[PATCH] D41539: [CodeGen] Decorate aggregate accesses with TBAA tags

[John McCall via Phabricator via cfe-commits]

rjmccall added a comment.

Yes, I think that's fine.

I think unconditionally attaching the tag is probably wrong; you need suppress it in the may_alias cases.  I'll grant that the existing code doesn't honor that, but that seems like a bug we should go ahead and fix.  You'll need to either propagate more information into this function or propagate out the memcpy instruction so that callers can decide how to apply TBAA.


https://reviews.llvm.org/D41539