Thu, 31 Oct 2019 18:56:45 +0000 (14:56 -0400)]
pan/midgard: Extend swizzle packing for vec4/16-bit

We would like to pack not just xyzw swizzles but also efgh swizzles.
This should work for vec4/16-bit. More work will be needed to pack
swizzles for vec8/16-bit and even more work for 8-bit, of course.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopan/midgard: Extend offset_swizzle to non-32-bit
Alyssa Rosenzweig [Thu, 31 Oct 2019 18:57:35 +0000 (14:57 -0400)]
pan/midgard: Extend offset_swizzle to non-32-bit

We take a size parameter; use it.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopan/midgard: offset_swizzle doesn't need dstsize
Alyssa Rosenzweig [Mon, 4 Nov 2019 15:44:54 +0000 (10:44 -0500)]
pan/midgard: offset_swizzle doesn't need dstsize

This argument should be omitted.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopan/midgard: Add bizarre corner case
Alyssa Rosenzweig [Mon, 4 Nov 2019 15:40:37 +0000 (10:40 -0500)]
pan/midgard: Add bizarre corner case

Someone really needs to look into this.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopan/midgard: Compute bundle interference
Alyssa Rosenzweig [Tue, 22 Oct 2019 01:32:35 +0000 (21:32 -0400)]
pan/midgard: Compute bundle interference

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopan/midgard: Fix quadword_count handling
Alyssa Rosenzweig [Mon, 4 Nov 2019 15:32:49 +0000 (10:32 -0500)]
pan/midgard: Fix quadword_count handling

Spilling can mess with this considerably.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopan/midgard: Validate tags when branching
Alyssa Rosenzweig [Mon, 4 Nov 2019 15:09:31 +0000 (10:09 -0500)]
pan/midgard: Validate tags when branching

Midgard prefetches instructions based on tag (ALU, LD/ST, texture *
size). To do so, the shader descriptor specifies the tag of the first
instruction, all instructions specify the tag of the next linear
instruction is, and all branches explicitly specify the tag of the
branch target.

If you mess this up, you get an INSTR_TYPE_MISMATCH, which unambiguously
refers to this problem, but it's still annoying to try to work out all
the branch targets in your head to debug.

Instead, let's track the tags of various blocks over time, so we can
automatically validate tags of branch targets, to make
INSTR_TYPE_MISMATCH issues immediately obvious in a disassembly.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agoaco: fix accidential reordering of instructions when scheduling
Daniel Schürmann [Thu, 31 Oct 2019 16:33:35 +0000 (17:33 +0100)]
aco: fix accidential reordering of instructions when scheduling

Fixes: 86786999189c43b4a2c8e1c1a18b55cd2f369fff "aco: implement VGPR spilling"
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
5 years agoaco: only use single-dword loads/stores for spilling
Daniel Schürmann [Fri, 1 Nov 2019 08:06:26 +0000 (09:06 +0100)]
aco: only use single-dword loads/stores for spilling

Fixes: 86786999189c43b4a2c8e1c1a18b55cd2f369fff "aco: implement VGPR spilling"
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
5 years agoaco: fix immediate offset for spills if scratch is used
Daniel Schürmann [Thu, 31 Oct 2019 12:25:44 +0000 (13:25 +0100)]
aco: fix immediate offset for spills if scratch is used

Fixes: 86786999189c43b4a2c8e1c1a18b55cd2f369fff "aco: implement VGPR spilling"
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
5 years agoanv: Properly handle host query reset of performance queries
Lionel Landwerlin [Wed, 30 Oct 2019 11:51:08 +0000 (13:51 +0200)]
anv: Properly handle host query reset of performance queries

The host query reset entry point didn't use the availability offset
for performance queries.

To fix this, reorder the availability of performance queries to match
other queries.

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Fixes: 2b5f30b1d9 ("anv: implement VK_INTEL_performance_query")
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agostate_tracker: Handle texture view min level in st_generate_mipmap()
Paul Gofman [Mon, 4 Nov 2019 12:31:10 +0000 (15:31 +0300)]
state_tracker: Handle texture view min level in st_generate_mipmap()

Signed-off-by: Paul Gofman <gofmanp@gmail.com>
Signed-off-by: Marek Olšák <marek.olsak@amd.com>
5 years agoiris: try to set the specified tiling when importing a dmabuf
James Xiong [Thu, 4 Apr 2019 15:27:33 +0000 (08:27 -0700)]
iris: try to set the specified tiling when importing a dmabuf

When importing a dmabuf with a specified tiling, the dmabuf user
should always try to set the tiling mode because: 1) the exporter
can set tiling AFTER exporting/importing. 2) a dmabuf could be
exported from a kernel driver other than i915, in this case the
dmabuf user and exporter need to set tiling separately.

This patch fixes a problem when running vkmark under weston with
iris on ICL, it crashed to console with the following assert. i965
doesn't have this problem as it always tries to set the specified
tiling mode.

weston: ../src/gallium/drivers/iris/iris_resource.c:990: iris_resource_from_handle: Assertion `res->bo->tiling_mode == isl_tiling_to_i915_tiling(res->surf.tiling)' failed.

Signed-off-by: James Xiong <james.xiong@intel.com>
Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
5 years agoiris: Fix "Force Zero RTA Index Enable" setting again
Kenneth Graunke [Mon, 4 Nov 2019 07:50:38 +0000 (23:50 -0800)]
iris: Fix "Force Zero RTA Index Enable" setting again

In 2ca0d913ea8, we began updating cso_fb->layers to the actual layer
count, rather than 0.  This fixed cases where we were setting "Force
Zero RTA Index Enable" even when doing layered rendering.  Sadly, it
also broke the check entirely: cso_fb->layers is now 1 for non-layered
cases, but the Force Zero RTA Index check was still comparing for 0.

Fixes: 2ca0d913ea8 ("iris: Fix framebuffer layer count")
5 years agonir: correct use of identity check in python
Dylan Baker [Fri, 25 Oct 2019 20:48:38 +0000 (13:48 -0700)]
nir: correct use of identity check in python

Python has the identity operator `is`, and the equality operator `==`.
Using `is` with strings sometimes works in CPython due to optimizations
(they have some kind of cache), but it may not always work.

Fixes: 96c4b135e34d0804e41bfbc28fc1b5050c49d71e
       ("nir/algebraic: Don't put quotes around floating point literals")
Reviewed-by: Matt Turner <mattst88@gmail.com>
5 years agopanfrost: MALI_DEPTH_TEST is actually MALI_DEPTH_WRITEMASK
Boris Brezillon [Mon, 4 Nov 2019 10:57:22 +0000 (11:57 +0100)]
panfrost: MALI_DEPTH_TEST is actually MALI_DEPTH_WRITEMASK

MALI_DEPTH_TEST should only be set when depth->writemask is true,
not when the depth test is enabled. Let's rename the flag and patch
panfrost_bind_depth_stencil_state() to do the right thing.

Signed-off-by: Boris Brezillon <boris.brezillon@collabora.com>
Reviewed-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agovulkan: bump headers/registry to 1.1.127
Lionel Landwerlin [Mon, 4 Nov 2019 08:21:00 +0000 (10:21 +0200)]
vulkan: bump headers/registry to 1.1.127

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agoradv: fix compute pipeline keys when optimizations are disabled
Samuel Pitoiset [Fri, 1 Nov 2019 08:34:12 +0000 (09:34 +0100)]
radv: fix compute pipeline keys when optimizations are disabled

If an app first creates a compute pipeline with
VK_PIPELINE_CREATE_DISABLE_OPTIMIZATION_BIT set, then re-compile it
without that flag, the driver should re-compile the compute shader.
Otherwise, it will return the unoptimized one.

Fixes: ce188813bfe ("radv: add initial support for VK_PIPELINE_CREATE_DISABLE_OPTIMIZATION_BIT")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
5 years agonv50/ir: fix crash in isUniform for undefined values
Karol Herbst [Sat, 2 Nov 2019 23:57:28 +0000 (00:57 +0100)]
nv50/ir: fix crash in isUniform for undefined values

Signed-off-by: Karol Herbst <kherbst@redhat.com>
Reviewed-by: Ilia Mirkin <imirkin@alum.mit.edu>
5 years agomesa: check draw buffer completeness on glClearBufferfi/glClearBufferiv
Lionel Landwerlin [Fri, 1 Nov 2019 17:43:00 +0000 (19:43 +0200)]
mesa: check draw buffer completeness on glClearBufferfi/glClearBufferiv

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
5 years agolima: set dithering flag when necessary
Vasily Khoruzhick [Sat, 26 Oct 2019 02:08:57 +0000 (19:08 -0700)]
lima: set dithering flag when necessary

Bit 13 in aux1 enables dithering

Reviewed-by: Qiang.Yu <yuq825@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
5 years agoglsl: encode struct/interface types better
Marek Olšák [Thu, 31 Oct 2019 00:38:44 +0000 (20:38 -0400)]
glsl: encode struct/interface types better

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
5 years agoglsl: encode array types better
Marek Olšák [Thu, 31 Oct 2019 00:27:08 +0000 (20:27 -0400)]
glsl: encode array types better

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
5 years agoglsl: encode explicit_stride for basic types better
Marek Olšák [Thu, 31 Oct 2019 00:22:49 +0000 (20:22 -0400)]
glsl: encode explicit_stride for basic types better

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
5 years agoglsl: encode vector_elements and matrix_columns better
Marek Olšák [Thu, 31 Oct 2019 00:18:05 +0000 (20:18 -0400)]
glsl: encode vector_elements and matrix_columns better

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
5 years agoglsl: encode/decode types using a union with bitfields for readability
Marek Olšák [Thu, 31 Oct 2019 00:15:36 +0000 (20:15 -0400)]
glsl: encode/decode types using a union with bitfields for readability

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
5 years agolima: ignore flags while looking for BO in cache
Vasily Khoruzhick [Fri, 25 Oct 2019 05:14:05 +0000 (22:14 -0700)]
lima: ignore flags while looking for BO in cache

Any BO would work, we don't have any BO types yet anyway. Moreover
lima_submit_add_bo() changes BO flags so they won't match allocation
flags.

Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
5 years agolima: align size before trying to fetch BO from cache
Vasily Khoruzhick [Fri, 25 Oct 2019 05:12:30 +0000 (22:12 -0700)]
lima: align size before trying to fetch BO from cache

Otherwise we may be looking in wrong bucket

Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
5 years agolima: add debug prints for BO cache
Vasily Khoruzhick [Fri, 25 Oct 2019 03:16:06 +0000 (20:16 -0700)]
lima: add debug prints for BO cache

LIMA_DEBUG=bocache now activates debug prints for BO allocation,
destruction and BO cache.

Reviewed-by: Erico Nunes <nunes.erico@gmail.com>
Signed-off-by: Vasily Khoruzhick <anarsoul@gmail.com>
5 years agopan/midgard: Use fp32 blend shaders
Alyssa Rosenzweig [Thu, 31 Oct 2019 19:50:45 +0000 (15:50 -0400)]
pan/midgard: Use fp32 blend shaders

Clearly we do want to have fp16 at some point ... but I kind of give up
debugging and it turns out the issues with fp16 support in 'frost are so
deeply rooted that I might as well disable this non-opt and land
LCRA now.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agoradv: Close all unnecessary fds in secure compile.
Bas Nieuwenhuizen [Fri, 1 Nov 2019 00:07:02 +0000 (01:07 +0100)]
radv: Close all unnecessary fds in secure compile.

The seccomp filter allows read/write, let us make sure nobody can
do anything with this.

Fixes: cff53da3748 "radv: enable secure compile support"
Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
5 years agoanv: remove incorrect polygonMode=point early-out
Erik Faye-Lund [Mon, 28 Oct 2019 15:25:10 +0000 (16:25 +0100)]
anv: remove incorrect polygonMode=point early-out

This is incorrect, because polygonMode only applies if the final
primitive type is a polygon; polygonMode doesn't apply to
line-primitives as the comment suggests.

The Vulkan 1.1 spec, section 26.11, "Polygons" defines that polygons are
separate from points and line segments:

" A polygon results from the decomposition of a triangle strip, triangle
  fan or a series of independent triangles. Like points and line segments,
  polygon rasterization is controlled by several variables in the
  VkPipelineRasterizationStateCreateInfo structure. "

Further, section 26.11.2, "Polygon Mode", only define polygonMode to
apply to polygons:

" Possible values of the VkPipelineRasterizationStateCreateInfo::polygonMode
  property of the currently active pipeline, specifying the method of
  rasterization for polygons, are: "

This seems to clearly define that polygonMode doesn't apply to points
and lines, so let's make sure that we don't early out with the wrong
value.

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
5 years agopan/midgard: Eliminate blank_alu_src
Alyssa Rosenzweig [Wed, 30 Oct 2019 20:29:28 +0000 (16:29 -0400)]
pan/midgard: Eliminate blank_alu_src

We don't need it in practice, so this is some more cleanup.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopan/midgard: Refactor swizzles
Alyssa Rosenzweig [Sat, 26 Oct 2019 18:06:17 +0000 (14:06 -0400)]
pan/midgard: Refactor swizzles

Rather than having hw-specific swizzles encoded directly in the
instructions, have a unified swizzle arary so we can manipulate swizzles
generically.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopan/midgard: Add a dummy source for loads
Alyssa Rosenzweig [Sat, 26 Oct 2019 19:50:38 +0000 (15:50 -0400)]
pan/midgard: Add a dummy source for loads

We want symmetry between loads and stores, so we add a dummy source. So
we get, e.g.

   st_int4 _,    val, arg_1, arg_2
   ld_int4 dest,   _, arg_1, arg_2

Semantically, this dummy source represents the data itself, as if the
load is simply a move. That means it has a swizzle that acts as a
source.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agopan/midgard: Remove OP_IS_STORE_VARY
Alyssa Rosenzweig [Sat, 26 Oct 2019 19:50:26 +0000 (15:50 -0400)]
pan/midgard: Remove OP_IS_STORE_VARY

Unused.

Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
5 years agoglsl: disable lower_fragdata_array() for NIR drivers
Timothy Arceri [Wed, 30 Oct 2019 00:34:27 +0000 (11:34 +1100)]
glsl: disable lower_fragdata_array() for NIR drivers

This function was added in 7e414b58640a to work around a defect in
lower_output_reads(). As of the previous commit no NIR driver calls
lower_output_reads().

This change means we don't need the special GLSL IR style
gl_FragData handling for building the resource list in a NIR based
linker.

No shader-db change on SKL i965.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
5 years agoglsl: just use NIR to lower outputs when driver can't read outputs
Timothy Arceri [Wed, 30 Oct 2019 00:24:37 +0000 (11:24 +1100)]
glsl: just use NIR to lower outputs when driver can't read outputs

This will allow us to stop lowering gl_FragData in GLSL IR for NIR
drivers which means we won't need the special GLSL IR type
handling for building the resource list in a NIR based linker.

i965 has been doing this since b828f7a27b2b.

Reviewed-by: Marek Olšák <marek.olsak@amd.com>
5 years agolima: support indexed draw with bias
Icenowy Zheng [Tue, 22 Oct 2019 14:14:56 +0000 (22:14 +0800)]
lima: support indexed draw with bias

When doing an indexed draw with index_bias set to a non-zero value (e.g.
by glDrawElementsBaseVertex), the vertex buffer should be offseted by
index_bias vertices.

Add this offset when setting the vertex buffer address.

Signed-off-by: Icenowy Zheng <icenowy@aosc.io>
Reviewed-by: Vasily Khoruzhick <anarsoul@gmail.com>
5 years agoanv: Move the RT BTI flush workaround to begin_subpass
Jason Ekstrand [Wed, 30 Oct 2019 20:05:29 +0000 (15:05 -0500)]
anv: Move the RT BTI flush workaround to begin_subpass

Now that we're no longer compacting binding table entries, the only time
they can possibly change is when we actually switch subpasses.

Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
5 years agoanv: Stop compacting render targets in the binding table
Jason Ekstrand [Wed, 30 Oct 2019 19:07:47 +0000 (14:07 -0500)]
anv: Stop compacting render targets in the binding table

Instead, always emit one entry for every color attachment in the subpass
or one NULL if there are no color attachments.  This will let us adjust
an Ice Lake workaround so we don't get a stall on every draw call.

Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
5 years agoanv: Don't claim the null RT as a valid color target
Jason Ekstrand [Wed, 30 Oct 2019 20:53:11 +0000 (15:53 -0500)]
anv: Don't claim the null RT as a valid color target

If it's NULL, we can let the compiler go ahead and delete it or flag it
as NULL.

Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
5 years agoanv: Don't delete fragment shaders that write sample mask
Jason Ekstrand [Wed, 30 Oct 2019 20:50:51 +0000 (15:50 -0500)]
anv: Don't delete fragment shaders that write sample mask

Also, use color_outputs_valid rather than nr_color_outputs since it
should be a bit more accurate.

Reviewed-by: Rafael Antognolli <rafael.antognolli@intel.com>
5 years agoglsl: Enable textureSize for samplerExternalOES
Yevhenii Kolesnikov [Mon, 7 Oct 2019 12:59:52 +0000 (15:59 +0300)]
glsl: Enable textureSize for samplerExternalOES

From OES_EGL_image_external_essl3

Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/1901
Signed-off-by: Yevhenii Kolesnikov <yevhenii.kolesnikov@globallogic.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
5 years agoci: Switch over to an autoscaling GKE cluster for builds.
Eric Anholt [Mon, 21 Oct 2019 20:36:14 +0000 (13:36 -0700)]
ci: Switch over to an autoscaling GKE cluster for builds.

The GKE pool we're using is 1-3 32-core VMs, preemptible (to keep
costs down), with 8 jobs concurrent per system.  We have plenty of
memory (4G/core), so we run make -j8 to try to keep the cores busy even
when one job is in a single-threaded step (docker image download, git
clone, artifacts processing, etc.)  When all jobs are generating work
for all the cores, they'll be scheduled fairly.

The nodes in the pool have 300GB boot disks (over-provisioned in space
to provide enough iops and throughput) mounted to /ccache, and
CACHE_DIR set pointing to them.  This means that once a new
autoscaled-up node has run some jobs, it should have a hot ccache from
then on (instead of having to rely on the docker container cache
having our ccache laying around and not getting wiped out by some
other fd.o job).  Local SSDs would provide higher performance, but
unfortunately are not supported with the cluster autoscaler.

For now, the softpipe/llvmpipe test runs are still on the shared
runners, until I can get them ported onto Bas's runner so they can be
parallelized in a single job.

Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
5 years agoci: Make lava inherit the ccache setup of the .build script.
Eric Anholt [Mon, 28 Oct 2019 22:21:44 +0000 (15:21 -0700)]
ci: Make lava inherit the ccache setup of the .build script.

It was just duplicating the code.

Reviewed-by: Michel Dänzer <mdaenzer@redhat.com>
5 years agomeson: revert glvnd workaround
Eric Engestrom [Wed, 25 Sep 2019 16:38:49 +0000 (17:38 +0100)]
meson: revert glvnd workaround

This effectively reverts MR !2112.

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
5 years agomeson: require glvnd 1.2.0
Eric Engestrom [Wed, 25 Sep 2019 16:42:18 +0000 (17:42 +0100)]
meson: require glvnd 1.2.0

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
5 years agogitlab-ci: build a recent enough version of GLVND (ie. 1.2.0)
Eric Engestrom [Wed, 25 Sep 2019 19:18:27 +0000 (20:18 +0100)]
gitlab-ci: build a recent enough version of GLVND (ie. 1.2.0)

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Reviewed-by: Dylan Baker <dylan@pnwbakers.com>
5 years agomeson: move idep_xmlconfig_headers to xmlpool/
Eric Engestrom [Thu, 31 Oct 2019 15:35:29 +0000 (15:35 +0000)]
meson: move idep_xmlconfig_headers to xmlpool/

That's where `xmlpool_options_h` is defined, and this way we can make sure
nobody starts making use of it in the future :)

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agoanv: Use the new BO alloc API for Android
Jason Ekstrand [Thu, 31 Oct 2019 15:28:05 +0000 (10:28 -0500)]
anv: Use the new BO alloc API for Android

Fixes: a44f5ee0d8b "anv: Rework the internal BO allocation API"
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agozink: emit line-width when using polygon line-mode
Erik Faye-Lund [Wed, 30 Oct 2019 12:57:21 +0000 (13:57 +0100)]
zink: emit line-width when using polygon line-mode

When switching this to dynamic state, I forgot that this also needs to
be emitted when we use a polygon-mode set to lines.

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Fixes: 6d30abb4f14 ("zink: use dynamic state for line-width")
5 years agoradeon: replace xmlpool_options_h with idep_xmlconfig_headers
Eric Engestrom [Sat, 26 Oct 2019 23:55:52 +0000 (00:55 +0100)]
radeon: replace xmlpool_options_h with idep_xmlconfig_headers

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agor200: replace xmlpool_options_h with idep_xmlconfig_headers
Eric Engestrom [Sat, 26 Oct 2019 23:55:45 +0000 (00:55 +0100)]
r200: replace xmlpool_options_h with idep_xmlconfig_headers

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agonouveau: replace xmlpool_options_h with idep_xmlconfig_headers
Eric Engestrom [Sat, 26 Oct 2019 23:55:37 +0000 (00:55 +0100)]
nouveau: replace xmlpool_options_h with idep_xmlconfig_headers

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agoi915: replace xmlpool_options_h with idep_xmlconfig_headers
Eric Engestrom [Sat, 26 Oct 2019 23:55:30 +0000 (00:55 +0100)]
i915: replace xmlpool_options_h with idep_xmlconfig_headers

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agodri: replace xmlpool_options_h with idep_xmlconfig_headers
Eric Engestrom [Sat, 26 Oct 2019 23:55:06 +0000 (00:55 +0100)]
dri: replace xmlpool_options_h with idep_xmlconfig_headers

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agotargets/xvmc: replace xmlpool_options_h with idep_xmlconfig_headers
Eric Engestrom [Sat, 26 Oct 2019 23:54:49 +0000 (00:54 +0100)]
targets/xvmc: replace xmlpool_options_h with idep_xmlconfig_headers

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agotargets/xa: replace xmlpool_options_h with idep_xmlconfig_headers
Eric Engestrom [Sat, 26 Oct 2019 23:54:38 +0000 (00:54 +0100)]
targets/xa: replace xmlpool_options_h with idep_xmlconfig_headers

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agotargets/vdpau: replace xmlpool_options_h with idep_xmlconfig_headers
Eric Engestrom [Sat, 26 Oct 2019 23:54:28 +0000 (00:54 +0100)]
targets/vdpau: replace xmlpool_options_h with idep_xmlconfig_headers

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agotargets/va: replace xmlpool_options_h with idep_xmlconfig_headers
Eric Engestrom [Sat, 26 Oct 2019 23:54:19 +0000 (00:54 +0100)]
targets/va: replace xmlpool_options_h with idep_xmlconfig_headers

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agotargets/omx: replace xmlpool_options_h with idep_xmlconfig_headers
Eric Engestrom [Sat, 26 Oct 2019 23:54:10 +0000 (00:54 +0100)]
targets/omx: replace xmlpool_options_h with idep_xmlconfig_headers

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agoloader: replace xmlpool_options_h with idep_xmlconfig_headers
Eric Engestrom [Sat, 26 Oct 2019 23:53:37 +0000 (00:53 +0100)]
loader: replace xmlpool_options_h with idep_xmlconfig_headers

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agopipe-loader: drop unnecessary xmlpool_options_h
Eric Engestrom [Sat, 26 Oct 2019 23:51:41 +0000 (00:51 +0100)]
pipe-loader: drop unnecessary xmlpool_options_h

idep_xmlconfig already covers that

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agoradv: drop unnecessary xmlpool_options_h
Eric Engestrom [Sat, 26 Oct 2019 23:51:22 +0000 (00:51 +0100)]
radv: drop unnecessary xmlpool_options_h

idep_xmlconfig already covers that

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agoanv: add missing xmlconfig headers dependency
Eric Engestrom [Sat, 26 Oct 2019 23:04:33 +0000 (00:04 +0100)]
anv: add missing xmlconfig headers dependency

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agomeson: split out idep_xmlconfig_headers from idep_xmlconfig
Eric Engestrom [Sat, 26 Oct 2019 23:48:36 +0000 (00:48 +0100)]
meson: split out idep_xmlconfig_headers from idep_xmlconfig

A bunch of components need the former but not the latter.

Signed-off-by: Eric Engestrom <eric.engestrom@intel.com>
Acked-by: Dylan Baker <dylan@pnwbakers.com>
5 years agopipe-loader: Build kmsro loader for with all kmsro targets
Alyssa Rosenzweig [Sat, 26 Oct 2019 19:10:22 +0000 (15:10 -0400)]
pipe-loader: Build kmsro loader for with all kmsro targets

Build failure reported by i965 CI, triggered by building dynamic
pipeloaders with kmsro drivers (besides 'frost). At this point, there's
no reason to actually do that -- mesa CI didn't mind -- but let's not
break the build.

v2: Simplify script. Add extra dependencies for v3d.

Fixes: afb0d08cb0f ("pipe-loader: Default to kmsro if probe fails")
Signed-off-by: Alyssa Rosenzweig <alyssa.rosenzweig@collabora.com>
Reported-by: Clayton Craft <clayton.a.craft@intel.com>
Tested-by: Clayton Craft <clayton.a.craft@intel.com>
Reviewed-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
5 years agozink: heap-allocate samplers objects
Erik Faye-Lund [Tue, 29 Oct 2019 09:07:53 +0000 (10:07 +0100)]
zink: heap-allocate samplers objects

VkSampler is 64-bit even on 32-bit systems, so casting it to a pointer
is a bad idea there. So let's heap-allocate the sampler-object instead.

Signed-off-by: Erik Faye-Lund <erik.faye-lund@collabora.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/issues/2017
Reviewed-by: Witold Baryluk <witold.baryluk@gmail.com>
Tested-by: Witold Baryluk <witold.baryluk@gmail.com>
5 years agoanv: Zero released anv_bo structs
Jason Ekstrand [Wed, 30 Oct 2019 17:31:12 +0000 (12:31 -0500)]
anv: Zero released anv_bo structs

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Use a bitset for tracking residency
Jason Ekstrand [Tue, 29 Oct 2019 19:26:15 +0000 (14:26 -0500)]
anv: Use a bitset for tracking residency

Now that we can conveniently map between GEM handles and struct anv_bo
pointers, we can use a simple bitset for residency tracking instead of
the complex hash set.  This shaves about 3% off of a CPU-limited example
running with the Dawn WebGPU implementation.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Set the batch allocator for compute pipelines
Jason Ekstrand [Wed, 30 Oct 2019 19:37:45 +0000 (14:37 -0500)]
anv: Set the batch allocator for compute pipelines

Otherwise relocations just up and crash.

Fixes: a3153162a9b "anv: Delay allocation of relocation lists"
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Add a device parameter to anv_execbuf_add_bo
Jason Ekstrand [Tue, 29 Oct 2019 20:18:16 +0000 (15:18 -0500)]
anv: Add a device parameter to anv_execbuf_add_bo

We're about to start needing to lookup BO pointers by GEM handle so we
need access to the device.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Drop anv_bo_init and anv_bo_init_new
Jason Ekstrand [Mon, 28 Oct 2019 23:03:32 +0000 (18:03 -0500)]
anv: Drop anv_bo_init and anv_bo_init_new

BOs are now only ever allocated through the BO cache so there's no need
to have these exposed.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Allocate misc BOs from the cache
Jason Ekstrand [Mon, 28 Oct 2019 22:28:09 +0000 (17:28 -0500)]
anv: Allocate misc BOs from the cache

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Allocate scratch BOs from the cache
Jason Ekstrand [Mon, 28 Oct 2019 21:42:02 +0000 (16:42 -0500)]
anv: Allocate scratch BOs from the cache

While we're here, we get rid of the locking and use a lock-free
algorithm.  The chances of spilling contention are low and this is
actually a bit simpler in some ways.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Allocate batch and fence buffers from the cache
Jason Ekstrand [Mon, 28 Oct 2019 20:42:20 +0000 (15:42 -0500)]
anv: Allocate batch and fence buffers from the cache

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoutil: Add a free list structure for use with util_sparse_array
Jason Ekstrand [Mon, 28 Oct 2019 19:49:38 +0000 (14:49 -0500)]
util: Add a free list structure for use with util_sparse_array

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Allocate descriptor buffers from the BO cache
Jason Ekstrand [Fri, 25 Oct 2019 22:15:31 +0000 (17:15 -0500)]
anv: Allocate descriptor buffers from the BO cache

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Set more flags on descriptor pool buffers
Jason Ekstrand [Fri, 25 Oct 2019 22:12:06 +0000 (17:12 -0500)]
anv: Set more flags on descriptor pool buffers

the ASYNC flag, in particular, has the potential to help performance
because it means less sync tracking in the kernel.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Allocate query pool BOs from the cache
Jason Ekstrand [Fri, 25 Oct 2019 22:07:36 +0000 (17:07 -0500)]
anv: Allocate query pool BOs from the cache

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Use the query_slot helper in vkResetQueryPoolEXT
Jason Ekstrand [Fri, 25 Oct 2019 22:07:02 +0000 (17:07 -0500)]
anv: Use the query_slot helper in vkResetQueryPoolEXT

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Allocate block pool BOs from the cache
Jason Ekstrand [Fri, 25 Oct 2019 21:29:29 +0000 (16:29 -0500)]
anv: Allocate block pool BOs from the cache

This commit switches block pools over to being allocated from the BO
cache rather than being allocated manually by the block pool.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv/tests: Initialize the BO cache and device mutex
Jason Ekstrand [Wed, 30 Oct 2019 16:44:12 +0000 (11:44 -0500)]
anv/tests: Initialize the BO cache and device mutex

We're about to start depending on the BO cache in the state and block
pools so we need them properly initialized for the tests to work.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv/tests: Zero-initialize instances
Jason Ekstrand [Wed, 30 Oct 2019 16:43:53 +0000 (11:43 -0500)]
anv/tests: Zero-initialize instances

Some of the tests were actually relying on some of those uninitialized
bits to be non-zero.  In particular, a couple want use_softpin = true.

Cc: mesa-stable@lists.freedesktop.org
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Choose BO flags internally in anv_block_pool
Jason Ekstrand [Fri, 25 Oct 2019 23:18:52 +0000 (18:18 -0500)]
anv: Choose BO flags internally in anv_block_pool

All block pools are allocated with the same flags.  There's no good
reason why it needs to be configurable.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Rework the internal BO allocation API
Jason Ekstrand [Fri, 25 Oct 2019 22:45:28 +0000 (17:45 -0500)]
anv: Rework the internal BO allocation API

This makes a number of changes to the current API:

 1. Everything is renamed to anv_device_* instead of anv_bo_cache_*
    because the BO cache is soon going to be the sole BO allocation path
    and not some special case to make import/export work.

 2. Drop the cache parameter.  It's totally redundant with the device
    and just annoying to keep typing.

 3. Rework flags so that they go the convenient direction for usage in
    ANV rather than whichever awkward way the i915 specified it to
    maintain backwards compatibility.  This also gives us the
    opportunity to set some defaults.

 4. Add flags for mapping and coherency.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Use anv_block_pool_foreach_bo in get_bo_from_pool
Jason Ekstrand [Fri, 25 Oct 2019 21:42:47 +0000 (16:42 -0500)]
anv: Use anv_block_pool_foreach_bo in get_bo_from_pool

While we're at it, use gen_48b_address().

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Rework anv_block_pool_expand_range
Jason Ekstrand [Fri, 25 Oct 2019 21:10:11 +0000 (16:10 -0500)]
anv: Rework anv_block_pool_expand_range

The growing algorithms for the softpin case and the userptr version are
almost entirely different.  Having this weird join doesn't make the code
more comprehensible.  This rework does a few things:

 1. Move the comment about 48-bit addresses to anv_device_init where we
    actually unset the EXEC_OBJECT_SUPPORTS_48B_ADDRESS flag.

 2. Separate the paths in anv_block_pool_expand_range so it's easier to
    see what happens in the two different cases.

 3. Use the anv_block_poo::bos array for storing all allocated BOs in
    both paths rather than using the cleanup list in both paths.  This
    lets us make the cleanups array only used for mmaps of the memfd for
    the userptr case.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Fix a potential BO handle leak
Jason Ekstrand [Fri, 25 Oct 2019 20:42:22 +0000 (15:42 -0500)]
anv: Fix a potential BO handle leak

Fixes: 731c4adcf9b "anv/allocator: Add support for non-userptr"
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Handle state pool relocations using "wrapper" BOs
Jason Ekstrand [Fri, 25 Oct 2019 19:52:37 +0000 (14:52 -0500)]
anv: Handle state pool relocations using "wrapper" BOs

Instead of depending on a mutable BO in the state pool for handling
growing state pools, add a concept of "wrapper" BOs which just wrap an
actual BO.  This way, the wrapper can exist once for all of time and we
can put it in relocation lists even if the actual BO it references gets
swapped out.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
5 years agoanv: Replace ANV_BO_EXTERNAL with anv_bo::is_external
Jason Ekstrand [Tue, 29 Oct 2019 01:12:24 +0000 (20:12 -0500)]
anv: Replace ANV_BO_EXTERNAL with anv_bo::is_external

We're not THAT strapped for space that we can't burn one extra bit for
a boolean.  If we're really worried about it, we can always shrink the
flags field to 16 bits because the kernel only uses 7 currently.

Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>