[PATCH] D24557: [Loop Vectorizer] Simplified GEP cloning. NFC.

Matthew Simpson via llvm-commits llvm-commits at lists.llvm.org
Wed Sep 14 09:05:28 PDT 2016


mssimpso added a subscriber: mssimpso.
mssimpso added a comment.

Hi Elena,

This looks good to me. Thanks! At first I was a bit surprised by this simplification, but it makes sense. If we have a consecutive GEP for a memory instruction that's not scalarized, we only need to compute it's value at VF=0. For the unroll iterations, we compute the offset from this GEP (at UF=0) later on.


================
Comment at: ../lib/Transforms/Vectorize/LoopVectorize.cpp:2794
@@ -2823,2 +2793,3 @@
       Ptr = Builder.Insert(Gep2);
+
     } else { // No GEP
----------------
Did you mean to add a blank line here?


Repository:
  rL LLVM

https://reviews.llvm.org/D24557





More information about the llvm-commits mailing list