[llvm-branch-commits] [llvm] [AMDGPU] Disable VALU sinking and hoisting with WWM (PR #123124)

Diana Picus via llvm-branch-commits llvm-branch-commits at lists.llvm.org
Fri Jan 17 02:12:55 PST 2025


================
@@ -2773,6 +2773,9 @@ void AMDGPUDAGToDAGISel::SelectINTRINSIC_WO_CHAIN(SDNode *N) {
   case Intrinsic::amdgcn_wwm:
   case Intrinsic::amdgcn_strict_wwm:
     Opcode = AMDGPU::STRICT_WWM;
+    CurDAG->getMachineFunction()
+        .getInfo<SIMachineFunctionInfo>()
+        ->setInitWholeWave();
----------------
rovka wrote:

I would not recommend using HasInitWholeWave for this. That has a very narrow meaning referring only to the use of the llvm.amdgcn.init.whole.wave intrinsic, not WWM in general. 

https://github.com/llvm/llvm-project/pull/123124


More information about the llvm-branch-commits mailing list