[PATCH] D64981: [AMDGPU] Save some work when an atomic op has no uses

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added a comment.

Test?



================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUAtomicOptimizer.cpp:494
 
-  // Create a PHI node to get our new atomic result into the exit block.
-  PHINode *const PHI = B.CreatePHI(Ty, 2);
-  PHI->addIncoming(UndefValue::get(Ty), EntryBB);
-  PHI->addIncoming(NewI, SingleLaneTerminator->getParent());
-
-  // We need to broadcast the value who was the lowest active lane (the first
-  // lane) to all other lanes in the wavefront. We use an intrinsic for this,
-  // but have to handle 64-bit broadcasts with two calls to this intrinsic.
-  Value *BroadcastI = nullptr;
-
-  if (TyBitWidth == 64) {
-    Value *const ExtractLo = B.CreateTrunc(PHI, B.getInt32Ty());
-    Value *const ExtractHi =
-        B.CreateTrunc(B.CreateLShr(PHI, B.getInt64(32)), B.getInt32Ty());
-    CallInst *const ReadFirstLaneLo =
-        B.CreateIntrinsic(Intrinsic::amdgcn_readfirstlane, {}, ExtractLo);
-    CallInst *const ReadFirstLaneHi =
-        B.CreateIntrinsic(Intrinsic::amdgcn_readfirstlane, {}, ExtractHi);
-    Value *const PartialInsert = B.CreateInsertElement(
-        UndefValue::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) {
-
-    BroadcastI = B.CreateIntrinsic(Intrinsic::amdgcn_readfirstlane, {}, PHI);
-  } else {
-    llvm_unreachable("Unhandled atomic bit width");
-  }
+  const bool NeedResult = !I.use_empty();
+  if (NeedResult) {
----------------
Not sure if this ignores debug uses, but it probably needs to


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D64981/new/

https://reviews.llvm.org/D64981