Lai-YT wrote: > Why is this a module flag, instead of a function attribute?
The other value handled by the _NVVMReflect_ pass is `"__CUDA_FTZ"`, which Clang currently sets via a module flag. I followed the same approach here for consistency. Also, since the decision to use precise square root functions seems to apply at the module level, using a module flag feels reasonable. @efriedma-quic, do you see any pros or cons to using function attributes instead in this case? :raising_hand: https://github.com/llvm/llvm-project/pull/134244 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
