[llvm] [LV] Teach the vectorizer to cost and vectorize llvm.sincos intrinsics (PR #123210)

[Benjamin Maxwell via llvm-commits]

================
@@ -2005,10 +2049,13 @@ class BasicTTIImplBase : public TargetTransformInfoImplCRTPBase<T> {
     InstructionCost ScalarizationCost = InstructionCost::getInvalid();
     if (RetVF.isVector() && !RetVF.isScalable()) {
       ScalarizationCost = 0;
-      if (!RetTy->isVoidTy())
-        ScalarizationCost += getScalarizationOverhead(
-            cast<VectorType>(RetTy),
-            /*Insert*/ true, /*Extract*/ false, CostKind);
+      if (!RetTy->isVoidTy()) {
+        for (Type *VectorTy : getContainedTypes(RetTy)) {
+          ScalarizationCost += getScalarizationOverhead(
+              cast<VectorType>(VectorTy),
+              /*Insert*/ true, /*Extract*/ false, CostKind);
+        }
+      }
----------------
MacDue wrote:

@arsenm this is what causes the changes in the AMDGPU `frexp.ll` costs. Previously, this would be skipped as `RetVF` was 1 for struct-of-vector returns. Now it costs the inserts for both returned vectors. 

https://github.com/llvm/llvm-project/pull/123210