[PATCH] D79641: [AMDGPU] Vectorize alloca thru bitcast

[Matt Arsenault via Phabricator via llvm-commits]

arsenm added inline comments.


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp:310
+    V = I->getOperand(0);
+  }
+  return V;
----------------
I think this needs to be careful around multiple uses


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp:435
+        for (User *CastUser : Inst->users()) {
+          if (isAssumeLikeIntrinsic(cast<Instruction>(CastUser)))
+            continue;
----------------
Needs test with assume intrinsic


================
Comment at: llvm/lib/Target/AMDGPU/AMDGPUPromoteAlloca.cpp:481
       Value *ExtractElement = Builder.CreateExtractElement(VecValue, Index);
+      if (Inst->getType() != VecEltTy)
+        ExtractElement = Builder.CreateBitCast(ExtractElement, Inst->getType());
----------------
IRBuilder does this check for you so you can omit it


================
Comment at: llvm/test/CodeGen/AMDGPU/vector-alloca-bitcast.ll:28
+  %tmp = alloca [4 x i32], addrspace(5)
+  %x = bitcast [4 x i32] addrspace(5)* %tmp to i32 addrspace(5)*
+  %y = getelementptr [4 x i32], [4 x i32] addrspace(5)* %tmp, i32 0, i32 1
----------------
Needs tests with multiple uses


CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D79641/new/

https://reviews.llvm.org/D79641