[clang] [CIR] Implement handling for VectorType with size 3 (PR #161232)

[Andy Kaylor via cfe-commits]

================
@@ -565,13 +580,31 @@ mlir::Value CIRGenFunction::emitLoadOfScalar(Address addr, bool isVolatile,
       return nullptr;
     }
 
-    const auto vecTy = cast<cir::VectorType>(eltTy);
+    // Handles vectors of sizes that are likely to be expanded to a larger size
+    // to optimize performance.
+    auto vecTy = cast<cir::VectorType>(eltTy);
+    cir::VectorType newVecTy =
+        cgm.getTargetCIRGenInfo().getABIInfo().getOptimalVectorMemoryType(
+            vecTy, getLangOpts());
+
+    if (vecTy != newVecTy) {
+      Address cast = addr.withElementType(builder, newVecTy);
+      mlir::Value value = builder.createLoad(cgm.getLoc(loc), cast, isVolatile);
+
+      unsigned oldNumElements = vecTy.getSize();
+      SmallVector<mlir::Attribute, 8> indices;
----------------
andykaylor wrote:

```suggestion
      SmallVector<mlir::Attribute> indices;
```

https://github.com/llvm/llvm-project/pull/161232