[llvm] [SLP][REVEC] Fix scalar mask is passed to getScalarizationOverhead but the type is vector. (PR #128476)

[Alexey Bataev via llvm-commits]

================
@@ -13963,15 +13963,31 @@ InstructionCost BoUpSLP::getGatherCost(ArrayRef<Value *> VL, bool ForPoisonSrc,
     ShuffledElements.setBit(I);
     ShuffleMask[I] = Res.first->second;
   }
-  if (!DemandedElements.isZero())
-    Cost +=
-        TTI->getScalarizationOverhead(VecTy, DemandedElements, /*Insert=*/true,
-                                      /*Extract=*/false, CostKind, VL);
+  if (!DemandedElements.isZero()) {
+    if (isa<FixedVectorType>(ScalarTy)) {
+      assert(SLPReVec && "Only supported by REVEC.");
+      // We don't need to insert elements one by one. Instead, we can insert the
+      // entire vector into the destination.
+      Cost = 0;
+      unsigned ScalarTyNumElements = getNumElements(ScalarTy);
+      for (unsigned I : seq<unsigned>(VL.size()))
+        if (DemandedElements[I])
+          Cost += TTI->getShuffleCost(
+              TTI::SK_InsertSubvector, VecTy, std::nullopt, CostKind,
+              I * ScalarTyNumElements, cast<FixedVectorType>(ScalarTy));
----------------
alexey-bataev wrote:

Use `::getShuffleCost'

https://github.com/llvm/llvm-project/pull/128476