[llvm] [AMDGPU][True16][CodeGen] do not legalize t16 operand during user scan (PR #145450)

Joe Nash via llvm-commits llvm-commits at lists.llvm.org
Tue Jun 24 12:09:10 PDT 2025


================
@@ -8733,20 +8733,20 @@ void SIInstrInfo::addUsersToMoveToVALUWorklist(
     case AMDGPU::INSERT_SUBREG:
       break;
     default:
-      OpNo = I.getOperandNo();
+      OpNo = MO.getOperandNo();
       break;
     }
 
     if (!RI.hasVectorRegisters(getOpRegClass(UseMI, OpNo))) {
-      Worklist.insert(&UseMI);
+      if (&UseMI == prevMI)
----------------
Sisyph wrote:

You shouldn't need this. SetVector should handle not inserting duplicate elements. Also, I'm not sure the iteration order of use_operands groups by instruction.

https://github.com/llvm/llvm-project/pull/145450


More information about the llvm-commits mailing list