spirv: don't split memory barriers

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>

diff --git a/.gitlab-ci/deqp-radv-navi10-aco-fails.txt b/.gitlab-ci/deqp-radv-navi10-aco-fails.txt
index d5e9453b0c58590b07fe0eddf4249eb092a8d6bf..1f4a94c4d9c853aa9cd1b115676870c41240f7fa 100644 (file)
--- a/.gitlab-ci/deqp-radv-navi10-aco-fails.txt
+++ b/.gitlab-ci/deqp-radv-navi10-aco-fails.txt
@@ -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

diff --git a/.gitlab-ci/deqp-radv-navi14-aco-fails.txt b/.gitlab-ci/deqp-radv-navi14-aco-fails.txt
index d5e9453b0c58590b07fe0eddf4249eb092a8d6bf..1f4a94c4d9c853aa9cd1b115676870c41240f7fa 100644 (file)
--- a/.gitlab-ci/deqp-radv-navi14-aco-fails.txt
+++ b/.gitlab-ci/deqp-radv-navi14-aco-fails.txt
@@ -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

diff --git a/.gitlab-ci/deqp-radv-pitcairn-aco-fails.txt b/.gitlab-ci/deqp-radv-pitcairn-aco-fails.txt
index dd37cbc223ab474a1ecfd4a8793a89a0da50e4fb..5ec617dcae798abd4324123969ffd8dc17d5bd6d 100644 (file)
--- a/.gitlab-ci/deqp-radv-pitcairn-aco-fails.txt
+++ b/.gitlab-ci/deqp-radv-pitcairn-aco-fails.txt
@@ -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

diff --git a/.gitlab-ci/deqp-radv-raven-aco-fails.txt b/.gitlab-ci/deqp-radv-raven-aco-fails.txt
index d5e9453b0c58590b07fe0eddf4249eb092a8d6bf..1f4a94c4d9c853aa9cd1b115676870c41240f7fa 100644 (file)
--- a/.gitlab-ci/deqp-radv-raven-aco-fails.txt
+++ b/.gitlab-ci/deqp-radv-raven-aco-fails.txt
@@ -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

diff --git a/src/compiler/spirv/spirv_to_nir.c b/src/compiler/spirv/spirv_to_nir.c
index b9f35ad0262627767134c0a3b82dc078f9667f87..e98728396fc7073a881ed286d64733d4c8b83389 100644 (file)
--- a/src/compiler/spirv/spirv_to_nir.c
+++ b/src/compiler/spirv/spirv_to_nir.c
@@ -3578,40 +3578,43 @@ 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) {
-       vtn_emit_barrier(b, nir_intrinsic_memory_barrier);
-       semantics &= ~(glsl_memory_barrier | SpvMemorySemanticsAtomicCounterMemoryMask);
-   }
-
-   /* Issue a bunch of more specific barriers */
-   uint32_t bits = semantics;
-   while (bits) {
-      SpvMemorySemanticsMask semantic = 1 << u_bit_scan(&bits);
-      switch (semantic) {
-      case SpvMemorySemanticsUniformMemoryMask:
-         vtn_emit_barrier(b, nir_intrinsic_memory_barrier_buffer);
-         break;
-      case SpvMemorySemanticsWorkgroupMemoryMask:
-         vtn_emit_barrier(b, nir_intrinsic_memory_barrier_shared);
-         break;
-      case SpvMemorySemanticsAtomicCounterMemoryMask:
-         vtn_emit_barrier(b, nir_intrinsic_memory_barrier_atomic_counter);
-         break;
-      case SpvMemorySemanticsImageMemoryMask:
-         vtn_emit_barrier(b, nir_intrinsic_memory_barrier_image);
-         break;
-      case SpvMemorySemanticsOutputMemoryMask:
-         if (b->nb.shader->info.stage == MESA_SHADER_TESS_CTRL)
-            vtn_emit_barrier(b, nir_intrinsic_memory_barrier_tcs_patch);
-         break;
-      default:
-         break;;
+   /* 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);
+      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 more specific barrier */
+   switch (semantics & all_memory_semantics) {
+   case SpvMemorySemanticsUniformMemoryMask:
+      vtn_emit_barrier(b, nir_intrinsic_memory_barrier_buffer);
+      break;
+   case SpvMemorySemanticsWorkgroupMemoryMask:
+      vtn_emit_barrier(b, nir_intrinsic_memory_barrier_shared);
+      break;
+   case SpvMemorySemanticsAtomicCounterMemoryMask:
+      vtn_emit_barrier(b, nir_intrinsic_memory_barrier_atomic_counter);
+      break;
+   case SpvMemorySemanticsImageMemoryMask:
+      vtn_emit_barrier(b, nir_intrinsic_memory_barrier_image);
+      break;
+   case SpvMemorySemanticsOutputMemoryMask:
+      if (b->nb.shader->info.stage == MESA_SHADER_TESS_CTRL)
+         vtn_emit_barrier(b, nir_intrinsic_memory_barrier_tcs_patch);
+      break;
+   default:
+      break;
    }
 }