[PATCH] D147114: [LV] Use BFI to adjust cost of predicated instructions

[Florian Hahn via Phabricator via llvm-commits]

fhahn added a comment.

Do we have a test with profile info? Might be worth having  dedicated tests for this feature with a range of branch probabilities if possible



================
Comment at: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp:4538
     // Scale the cost by the probability of executing the predicated blocks.
     // This assumes the predicated block for each vector lane is equally
     // likely.
----------------
this needs updating now I think


================
Comment at: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp:5596
+    InstructionCost &Cost, const BasicBlock *BB) const {
+  assert(Cost.isValid() && "Can't scale invalid cost");
+
----------------
dmgreen wrote:
> Could invalid cost handling be pushed into this function. It would help not needing it at all the call sites.
Yeah I think that would help readability


================
Comment at: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp:5602
+  auto HeaderFreq = BFI->getBlockFreq(TheLoop->getHeader()).getFrequency();
+  if (HeaderFreq == 0)
+    return Cost;
----------------
Could you explain the reasoning here? Maybe fall back to the original code using `getReciprocalPredBlockProb` here?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D147114/new/

https://reviews.llvm.org/D147114