https://gcc.gnu.org/bugzilla/show_bug.cgi?id=104283
Bug ID: 104283 Summary: nvptx-none needs more user friendly architecture handling Product: gcc Version: 12.0 Status: UNCONFIRMED Severity: normal Priority: P3 Component: target Assignee: unassigned at gcc dot gnu.org Reporter: xw111luoye at gmail dot com Target Milestone: --- x86_64-pc-linux-gnu-accel-nvptx-none-gcc needs both -misa and -mptx to select a specific target. When I use OpenMP offload. -foffload-options=nvptx-none="-misa=sm_80" I got error ``` ptxas /tmp/ccW7wYQT.o, line 3; error : PTX .version 3.1 does not support .target sm_80 ptxas /tmp/ccW7wYQT.o, line 2622; warning : Instruction 'shfl' without '.sync' may produce unpredictable results on sm_70 and later architectures ``` Then I have to get it through with -foffload-options=nvptx-none="-misa=sm_80 -mptx=7.0" I think GPU users know what sm_80 is but they rarely know PTX. Can we -mptx default to highest -misa selection supports? Or introduce something like -march=sm_80 and map it to appropriate -misa and -mptx. Simplify the need of options is much appreciated.