[llvm] [AMDGPU] Exclude implicit uses and defs in SIInsertWaitcnts. NFC. (PR #108953)

Stanislav Mekhanoshin via llvm-commits llvm-commits at lists.llvm.org
Tue Sep 17 03:30:39 PDT 2024


================
@@ -2326,7 +2326,7 @@ bool SIInsertWaitcnts::shouldFlushVmCnt(MachineLoop *ML,
         if (MI.mayStore())
           HasVMemStore = true;
       }
-      for (const MachineOperand &Op : MI.all_uses()) {
+      for (const MachineOperand &Op : MI.uses()) {
----------------
rampitec wrote:

Same here, I am excluding VGPR implicit uses, which are never real. Besides the whole function is an opportunistic optimization.

https://github.com/llvm/llvm-project/pull/108953


More information about the llvm-commits mailing list