[PATCH] D121613: [LV] Use usesScalars in widenPHIInstruction.

Thu Mar 17 04:59:33 PDT 2022

fhahn marked an inline comment as done.
fhahn added inline comments.

================
Comment at: llvm/test/Transforms/LoopVectorize/AArch64/sve-widen-gep.ll:147
+; CHECK-NEXT:    [[NEXT_GEP_0:%.*]] = getelementptr i8, i8* [[NEXT_GEP]], i32 0
+; CHECK-NEXT:    [[TMP16:%.*]] = bitcast i8* [[NEXT_GEP_0]] to <vscale x 2 x i8>*
 ; CHECK-NEXT:    [[WIDE_LOAD:%.*]] = load <vscale x 2 x i8>, <vscale x 2 x i8>* [[TMP16]], align 1
----------------
Ayal wrote:
> This is a case where we now know all users use scalars more accurately, where previously we generated vectors redundantly?
Yes exactly. Previously we had to compute the vector GEP and then just extract lane 0.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D121613/new/

https://reviews.llvm.org/D121613