AlexVlx wrote: Just because somebody wrote something we should not assume it is necessarily sound or valid. Is there some illustration of said valid uses / have the users made an argument around why they absolutely must have the macro?
CUDA doesn't have a macro like this at all AFAIK, and everything was fine dynamically querying `warpSize` on the host and relying on non-`constexpr` `warpSize` on the device side. Not only does it not impair the single-source-ness of CUDA, but I'd say it actually enhances it. They are solving the same problem, can we identify why their solution is inadequate and we should not migrate people towards that, removing an arbitrary point of divergence between HIP & CUDA? https://github.com/llvm/llvm-project/pull/109663 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
