b-sumner wrote: > I'm referring to the fact that for a normal compile, if the user passes > `-fsanitize=address`, should that enable it for both the host and device? My > argument is that the user is capable of passing something like `-Xarch_host > -fsanitize=undefined -Xarch_device -fsanitize=address` if they want > fine-grained control, or don't want it on the CPU but want it on the CPU and > vice-versa. This patch adds `-fanitize-gpu` which is a completely separate > option that only works in this single case.
I agree, -fsanitize=address is supposed to enable instrumentation on both host and device...and that is exactly what happens now. If this patch is affecting the existing behavior then something is wrong. https://github.com/llvm/llvm-project/pull/124754 _______________________________________________ cfe-commits mailing list cfe-commits@lists.llvm.org https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits
