Revert "zink: refcount the shader cache"

This reverts commit b9fdc21bba0724271520462f3f04ba72ae106a26.

Fixes: b9fdc21bba0 ("zink: refcount the shader cache")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7754>

Revert "zink: move shader key structs into their own header"

This reverts commit ed7a5a55689ddcf10aae3f671e390252aad2d62f.

Fixes: ed7a5a55689 ("zink: move shader key structs into their own header")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7754>

Revert "zink: fill in params for fs shader keys and flag shader for rebuild"

This reverts commit 9aa08221fa6653002dec2a07641d58d3c7fa227a.

Fixes: 9aa08221fa6 ("zink: fill in params for fs shader keys and flag shader for rebuild")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7754>

Revert "zink: put those shader keys to work fixing up fragment shaders"

This reverts commit eeff625ab31f0a1bf94739e09ca6b08de00c94f5.

Fixes: eeff625ab31 ("zink: put those shader keys to work fixing up fragment shaders")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7754>

Revert "zink: update shader modules in gfx program when flagged dirty"

This reverts commit e96afeeb7bb2306aa0fba76e27163d4c7a86860b.

Fixes: e96afeeb7bb ("zink: update shader modules in gfx program when flagged dirty")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7754>

radv: Fix RB+ blending for VK_FORMAT_E5B9G9R9_UFLOAT_PACK32.

Fixes: e893102bcf5 ("radv: Add VK_FORMAT_E5B9G9R9_UFLOAT_PACK32 rendering support.")
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7716>

mesa: Synchronize get_gl_override()

Fixes helgrind complaint found with piglit glx-multithread-clearbuffer.
This is a legit race because override[api].version is cleared before
parsing the override string.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7644>

tgsi: Fix helgrind complaint about one-time init

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7644>

gallium/trace: Fix helgrind complaint about one-time init

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7644>

mesa: Fix helgrind complaint about one-time init

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7644>

util: Fix helgrind complaint about one-time init

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7644>

mesa/st: Use do_once for one-time init

Also, since there is a second call-path into st_init_extensions() from
get_version(), add an extra st_debug_init() call.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7644>

gallium/hud: Use do_once for one-time init

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7644>

freedreno/ir3: Use get_once() for one-time init

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7644>

nir: Use get_once() helper for one-time init's

Makes the code more concise, and makes helgrind/drd happy at the same
time!

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7644>

util: Add helpers for various one-time-init patters

A fairly common pattern for debug envvars is something like:

   static int should_print = -1;
   if (should_print < 0)
      should_print = env_var_as_unsigned("NIR_PRINT", 0);

Unfortunately helgrind doesn't realize that we expect to always get the
same return value, so we don't actually care about the race condition
here.

Add a helper get_once() and do_once macros, with extra locking to make
helgrind/drd happy.  Note that other than the nir usages (which are
limited to debug builds), other usages are not in hot-paths.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7644>

util: Add helgrind support for simple_mtx

Annoyingly mtypes.h pulls in simple_mtx, which means we end up needing
to sprinkle a lot of idep_mesautil around.

Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3773
Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Kristian H. Kristensen <hoegsberg@google.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7644>

spirv: Implement SpvOpConvertUToAccelerationStructureKHR

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7734>

spirv: Implement OpTerminateRayKHR and OpIgnoreIntersectionKHR

In the final version of SPV_KHR_ray_tracing, these are now block
terminators like OpKill or OpReturn.  This means that they need special
handling in vtn_cfg.c.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7734>

spirv: Call repair SSA for OpTerminateInvocation

Fixes: 886d2d1a9abcb "spirv: Handle SpvOpTerminateInvocation"

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7734>

spirv: Implement OpTraceRayKHR and OpExecuteCallableKHR

The old NV version (and the provisional KHR version) specified the data
payload via an integer location.  This was quite annoying for the parser
and potentially error-prone.  The final KHR version of the SPIR-V
ray-tracing spec replaces these integers with actual pointers.  We don't
really need to implement the NV versions but we have the code and
someone might want to parse some NV ray-tracing shaders.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7734>

spirv: Update JSON and headers from Khronos main

This corresponds to 104ecc356c1bea4476320faca64440cd1df655a3 ("Merge
pull request #180 from dneto0/issue-179") in
https://github.com/KhronosGroup/SPIRV-Headers

Acked-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7734>

spirv: Rename some ray-tracing intrinsics to NV

For these intrinsics, the NV version and the provisional KHR version
have the same enum value and semantics but the final KHR version is
different on both counts.  Re-name them to NV before we update the
header so the header update isn't a functional change.

Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7734>

vulkan: Update XML and headers to 1.2.162

This brings in the following new extensions:

 - VK_KHR_deferred_host_operations
 - VK_KHR_pipeline_library
 - VK_KHR_acceleration_structure
 - VK_KHR_ray_tracing_pipeline
 - VK_KHR_ray_query

Acked-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Acked-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7734>

zink: fall back to util_blitter for scaled resolves

Vulkan can't scale while resolving using vkCmdResolveImage. For this we
need to use util_blitter.

The reason this wasn't a problem in the past, was that glBlitFramebuffer
always set pipe_blit_info::render_condition_enable, and we always used
that to bail out to util_blitter. When the latter changed, this broke.

Fixes: 19906022e22 ("zink: more accurately track supported blits")
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7745>

nir/opt_peephole_select: respect selection_control when collapsing ifs

Totals from 34 (0.02% of 138013) affected shaders (RAVEN):
CodeSize: 625888 -> 626336 (+0.07%); split: -0.00%, +0.08%
Instrs: 124121 -> 124229 (+0.09%); split: -0.00%, +0.09%
Cycles: 1403072 -> 1403588 (+0.04%); split: -0.01%, +0.04%
VMEM: 5308 -> 5364 (+1.06%); split: +1.07%, -0.02%
Copies: 12773 -> 12838 (+0.51%); split: -0.08%, +0.59%
Branches: 5758 -> 5801 (+0.75%); split: -0.21%, +0.96%

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7478>

nir/opt_peephole_select: collapse nested IFs if applicable

Single-sided nested IFs can sometimes be collapsed
even if they cannot be flattened.
This optimization re-uses block_check_for_allowed_instrs()
to determine if it is beneficial to collapse the IFs.
Additionally, it is required that the phis of the outer IF
become trivial after this optimization, so that no additional
bcsel instructions are added.
This optimization turns

   if (cond1) {
      <allowed instruction>
      if (cond2) {
         <any code>
      } else {
      }
   } else {
   }

into

   <allowed instruction>
   if (cond1 && cond2) {
      <any code>
   } else {
   }

Totals from 17044 (12.35% of 138013) affected shaders (RAVEN):
SGPRs: 1246416 -> 1246256 (-0.01%); split: -0.01%, +0.00%
VGPRs: 802752 -> 802736 (-0.00%); split: -0.01%, +0.01%
SpillSGPRs: 45857 -> 45850 (-0.02%); split: -0.07%, +0.05%
CodeSize: 85318240 -> 85208592 (-0.13%); split: -0.15%, +0.02%
Instrs: 16769049 -> 16738195 (-0.18%); split: -0.20%, +0.02%
Cycles: 947328732 -> 947145796 (-0.02%); split: -0.03%, +0.01%
VMEM: 7271539 -> 7274090 (+0.04%); split: +0.05%, -0.01%
SMEM: 925983 -> 927374 (+0.15%); split: +0.19%, -0.04%
VClause: 294334 -> 294340 (+0.00%); split: -0.00%, +0.00%
SClause: 633600 -> 634048 (+0.07%); split: -0.01%, +0.08%
Copies: 1589650 -> 1580573 (-0.57%); split: -0.66%, +0.09%
Branches: 540830 -> 525767 (-2.79%); split: -2.79%, +0.00%
PreSGPRs: 902500 -> 902415 (-0.01%); split: -0.02%, +0.01%
PreVGPRs: 759992 -> 760019 (+0.00%); split: -0.00%, +0.01%

Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7478>

v3dv: fix allocation size for BO handles

We were allocating twice the size we need for this array. This was
probably caused by a copy and paste error from the GL driver which
grows this dynamically as BOs are added to the job.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7733>

v3dv: remove obsolete comment

We already check that the feature is present in our kernel when we
initialize the physical device.

Reviewed-by: Alejandro Piñeiro <apinheiro@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7731>

clover: Fix typo in comment.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7702>

gallium: fix missing bit field in p_state.h

Marek pointed this out, not sure how we missed it.

Fixes: 3dc6da1ac1f5 ("gallium: add a non-multisample sample mask out behaviour flag.")

Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7741>

lavapipe: enable pipeline stats queries

These pass CTS, but I think are missing some stuff CTS doesn't test.

This is one of the base Vulkan 1.0 features and I'd like to support
it for conformance.

Cc: "20.3" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7705>

lavapipe: fixup mipmap precsion bits

8 seems more correct, however it fixes a bunch of explict lod
tests but breaks some lod query tests.

Cc: "20.3" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7705>

llvmpipe: fix multisample lines.

This also needs another lines fix, but at least align the code
with tri and points

Cc: "20.3" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7705>

llvmpipe: fix multisample point rendering.

Fixes one case in
dEQP-VK.rasterization.primitives_multisample_4_bit.no_stipple.points

Cc: "20.3" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7705>

llvmpipe/setup: move point stats collection earlier.

You have to count the stats pre-culling here.

Just like dc261cdd42380 did for lines.

VK-GL-CTS dEQP-VK.query_pool.statistics_query.clipping_primitives*point_list

Cc: <mesa-stable@lists.freedesktop.org>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7705>

lavapipe: fix wsi acquire fences

Fixes:
dEQP-VK.wsi.xcb.swapchain.acquire.too_many

Cc: "20.3" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7705>

lavapipe: fixup device allocate + enable private data

I'd only half ported private memory support, finish the job.

Cc: "20.3" <mesa-stable@lists.freedesktop.org>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Roland Scheidegger <sroland@vmware.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7705>

zink: fix layered resolves

Until recently, we ended up using u_blitter here, because
info->render_condition_enable was always true here. But when we recently
fixed that overly broad check, this broke.

So let's fix layered-resolves, by actually checking if the resource has
layers respect them in that case, similar to what we do in blit_native.

Fixes: 19906022e22 ("zink: more accurately track supported blits")
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3843
Reviewed-By: Mike Blumenkrantz <michael.blumenkrantz@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7737>

release-calender: Update 20.3

I've been forgetting to remove completed rc's

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7739>

docs: update calendar and link releases notes for 20.2.3

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7739>

docs: Add relnotes for 20.2.3

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7739>

docs: add release notes for 20.2.3

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7739>

aco: optimize v_add+v_lshlrev to v_mad_u32_u24 on GFX6-8

This optimizes v_add(c, v_lshlrev(a, b)) to v_mad_u32_u24(b, 1<<a, c)
if 'a' is a constant (less than or equal to 6 to avoid creating
literals) and 'b' known to be a 16-bit or a 24-bit value.

On GFX9+, this is already optimized to v_lshl_add_u32.

No fossils-db changes.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7673>

aco: optimize v_add+s_lshl to v_mad_u32_u24 on GFX6-8

This optimizes v_add(c, s_lshl(a, b)) to v_mad_u32_u24(a, 1<<b, c)
if 'b' is a constant (less than or equal to 6 to avoid creating
literals) and 'a' known to be a 16-bit or a 24-bit value.

On GFX9+, this is already optimized to v_lshl_add_u32.

fossils-db (Polaris10):
Totals from 1916 (1.36% of 140385) affected shaders:
SGPRs: 88322 -> 87780 (-0.61%); split: -0.66%, +0.05%
CodeSize: 7852668 -> 7851800 (-0.01%); split: -0.01%, +0.00%
Instrs: 1533965 -> 1530459 (-0.23%); split: -0.23%, +0.00%
Cycles: 57001852 -> 56983244 (-0.03%); split: -0.03%, +0.00%
VMEM: 372561 -> 371733 (-0.22%); split: +0.03%, -0.25%
SMEM: 108859 -> 103711 (-4.73%); split: +0.23%, -4.96%
VClause: 37231 -> 37204 (-0.07%)
SClause: 58116 -> 58086 (-0.05%); split: -0.06%, +0.01%
Copies: 199953 -> 199931 (-0.01%); split: -0.03%, +0.02%
Branches: 63478 -> 63477 (-0.00%)
PreSGPRs: 61818 -> 61816 (-0.00%)

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7673>

aco: allow to use the range analysis UB in emit_{sop2,vop2}_instruction()

It will allow to combine v_add+s_lshl or v_add+v_lshlrev to
v_mad_u32_u24 on GFX6-8 if operands are known to be 16-bit or 24-bit.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7673>

aco: add a new Operand flag to indicate that is 24-bit

To indicate that the upper 8-bits are always 0 to optimize more MADs.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7673>

aco/tests: extend the optimize.add_lshl tests to GFX8

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Rhys Perry <pendingchaos02@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7673>

ac,radv: use better export formats for 8-bit when RB+ isn't allowed

When RB+ is enabled, R8_UINT/R8_SINT/R8_UNORM should use FP16_ABGR
for 2x exporting performance. Otherwise, use 32_R to remove useless
instructions needed for 16-bit compressed exports.

fossils-db (Vega10):
Totals from 8858 (6.35% of 139517) affected shaders:
SGPRs: 801248 -> 801210 (-0.00%); split: -0.01%, +0.00%
VGPRs: 596224 -> 596120 (-0.02%); split: -0.02%, +0.01%
CodeSize: 71462452 -> 71356684 (-0.15%); split: -0.15%, +0.00%
MaxWaves: 37097 -> 37105 (+0.02%); split: +0.04%, -0.02%
Instrs: 13963177 -> 13950809 (-0.09%); split: -0.09%, +0.00%
Cycles: 1476539360 -> 1476489996 (-0.00%); split: -0.00%, +0.00%
VMEM: 2363008 -> 2361349 (-0.07%); split: +0.04%, -0.11%
SMEM: 550362 -> 549977 (-0.07%); split: +0.01%, -0.08%
VClause: 245704 -> 245727 (+0.01%); split: -0.01%, +0.02%
SClause: 485161 -> 485104 (-0.01%); split: -0.01%, +0.00%
Copies: 1420034 -> 1422310 (+0.16%); split: -0.01%, +0.17%
Branches: 518710 -> 518705 (-0.00%)
PreSGPRs: 706633 -> 706584 (-0.01%)
PreVGPRs: 547163 -> 547007 (-0.03%); split: -0.03%, +0.01%

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7512>

radv: add new vk_format_is_*() helpers

I think we should make RADV uses util_format everywhere.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7512>

meson: use a feature option for microsoft-clc

It's less code and makes the configuration easier to fine tune.

Fixes: ff05da7f8dc4aa531704d48f718514e3b1fff45d
       ("microsoft: Add CLC frontend and kernel/compute support to DXIL converter")

Reviewed-by: Jesse Natalie <jenatali@microsoft.com>
Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7699>

meson: Don't add extra values to shader-cache

We're trying to move to using a feature here, adding more values breaks
that.

Fixes: 5de56937a3d009659076dc67de6a57379fc7a31b
       ("disk_cache: build option for disabled-by-default")

Reviewed-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7699>

freedreno/ir3: Fix valgrind complaint about streamout state

The warning is a bit misleading about where it shows up.. it complains
about the shader key, due to shader key being calculated from (among
other things) stream_output state that had some uninitialized garbage
in the padding.

==84572== Uninitialised byte(s) found during client check request
==84572==    at 0x60548E8: blob_write_bytes (blob.c:163)
==84572==    by 0x6534EF7: compute_variant_key (ir3_disk_cache.c:111)
==84572==    by 0x6535143: ir3_disk_cache_retrieve (ir3_disk_cache.c:171)
==84572==    by 0x654D82F: create_variant (ir3_shader.c:251)
==84572==    by 0x654DA2B: ir3_shader_get_variant (ir3_shader.c:301)
==84572==    by 0x645B2CB: ir3_shader_variant (ir3_gallium.c:113)
==84572==    by 0x645B7EB: ir3_shader_create (ir3_gallium.c:219)
==84572==    by 0x645BAA7: ir3_shader_state_create (ir3_gallium.c:285)
==84572==    by 0x6506003: fd6_shader_state_create (fd6_program.c:1136)
==84572==    by 0x64676C7: assemble_tgsi (freedreno_program.c:105)
==84572==    by 0x64679DF: fd_prog_init (freedreno_program.c:188)
==84572==    by 0x6506157: fd6_prog_init (fd6_program.c:1172)
==84572==  Address 0xeff1588 is 424 bytes inside a block of size 480 alloc'd
==84572==    at 0x4866FA4: malloc (vg_replace_malloc.c:307)
==84572==    by 0x605D46F: ralloc_size (ralloc.c:133)
==84572==    by 0x605D52F: rzalloc_size (ralloc.c:166)
==84572==    by 0x654DFF7: ir3_shader_from_nir (ir3_shader.c:473)
==84572==    by 0x645B6C7: ir3_shader_create (ir3_gallium.c:182)
==84572==    by 0x645BAA7: ir3_shader_state_create (ir3_gallium.c:285)
==84572==    by 0x6506003: fd6_shader_state_create (fd6_program.c:1136)
==84572==    by 0x64676C7: assemble_tgsi (freedreno_program.c:105)
==84572==    by 0x64679DF: fd_prog_init (freedreno_program.c:188)
==84572==    by 0x6506157: fd6_prog_init (fd6_program.c:1172)
==84572==    by 0x64CB36F: fd6_context_create (fd6_context.c:154)
==84572==    by 0x59D93BB: st_api_create_context (st_manager.c:917)

Somehow this was showing up with dEQP-GLES31.info.vendor but not other
things.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7717>

freedreno/drm: Quiet timedout error msg

This isn't terribly interesting, but got more chatty when we converted
to mesa_loge() vs debug_printf()

Fixes: 156d7e45f74 ("freedreno: Convert to mesa_log*()")
Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7717>

freedreno/a6xx: Clear control mem at context create

We could be getting a recycled bo containing random garbage, which can
confuse check_vsc_overflow().

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7717>

freedreno: Convert one last mtx_t -> simple_mtx_t

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7717>

freedreno: emit_marker() cleanup

1) Propagate the change to only emit markers in debug builds (and add
   the WFI that ensures they are synchronized with GPU.  We could
   consider dropping them entirely, since the GPU devcoredump support
   in newer kernels is more useful.  But it is still an occasionally
   useful fallback.

2) Use p_atomic_inc_return() to placate helgrind

Signed-off-by: Rob Clark <robdclark@chromium.org>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7717>

mesa: add an environment variable to default enable INTEL_blackhole

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7640>

st: trigger noop if the default value is not true

v2: Verify that PIPE_CAP_FRONTEND_NOOP is available before calling vfunc (Icecream95)

Signed-off-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com> (v1)
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7640>

ir3/ra: Fix array reg liveness in scalar pass

Assigning an array reg removes IR3_REG_ARRAY, which means that
definitions and uses can't be tracked back to the array register's name
and liveness for the components of the array aren't correctly
calculated. To fix this we delay assigning array registers until the
scalar pass.

Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7711>

nir: fix gathering cross invocation info

Fixes: 5b77b14448c ("nir: Use src_is_invocation_id in get_deref_info.")
Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7730>

swr: Pass draw start information to state update mechanism

This fixes crash in many workloads/tests

Reviewed-by: Krzysztof Raszkowski <krzysztof.raszkowski@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7728>

ci: fix name of the Sienna Cichlid expected failures file

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7729>

v3dv/cmd_buffer: missing (uint8_t *) casting when calling memcmp

Caused to return early wrongly on CmdPushConstants with some tests
using several calls to that method. As we are here we are also
replacing the (void *) casting at the memcpy below.

Fixes: e1c8041cde64 ("v3dv: try harder to skip emission of redundant state")

Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7718>

radv: dump BO ranges into bo_ranges.log instead of stderr

Like other dumps during GPU hang detection.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7706>

radv: add RADV_DEBUG=noumr to disable UMR logs during GPU hang detection

Sometimes UMR logs can't be dumped and you would get permission
denied, even if the UMR binary has the setuid bit enabled.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7706>

radv: dump application info in the GPU hang report

Like the name, version, as well as the engine and the API version.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7706>

radv: append a time string to the hang report dump directory

Using the PID only isn't really informative.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7706>

radv: print more debug messages when generating a hang report

If for some reasons the driver can't generate the hang report properly.

Signed-off-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Reviewed-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7706>

radeonsi: don't allocate LDS for TCS inputs if it's not used

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

radeonsi: don't insert barrier between VS/TCS if all TCS inputs come from VGPRs

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

radeonsi: pass VS->TCS IO via VGPRs if VS and TCS have the same thread count

It can only be done if a TCS input is accessed without indirect indexing and
with gl_InvocationID as the vertex index, and the number of VS and TCS threads
is the same.

This eliminates LDS stores and loads for VS->TCS IO, reducing shader lifetime
and LDS traffic.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

ac/llvm: prepare for passing VS->TCS IO via VGPRs

- bump AC_MAX_ARGS
- add vertex_index_is_invoc_id parameter into load_tess_varyings

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

radeonsi: remove unnecessary NULL checking in NIR tess functions

param_index is always checked for non-NULL later.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

radeonsi: if VS and TCS have the same number of threads, merge the conditonals

Instead of:
    if (VS) {
VS;
    }
    if (TCS) {
TCS;
    }

Do this if the number of threads is the same in VS and TCS:
    exec = enabled_threads;
    VS;
    TCS;

Skipping declare_vb_descriptor_input_sgprs is needed to match the VS return
values.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

radeonsi: always return void from si_build_wrapper_function

It's the end of the shader, there are no return values.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

radeonsi: merge TCS and TCS epilog conditional blocks

Instead of:
    if (TCS) {
       TCS;
    }
    if (TCS && epilog) {
       epilog;
    }

Do:
    if (TCS) {
TCS;
if (epilog) {
    epilog;
    }

Only monolithic shaders can do it.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

radeonsi: don't generate a dead conditional in si_write_tess_factors on gfx9+

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

radeonsi: limit HS LDS usage per workgroup to 16K to allow at least 2 WGs/CU

This increases occupancy when the LDS size is e.g. 20K for 3 waves.
If we limit the size to 16K, we can fit 2 workgroups with 2 waves each,
so 4 waves in total.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

radeonsi: don't allocate LDS for TCS outputs if they are not read

This reduces LDS usage by 50% in Unigine Heaven.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

radeonsi: don't leave more than 8 unoccupied lanes in HS

Previously it was 16 and bigger patches would always trim the patch count
needlessly.

There are 2 variables to consider:
- lane occupancy
- LDS usage (limiting wave occupancy)

If LDS size is 32 KB (max limit per CU) for 3 waves and we can't maximize
occupancy, it's better to leave some lanes unoccupied because using 2
waves would decrease the LDS size to 21 KB, which is not enough to fit
another workgroup on the CU.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

radeonsi: adjust tess SGPRs to allow fully occupied 3 HS waves of triangles

With triangles and 3 HS waves, 3 lanes were unoccupied. Adjust the SGPR
encoding to allow 1 more triangle to fit there.

Some of the fields are not large enough, but they weren't large enough
before either.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

ac/nir: fix a typo in ac_are_tessfactors_def_in_all_invocs

I think it only made the pass return false if there was a barrier

Fixes: 2832bc972bf - ac/nir_to_llvm: add ac_are_tessfactors_def_in_all_invocs()

Reviewed-by: Timothy Arceri <tarceri@itsqueeze.com>
Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7623>

r600/sfn: Correctly lower all int64

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7724>

nv50/ir: Initialize Program members in constructor.

Fix defects reported by Coverity Scan.

uninit_member: Non-static class member tlsSize is not initialized in this constructor nor in any functions that it calls.
uninit_member: Non-static class member driver is not initialized in this constructor nor in any functions that it calls.
uninit_member: Non-static class member driver_out is not initialized in this constructor nor in any functions that it calls.

Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7703>

r600/sfn: use a per stream index register in GS

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7714>

r600/sfn: lower bool to int32 only after common optimizations

Fixes: f79b7fcf7c7f5db626efdb63f27e8bc64d0aed77
   r600/sfn: use 32 bit bools

Signed-off-by: Gert Wollny <gert.wollny@collabora.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7714>

nir: Use src_is_invocation_id in get_deref_info.

Signed-off-by: Timur Kristof <timur.kristof@gmail.com>
Reviewed-by: Marek Olšák <marek.olsak@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7622>

nir: fix gathering patch IO usage with lowered IO

Fixes: 17af07024df - nir: gather all IO info from IO intrinsics

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7622>

nir: fix gathering TCS cross invocation access with lowered IO

Fixes: abe9588ff04 - nir: gather tess.tcs_cross_invocation info from lowered IO intrinsics

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7622>

mesa: lock Shared->TexMutex only once for a glthread batch

This removes a lot of locking from the driver thread.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7053>

mesa: lock Shared->BufferObjects only once for a glthread batch

This removes a lot of locking from the driver thread.

If multiple contexts sharing buffers submit GL calls from multiple threads,
they will be serialized by this mutex. I can add a driconf option to turn
off this optimization if needed, but I currently don't anticipate to see
GL apps that use multiple shared contexts in different threads
simultaneously.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7053>

glthread: make glGetActiveUniform return without syncing

We just need to track glLinkProgram and glDeleteProgram.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7053>

mesa: make error handling for glGetActiveUniform glthread-safe

glGetActiveUniform will be called from the app thread directly.
This is safe if a few conditions are met.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7053>

mesa: add glInternalSetError for glthread

glthread wants to set GL errors, but has to do it by adding the SetError
call into the queue for it to be thread-safe.

Acked-by: Pierre-Eric Pelloux-Prayer <pierre-eric.pelloux-prayer@amd.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7053>

radv: Dump BO VA ranges on hang.

To make it easier to figure out if a given VA is valid or not.

Gitlab: https://gitlab.freedesktop.org/mesa/mesa/-/issues/3620
Reviewed-by: Samuel Pitoiset <samuel.pitoiset@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7205>

clover: Initialize command_queue member _props.

Fix defect reported by Coverity Scan.

Uninitialized scalar variable (UNINIT)
uninit_use: Using uninitialized value this->_props.
    _props |= properties[i + 1];

Fixes: e42a7fa0378 ("clover: add support command queue properties")
Signed-off-by: Vinson Lee <vlee@freedesktop.org>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Karol Herbst <kherbst@redhat.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/7701>

meson: drop deprecated EGL platform build options

These two options were deprecated and announced to be removed in 20.3,
so let's drop them now.

Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Acked-by: Emil Velikov <emil.velikov@collabora.com>
Acked-by: Matt Turner <mattst88@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Cc: 20.3 <mesa-stable>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5844>

gitlab-ci: drop deprecated platforms that snuck in when nobody was watching

... even though I did review that commit. My bad :]

Fixes: c56f09124b195c5cbaab ("gitlab-ci: Move classic driver testing to a new meson-classic job")
Signed-off-by: Eric Engestrom <eric@engestrom.ch>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5844>

aco/tests: add Builder::v_mul_imm() tests

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Timur Kristóf <timur.kristof@gmail.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5390>