[PATCH] D136601: [OpenMP] [OMPIRBuilder] Create a new datatype to hold the unique target region info

[Jan Sjödin via Phabricator via llvm-commits]

jsjodin added a comment.

What is the procedure to redo a patch that was reverted?



================
Comment at: clang/lib/CodeGen/CGOpenMPRuntime.cpp:1909
+    CtorEntryInfo.ParentName = Twine(Buffer, "_ctor").toStringRef(Out);
+    llvm::errs() << "Registering var ctor: " << Twine(Buffer, "_ctor") << "\n";
     OffloadEntriesInfoManager.registerTargetRegionEntryInfo(
----------------
mikerice wrote:
> Just noticed this debug code left in here.
Good catch! I am updating the patch again, on top of the reference fix.


================
Comment at: clang/lib/CodeGen/CGOpenMPRuntime.cpp:3031
+        OrderedEntries[E.getOrder()] =
+            std::make_tuple(&E, Loc, EntryInfo.ParentName);
+        ParentFunctions[E.getOrder()] = EntryInfo.ParentName;
----------------
jsjodin wrote:
> mikerice wrote:
> > vitalybuka wrote:
> > > mikerice wrote:
> > > > Needed StringRef(EntryInfo.ParentName) for this to compile for me.
> > > StringRef(EntryInfo.ParentName) is ref to the scope-local EntryInfo info, so you have HWASAN report here
> > Probably meant to change this to const llvm::TargetRegionEntryInfo &EntryInfo to match OffloadTargetRegionEntryInfoActTy?
> Yes, I think there was an inconsistency introduced when I was splitting up my patch. I will work on a fix.
> Probably meant to change this to const llvm::TargetRegionEntryInfo &EntryInfo to match OffloadTargetRegionEntryInfoActTy?




CHANGES SINCE LAST ACTION
  https://reviews.llvm.org/D136601/new/

https://reviews.llvm.org/D136601