[llvm] [VPlan] Introduce ComputeReductionResult VPInstruction opcode. (PR #70253)

[Florian Hahn via llvm-commits]

================
@@ -7640,6 +7463,42 @@ SCEV2ValueTy LoopVectorizationPlanner::executePlan(
 
   BestVPlan.execute(&State);
 
+  DenseMap<const RecurrenceDescriptor *, Value *> ReductionResumeValues;
+  BasicBlock *OrigScalarPH = OrigLoop->getLoopPreheader();
+  // Check if ResResult is a CompueReductionResult instruction, and if it is
+  // return the merge phi node for it.
+  auto GetMergePhiForReduction = [&State,
+                                  OrigScalarPH](VPInstruction *RedResult)
+      -> std::pair<const RecurrenceDescriptor *, PHINode *> {
+    if (!RedResult ||
+        RedResult->getOpcode() != VPInstruction::ComputeReductionResult)
+      return {nullptr, nullptr};
+
+    Value *FinalValue =
+        State.get(RedResult, VPIteration(State.UF - 1,
+                                         VPLane::getLastLaneForVF(State.VF)));
+    auto *PhiR = cast<VPReductionPHIRecipe>(RedResult->getOperand(0));
+    assert(FinalValue->hasOneUse() || PhiR->isInLoop() ||
+           PhiR->getRecurrenceDescriptor().IntermediateStore &&
----------------
fhahn wrote:

Hmm, inLoop is only used in the assert I think, and it is just a single step to get them, not duplicating them has the benefit of ensuring they are always kept in sync

https://github.com/llvm/llvm-project/pull/70253