[PATCH] D100570: [LoopVectorize] Prevent multiple Phis being generated with in-order reductions

[David Sherwood via Phabricator via llvm-commits]

david-arm added inline comments.


================
Comment at: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp:4292
   for (unsigned Part = 0; Part < UF; ++Part) {
+    if (IsOrdered && Part > 0)
+      break;
----------------
Hi @kmclaughlin, it feels a bit inconsistent that we're not checking for VF.isVector() here, but we do below? Maybe the VF.isVector() check can be folded into the IsOrdered definition above? Or perhaps we can avoid all the extra VF.isVector() checks completely if we move the check into useOrderedReductions?


================
Comment at: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp:4723
 
+    bool IsOrdered = Cost->isInLoopReduction(cast<PHINode>(PN)) &&
+                     useOrderedReductions(*RdxDesc);
----------------
To be consistent with the other changes above should we add a `&& VF.isVector()` check here too?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D100570/new/

https://reviews.llvm.org/D100570