https://gcc.gnu.org/bugzilla/show_bug.cgi?id=122612
--- Comment #2 from Sam James <sjames at gcc dot gnu.org> --- (In reply to Thomas Schwinge from comment #1) > Yeah. > > This will need resolving via nvptx-tools 'as', see > <https://github.com/SourceryTools/nvptx-tools/issues/30> "[RFC] Handle sm_* > which is no longer supported by CUDA / ptxas exec check or configure check?". > > As a work-around, you should be able to change 'target_arg = "sm_50";' near > the end in nvptx-tools 'nvptx-as.cc' to whatever the lowest 'sm_N' is that > your 'ptxas' supports. (Or, build GCC/nvptx without CUDA/'ptxas' available, > or 'configure' GCC/nvptx with a '--with-arch=[...]' that implies a 'sm_N' > that your 'ptxas' supports; see > 'gcc/config/nvptx/nvptx.cc:first_ptx_version_supporting_sm'.) Thank you! It sounds like I should be able to do this dynamically with a sed + map of CUDA versions for now.
