[llvm] [AMDGPU] Cleanup bitcast spam in atomic optimizer (PR #96933)

[Vikram Hegde via llvm-commits]

================
@@ -922,26 +887,8 @@ void AMDGPUAtomicOptimizerImpl::optimizeAtomic(Instruction &I,
     // but have to handle 64-bit broadcasts with two calls to this intrinsic.
     Value *BroadcastI = nullptr;
 
-    if (TyBitWidth == 64) {
-      Value *CastedPhi = B.CreateBitCast(PHI, IntNTy);
-      Value *const ExtractLo = B.CreateTrunc(CastedPhi, Int32Ty);
-      Value *const ExtractHi =
-          B.CreateTrunc(B.CreateLShr(CastedPhi, 32), Int32Ty);
-      CallInst *const ReadFirstLaneLo = B.CreateIntrinsic(
-          Int32Ty, Intrinsic::amdgcn_readfirstlane, ExtractLo);
-      CallInst *const ReadFirstLaneHi = B.CreateIntrinsic(
-          Int32Ty, Intrinsic::amdgcn_readfirstlane, ExtractHi);
-      Value *const PartialInsert = B.CreateInsertElement(
-          PoisonValue::get(VecTy), ReadFirstLaneLo, B.getInt32(0));
-      Value *const Insert =
-          B.CreateInsertElement(PartialInsert, ReadFirstLaneHi, B.getInt32(1));
-      BroadcastI = B.CreateBitCast(Insert, Ty);
-    } else if (TyBitWidth == 32) {
-      Value *CastedPhi = B.CreateBitCast(PHI, IntNTy);
-      BroadcastI =
-          B.CreateIntrinsic(IntNTy, Intrinsic::amdgcn_readfirstlane, CastedPhi);
-      BroadcastI = B.CreateBitCast(BroadcastI, Ty);
-
+    if (TyBitWidth == 32 || TyBitWidth == 64) {
----------------
vikramRH wrote:

done

https://github.com/llvm/llvm-project/pull/96933