OpenHarmony/third_party_mesa3d
2
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

spirv: don't split memory barriers

Browse Source 
If the SPIR-V had a shared+image memory barrier, we would emit two NIR
barriers: a shared barrier and an image barrier.

Unlike a single barrier, two barriers allows transformations such as:

intrinsic image_deref_store (ssa_27, ssa_33, ssa_34, ssa_32, ssa_25) (1)
intrinsic memory_barrier_shared () ()
intrinsic memory_barrier_image () ()
intrinsic store_shared (ssa_35, ssa_24) (0, 1, 4, 0)
->
intrinsic memory_barrier_shared () ()
intrinsic store_shared (ssa_35, ssa_24) (0, 1, 4, 0)
intrinsic image_deref_store (ssa_27, ssa_33, ssa_34, ssa_32, ssa_25) (1)
intrinsic memory_barrier_image () ()

This commit fixes two dEQP-VK.memory_model.* CTS tests with ACO.

Signed-off-by: Rhys Perry <pendingchaos02@gmail.com>
Reviewed-by: Caio Marcelo de Oliveira Filho <caio.oliveira@intel.com>
Part-of: <https://gitlab.freedesktop.org/mesa/mesa/-/merge_requests/5951>
This commit is contained in:
Rhys Perry
2020-07-17 11:46:47 +01:00
committed by Marge Bot
parent 28c227c7ca
commit fac813dc61
5 changed files with 35 additions and 40 deletions
Download Patch File Download Diff File Expand all files Collapse all files

2
.gitlab-ci/deqp-radv-navi10-aco-fails.txt
View File

@@ -2,8 +2,6 @@
dEQP-VK.transform_feedback.simple.multistreams_1
dEQP-VK.transform_feedback.simple.multistreams_3
dEQP-VK.memory_model.message_passing.core11.u32.coherent.fence_fence.atomicwrite.device.payload_local.image.guard_nonlocal.workgroup.comp
dEQP-VK.memory_model.message_passing.core11.u32.coherent.fence_fence.atomicwrite.device.payload_nonlocal.image.guard_nonlocal.workgroup.comp
dEQP-VK.rasterization.flatshading.line_strip_wide
dEQP-VK.rasterization.flatshading.non_strict_line_strip_wide
dEQP-VK.rasterization.flatshading.non_strict_lines_wide

2
.gitlab-ci/deqp-radv-navi14-aco-fails.txt
View File

@@ -2,8 +2,6 @@
dEQP-VK.transform_feedback.simple.multistreams_1
dEQP-VK.transform_feedback.simple.multistreams_3
dEQP-VK.memory_model.message_passing.core11.u32.coherent.fence_fence.atomicwrite.device.payload_local.image.guard_nonlocal.workgroup.comp
dEQP-VK.memory_model.message_passing.core11.u32.coherent.fence_fence.atomicwrite.device.payload_nonlocal.image.guard_nonlocal.workgroup.comp
dEQP-VK.rasterization.flatshading.line_strip_wide
dEQP-VK.rasterization.flatshading.non_strict_line_strip_wide
dEQP-VK.rasterization.flatshading.non_strict_lines_wide

2
.gitlab-ci/deqp-radv-pitcairn-aco-fails.txt
View File

@@ -1,5 +1,3 @@
dEQP-VK.memory_model.message_passing.core11.u32.coherent.fence_fence.atomicwrite.device.payload_local.image.guard_nonlocal.workgroup.comp
dEQP-VK.memory_model.message_passing.core11.u32.coherent.fence_fence.atomicwrite.device.payload_nonlocal.image.guard_nonlocal.workgroup.comp
dEQP-VK.pipeline.depth.format.d16_unorm.compare_ops.never_zerodepthbounds_depthdisabled_stencilenabled
dEQP-VK.pipeline.depth.format.d32_sfloat.compare_ops.never_zerodepthbounds_depthdisabled_stencilenabled
dEQP-VK.rasterization.flatshading.line_strip_wide

2
.gitlab-ci/deqp-radv-raven-aco-fails.txt
View File

@@ -2,8 +2,6 @@
dEQP-VK.transform_feedback.simple.multistreams_1
dEQP-VK.transform_feedback.simple.multistreams_3
dEQP-VK.memory_model.message_passing.core11.u32.coherent.fence_fence.atomicwrite.device.payload_local.image.guard_nonlocal.workgroup.comp
dEQP-VK.memory_model.message_passing.core11.u32.coherent.fence_fence.atomicwrite.device.payload_nonlocal.image.guard_nonlocal.workgroup.comp
dEQP-VK.rasterization.flatshading.line_strip_wide
dEQP-VK.rasterization.flatshading.non_strict_line_strip_wide
dEQP-VK.rasterization.flatshading.non_strict_lines_wide

31
src/compiler/spirv/spirv_to_nir.c
View File

@@ -3578,21 +3578,25 @@ vtn_emit_memory_barrier(struct vtn_builder *b, SpvScope scope,
/* There's only two scopes thing left */
vtn_assert(scope == SpvScopeInvocation || scope == SpvScopeDevice);
/* Map the GLSL memoryBarrier() construct to the corresponding NIR one. */
static const SpvMemorySemanticsMask glsl_memory_barrier =
SpvMemorySemanticsUniformMemoryMask |
SpvMemorySemanticsWorkgroupMemoryMask |
SpvMemorySemanticsImageMemoryMask;
if ((semantics & glsl_memory_barrier) == glsl_memory_barrier) {
/* Map the GLSL memoryBarrier() construct and any barriers with more than one
* semantic to the corresponding NIR one.
*/
if (util_bitcount(semantics & all_memory_semantics) > 1) {
vtn_emit_barrier(b, nir_intrinsic_memory_barrier);
semantics &= ~(glsl_memory_barrier | SpvMemorySemanticsAtomicCounterMemoryMask);
if (semantics & SpvMemorySemanticsOutputMemoryMask) {
/* GLSL memoryBarrier() (and the corresponding NIR one) doesn't include
* TCS outputs, so we have to emit it's own intrinsic for that. We
* then need to emit another memory_barrier to prevent moving
* non-output operations to before the tcs_patch barrier.
*/
vtn_emit_barrier(b, nir_intrinsic_memory_barrier_tcs_patch);
vtn_emit_barrier(b, nir_intrinsic_memory_barrier);
}
return;
}
/* Issue a bunch of more specific barriers */
uint32_t bits = semantics;
while (bits) {
SpvMemorySemanticsMask semantic = 1 << u_bit_scan(&bits);
switch (semantic) {
/* Issue a more specific barrier */
switch (semantics & all_memory_semantics) {
case SpvMemorySemanticsUniformMemoryMask:
vtn_emit_barrier(b, nir_intrinsic_memory_barrier_buffer);
break;
@@ -3610,8 +3614,7 @@ vtn_emit_memory_barrier(struct vtn_builder *b, SpvScope scope,
vtn_emit_barrier(b, nir_intrinsic_memory_barrier_tcs_patch);
break;
default:
break;;
}
break;
}
}