[clang] [CIR] cir.call with scalar return type (PR #135552)

[Erich Keane via cfe-commits]

================
@@ -303,14 +305,23 @@ const CIRGenFunctionInfo &CIRGenTypes::arrangeCIRFunctionInfo() {
   assert(!cir::MissingFeatures::opCallCallConv());
 
   // Construction the function info. We co-allocate the ArgInfos.
-  fi = CIRGenFunctionInfo::create();
+  fi = CIRGenFunctionInfo::create(returnType);
   functionInfos.InsertNode(fi, insertPos);
 
   bool inserted = functionsBeingProcessed.insert(fi).second;
   (void)inserted;
   assert(inserted && "Are functions being processed recursively?");
 
   assert(!cir::MissingFeatures::opCallCallConv());
+  getABIInfo().computeInfo(*fi);
+
+  // Loop over all of the computed argument and return value info. If any of
+  // them are direct or extend without a specified coerce type, specify the
+  // default now.
+  cir::ABIArgInfo &retInfo = fi->getReturnInfo();
+  if (retInfo.canHaveCoerceToType() && retInfo.getCoerceToType() == nullptr)
+    retInfo.setCoerceToType(convertType(fi->getReturnType()));
+
----------------
erichkeane wrote:

I see we return this by reference below... so this is leaked... Or am I missing something?  There is something goofy happening here that I'm misunderstanding perhaps.... @AaronBallman / @efriedma-quic / @rjmccall : Is there something we're not doing right here, that classic-CodeGen does, or does that leak these too?  Or am I missing something?

https://github.com/llvm/llvm-project/pull/135552