[llvm] [LoopVectorize] Enhance Vectorization decisions for predicate tail-folded loops with low trip counts (PR #69588)

[David Sherwood via llvm-commits]

================
@@ -9816,11 +9817,13 @@ static bool areRuntimeChecksProfitable(GeneratedRTChecks &Checks,
   //   RtC < ScalarC * TC * (1 / X)  ==>  RtC * X / ScalarC < TC
   double MinTC2 = RtC * 10 / ScalarC;
 
-  // Now pick the larger minimum. If it is not a multiple of VF, choose the
-  // next closest multiple of VF. This should partly compensate for ignoring
-  // the epilogue cost.
+  // Now pick the larger minimum. If it is not a multiple of VF and a scalar
+  // epilogue is allowed, choose the next closest multiple of VF. This should
+  // partly compensate for ignoring the epilogue cost.
   uint64_t MinTC = std::ceil(std::max(MinTC1, MinTC2));
-  VF.MinProfitableTripCount = ElementCount::getFixed(alignTo(MinTC, IntVF));
+  if (SEL == CM_ScalarEpilogueAllowed)
----------------
david-arm wrote:

Can you just put an assert here for the CM_ScalarEpilogueAllowed case and see which tests fail? Then you can just make a copy of the test and adapt it to use PGO data for the trip count information and compile it with tail-folding enabled?


https://github.com/llvm/llvm-project/pull/69588