[PATCH] D133762: [VPlan] Remove dead recipes before sinking.

Wed Oct 12 04:47:03 PDT 2022

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

================
Comment at: llvm/test/Transforms/LoopVectorize/pointer-induction.ll:39
 ; CHECK:       pred.store.if:
-; CHECK-NEXT:    store i8 95, i8* [[TMP4]], align 1
+; CHECK-NEXT:    [[TMP11:%.*]] = getelementptr inbounds i8, i8* [[NEXT_GEP]], i64 -1
+; CHECK-NEXT:    store i8 95, i8* [[TMP11]], align 1
----------------
Ayal wrote:
> Is this GEP replicated sinking intentional/helpful/harmful?
This is intentional and a consequence of VPlan-based sinking. We will keep a uniform GEP in the header and create a new replicate recipe that is sunk to the users in the replicate region. This comes at the cost of not reusing the uniform GEP in the first `pred.store.block`, but this should be cleaned up by different passes.

Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D133762/new/

https://reviews.llvm.org/D133762