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

[Alexey Bataev via llvm-commits]

================
@@ -7729,6 +7552,33 @@ SCEV2ValueTy LoopVectorizationPlanner::executePlan(
 
   BestVPlan.execute(&State);
 
+  DenseMap<const RecurrenceDescriptor *, Value *> ReductionResumeValues;
+  BasicBlock *OrigScalarPH = OrigLoop->getLoopPreheader();
+  for (VPRecipeBase &R : *cast<VPBasicBlock>(
+           BestVPlan.getVectorLoopRegion()->getSingleSuccessor())) {
+
+    auto *Red = dyn_cast<VPInstruction>(&R);
+    if (!Red || Red->getOpcode() != VPInstruction::ComputeReductionResult)
+      continue;
+
+    Value *FinalValue = State.get(
+        Red, VPIteration(State.UF - 1, VPLane::getLastLaneForVF(State.VF)));
+    auto *PhiR = cast<VPReductionPHIRecipe>(Red->getOperand(0));
+    assert(FinalValue->hasOneUse() || PhiR->isInLoop() ||
+           PhiR->getRecurrenceDescriptor().IntermediateStore);
+
+    // Find the merge phi, which must be located in the preheader of the
+    // original scalar loop.
+    PHINode *MergePhi = nullptr;
+    for (Value *U : FinalValue->users()) {
+      auto *P = dyn_cast<PHINode>(U);
+      if (P && OrigScalarPH == P->getParent()) {
+        MergePhi = P;
+        break;
+      }
+    }
----------------
alexey-bataev wrote:

Better to outline this as lambda

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