Anastasia added a comment. In https://reviews.llvm.org/D41699#990894, @yaxunl wrote:
> whereas the old definition may fail to detect (not always, only for certain > combinations). For this reason I think it would be good to fix this. I don't think it's good to report errors inconsistently > > > 2. the new definition is less extendable. If new address mode is added, it is > easy for the old definition to maintain binary backward compatibility since > the old definition has room for new address mode. However for the new > definition, it is hard to maintain binary backward compatibility and a clean > definition at the same time. If the new bit is added at higher bits they are > going to be separated from the other bits for address mode. > > Probably we should seek opinions from more OpenCL platform/library > developers since this change could be disruptive for them. Sure, let's hope some over developers can jump in. As for sampler modes I am not sure how much it is going to change in the future OpenCL versions. And as for binary compatibility I think this is a valid point, however it is solvable (by maintaining binary versions). I think we won't be able to do many interesting compiler refactoring tasks if we are to keep binary format unchanged.... It would be very pity to restrict us on that. Perhaps we should think of the solution to that overall long term. https://reviews.llvm.org/D41699 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org http://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
