[llvm-branch-commits] [llvm] [AMDGPU] Fix duplicate s_wait_asynccnt on gfx12-plus (PR #190777)

[via llvm-branch-commits]

llvmbot wrote:


<!--LLVM PR SUMMARY COMMENT-->

@llvm/pr-subscribers-backend-amdgpu

Author: Sameer Sahasrabuddhe (ssahasra)

<details>
<summary>Changes</summary>

S_WAIT_ASYNCCNT was missing from counterTypeForInstr(), so isWaitInstr() did not recognize it as a wait instruction. On the fixpoint algorithm's second pass over a loop body, the already-inserted S_WAIT_ASYNCCNT was treated as a normal instruction, causing WAIT_ASYNCMARK to be re-processed and a duplicate S_WAIT_ASYNCCNT to be emitted.

Assisted-By: Claude Opus 4.6

---
Full diff: https://github.com/llvm/llvm-project/pull/190777.diff


2 Files Affected:

- (modified) llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp (+2) 
- (modified) llvm/test/CodeGen/AMDGPU/asyncmark-gfx12plus.ll (-1) 


``````````diff
diff --git a/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp b/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
index eb0b12c3b5bc1..3d6cd101274b2 100644
--- a/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
+++ b/llvm/lib/Target/AMDGPU/SIInsertWaitcnts.cpp
@@ -1745,6 +1745,8 @@ static std::optional<InstCounterType> counterTypeForInstr(unsigned Opcode) {
     return KM_CNT;
   case AMDGPU::S_WAIT_XCNT:
     return X_CNT;
+  case AMDGPU::S_WAIT_ASYNCCNT:
+    return ASYNC_CNT;
   default:
     return {};
   }
diff --git a/llvm/test/CodeGen/AMDGPU/asyncmark-gfx12plus.ll b/llvm/test/CodeGen/AMDGPU/asyncmark-gfx12plus.ll
index cfb296fb2d529..18064c6e60a77 100644
--- a/llvm/test/CodeGen/AMDGPU/asyncmark-gfx12plus.ll
+++ b/llvm/test/CodeGen/AMDGPU/asyncmark-gfx12plus.ll
@@ -248,7 +248,6 @@ define amdgpu_kernel void @test_pipelined_loop_with_global(ptr addrspace(1) %foo
 ; GFX1250-NEXT:    ; asyncmark
 ; GFX1250-NEXT:    ; wait_asyncmark(2)
 ; GFX1250-NEXT:    s_wait_asynccnt 0x2
-; GFX1250-NEXT:    s_wait_asynccnt 0x2
 ; GFX1250-NEXT:    s_add_co_i32 s8, s8, 1
 ; GFX1250-NEXT:    s_add_co_i32 s9, s9, 4
 ; GFX1250-NEXT:    ds_load_b32 v9, v9

``````````

</details>


https://github.com/llvm/llvm-project/pull/190777