[llvm] [SLP][REVEC] Make tryToReduce and related functions support vector instructions. (PR #102327)

[Alexey Bataev via llvm-commits]

================
@@ -17939,9 +17968,24 @@ class HorizontalReduction {
     case RecurKind::FAdd:
     case RecurKind::FMul: {
       unsigned RdxOpcode = RecurrenceDescriptor::getOpcode(RdxKind);
-      if (!AllConsts)
-        VectorCost =
-            TTI->getArithmeticReductionCost(RdxOpcode, VectorTy, FMF, CostKind);
+      if (!AllConsts) {
+        if (auto *VecTy = dyn_cast<FixedVectorType>(ScalarTy)) {
+          assert(SLPReVec && "FixedVectorType is not expected.");
+          unsigned ScalarTyNumElements = VecTy->getNumElements();
+          for (unsigned I : seq<unsigned>(ReducedVals.size())) {
+            VectorCost += TTI->getShuffleCost(
+                TTI::SK_PermuteSingleSrc, VectorTy,
+                createStrideMask(I, ScalarTyNumElements, ReducedVals.size()));
+            VectorCost += TTI->getArithmeticReductionCost(RdxOpcode, VecTy, FMF,
+                                                          CostKind);
+            VectorCost += TTI->getVectorInstrCost(
+                Instruction::InsertElement, VecTy, TTI::TCK_RecipThroughput, I);
----------------
alexey-bataev wrote:

Better to gather all the required indices and then use getScalarizationOverhead

https://github.com/llvm/llvm-project/pull/102327