[llvm] [LV] Pure runtime check for minimum profitable trip count. (PR #115833)

[David Sherwood via llvm-commits]

================
@@ -2502,6 +2502,12 @@ void InnerLoopVectorizer::emitIterationCountCheck(BasicBlock *Bypass) {
 #endif
     // Don't execute the vector loop if (UMax - n) < (VF * UF).
     CheckMinIters = Builder.CreateICmp(ICmpInst::ICMP_ULT, LHS, Step);
+  } else if (MinProfitableTripCount.isNonZero()) {
----------------
david-arm wrote:

Actually I can see in `areRuntimeChecksProfitable` there is something about tail-folding near the end:

```
  // 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::max(MinTC1, MinTC2);
  if (SEL == CM_ScalarEpilogueAllowed)
    MinTC = alignTo(MinTC, IntVF);
  VF.MinProfitableTripCount = ElementCount::getFixed(MinTC);
```

so at least the min trip count isn't being rounded up

https://github.com/llvm/llvm-project/pull/115833