[PATCH] D115713: [LV] Don't apply "TinyTripCountVectorThreshold" for loops with compile time known TC.

[Paul Walker via Phabricator via llvm-commits]

paulwalker-arm added inline comments.


================
Comment at: llvm/lib/Transforms/Vectorize/LoopVectorize.cpp:9870-9875
+      // If trip count is known compile time constant there is no any extra
+      // overhead from vectorization with scalar epilogue. Let cost model to
+      // decide on profitability in that case.
+      auto TC = SE.getSmallConstantTripCount(L);
+      if (!TC) {
+        if (*ExpectedTC <= TTI->getMinTripCountTailFoldingThreshold()) {
----------------
I don't quite understand this change.  The whole point of `getMinTripCountTailFoldingThreshold()` was to give targets control over this behaviour based on their understanding how the cost model has been implemented.

Admittedly this was in part due to the immaturity of the cost modelling but this change essentially removes that flexibility to the point where there's no value in keeping `getMinTripCountTailFoldingThreshold()`?

If your previous patches improve the cost model in this regard then I'd rather `getMinTripCountTailFoldingThreshold()` be removed.  That said, @dtemirbulatov can you help here in ascertaining if this option is still required based on this new patch series?


Repository:
  rG LLVM Github Monorepo

CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D115713/new/

https://reviews.llvm.org/D115713