[clang] [CIR] Add support for copy elision (PR #157713)

[Andy Kaylor via cfe-commits]

================
@@ -44,16 +46,37 @@ CIRGenFunction::emitAutoVarAlloca(const VarDecl &d,
   if (ty->isVariablyModifiedType())
     cgm.errorNYI(d.getSourceRange(), "emitAutoVarDecl: variably modified type");
 
+  assert(!cir::MissingFeatures::openMP());
+
   Address address = Address::invalid();
   if (!ty->isConstantSizeType())
     cgm.errorNYI(d.getSourceRange(), "emitAutoVarDecl: non-constant size type");
 
   // A normal fixed sized variable becomes an alloca in the entry block,
-  mlir::Type allocaTy = convertTypeForMem(ty);
-  // Create the temp alloca and declare variable using it.
-  address = createTempAlloca(allocaTy, alignment, loc, d.getName(),
-                             /*arraySize=*/nullptr, /*alloca=*/nullptr, ip);
-  declare(address.getPointer(), &d, ty, getLoc(d.getSourceRange()), alignment);
+  // unless:
+  // - it's an NRVO variable.
+  // - we are compiling OpenMP and it's an OpenMP local variable.
+  if (nrvo) {
+    // The named return value optimization: allocate this variable in the
+    // return slot, so that we can elide the copy when returning this
+    // variable (C++0x [class.copy]p34).
+    address = returnValue;
+
+    if (const auto *rd = ty->getAsRecordDecl()) {
----------------
andykaylor wrote:

I made the change anyway to avoid any potential confusion between `RecordDecl` and `CXXRecordDecl`.

https://github.com/llvm/llvm-project/pull/157713