[llvm] [SLP]Fix perfect diamond match with extractelements in scalars (PR #132466)

[Alexey Bataev via llvm-commits]

================
@@ -5339,8 +5338,19 @@ static InstructionCost getScalarizationOverhead(const TargetTransformInfo &TTI,
     }
     return Cost;
   }
-  return TTI.getScalarizationOverhead(Ty, DemandedElts, Insert, Extract,
-                                      CostKind, VL);
+  APInt NewDemandedElts = DemandedElts;
+  InstructionCost Cost = 0;
+  if (!ForPoisonSrc && Insert) {
+    // Handle insert into non-poison vector.
----------------
alexey-bataev wrote:

Let's use a conservative solution for now. Currently, codegen just generates insertelement for each unique scalar here, so I just replicated this for non-poison input value.
Need to teach ScalarizationOverhead of the non-poison input vector. Also, not sure the current generic implementation is fully correct. It currently supposes that all insertvector instructions are inserted into poison vector, but it is true only for the first instruction.

https://github.com/llvm/llvm-project/pull/132466