[llvm] [LV][EVL] Emit vp.merge intrinsic to enable out-loop reduction in EVL vectorization. (PR #101641)

[Mel Chen via llvm-commits]

================
@@ -9386,10 +9397,8 @@ void LoopVectorizationPlanner::adjustRecipesForReductions(
                cast<VPInstruction>(&U)->getOpcode() ==
                    VPInstruction::ComputeReductionResult;
       });
-      if (PreferPredicatedReductionSelect ||
-          TTI.preferPredicatedReductionSelect(
-              PhiR->getRecurrenceDescriptor().getOpcode(), PhiTy,
-              TargetTransformInfo::ReductionFlags()))
+      if (CM.usePredicatedReductionSelect(
+              PhiR->getRecurrenceDescriptor().getOpcode(), PhiTy))
----------------
Mel-Chen wrote:

It can be adjusted in the EVL transformation. patch 8a3982fa6c65c08271b43194698d27e4cb29d0f4
But I don't recommend this. Such an implementation is more complicated and may lose the opportunity to sink the non-predicated reduction select outside the loop in the future VPlan transformation.
Could you point out why you want to adjust it in the EVL transformation?

https://github.com/llvm/llvm-project/pull/101641