[llvm] [LV] Add initial support for vectorizing literal struct return values (PR #109833)

[David Sherwood via llvm-commits]

================
@@ -6081,13 +6085,17 @@ InstructionCost LoopVectorizationCostModel::getScalarizationOverhead(
     return 0;
 
   InstructionCost Cost = 0;
-  Type *RetTy = toVectorTy(I->getType(), VF);
+  Type *RetTy = toVectorizedTy(I->getType(), VF);
   if (!RetTy->isVoidTy() &&
-      (!isa<LoadInst>(I) || !TTI.supportsEfficientVectorElementLoadStore()))
-    Cost += TTI.getScalarizationOverhead(
-        cast<VectorType>(RetTy), APInt::getAllOnes(VF.getKnownMinValue()),
-        /*Insert*/ true,
-        /*Extract*/ false, CostKind);
+      (!isa<LoadInst>(I) || !TTI.supportsEfficientVectorElementLoadStore())) {
+
+    for (Type *VectorTy : getContainedTypes(RetTy)) {
----------------
david-arm wrote:

Again, would be good to have cost model tests that defend this code path. Looks like it's called from `setVectorizedCallDecision`. There are some examples of how we check the costs in `Transforms/LoopVectorize/AArch64/intrinsiccost.ll`.

https://github.com/llvm/llvm-project/pull/109833