[PATCH] D26790: [X86] Add a hasOneUse check to selectScalarSSELoad to keep the same load from being folded multiple times

[Craig Topper via llvm-commits]

craig.topper added a comment.

There's a potential correctness problem here too. This IR finishes isel with the store only depending on the chain from the minss instruction but not the maxss instruction. So the duplication of the load breaks the chain dependency.

define void @double_fold(float* %x, <4 x float> %y, <4 x float>* %z) {
entry:

  %0 = load float, float* %x, align 1
  %vecinit.i = insertelement <4 x float> undef, float %0, i32 0
  %1 = tail call <4 x float> @llvm.x86.sse.min.ss(<4 x float> %y, <4 x float> %vecinit.i)
  %2 = tail call <4 x float> @llvm.x86.sse.max.ss(<4 x float> %y, <4 x float> %vecinit.i)
  %3 = fadd <4 x float> %1, %2
  store <4 x float> %3, <4 x float>* %z
  ret void

}

declare <4 x float> @llvm.x86.sse.min.ss(<4 x float>, <4 x float>)

declare <4 x float> @llvm.x86.sse.max.ss(<4 x float>, <4 x float>)


https://reviews.llvm.org/D26790