https://github.com/AlexMaclean commented:
Backend changes look reasonable so far. One concern I have with this change is that until now we've assumed specific address-spaces are non-overlapping. You've addressed some of the places where this assumption is encoded but I think there are others you have not. For example, addrspacecasts between shared::cta and shared::cluster seem like they should be valid and should be expanded via a pair of cvta instructions through generic space, I think this would produce an error right now. I also wonder if InferAS or other places have made this assumption as well. https://github.com/llvm/llvm-project/pull/135444 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
