NOTE: There is no known need to do this for TEI, howeverNvidia encourages retaining the major CC and any minors in-between when usingnvprune on cuBLAS.

Feel free to close the PR if you prefer to avoid until there's a relevant bug report. My understanding is it should only be an issue when using a kernel from cuBLAS that would defer tosm_70 when it'd have been equivalent forsm_75.

For example in the current base image used to build,sm_70 has 184 cubins vssm_75 containing only 8:

$cuobjdump --list-elf /usr/local/cuda/lib64/libcublas_static.a| grep -oE'\.sm_70.*\.'| wc -l184$cuobjdump --list-elf /usr/local/cuda/lib64/libcublas_static.a| grep -oE'\.sm_75.*\.'| wc -l8#Individual cubins:$cuobjdump --list-elf /usr/local/cuda/lib64/libcublas_static.a| grep -E'\.sm_75.*\.'ELF file    5: libcublas_static.5.sm_75.cubinELF file   13: libcublas_static.13.sm_75.cubinELF file   21: libcublas_static.21.sm_75.cubinELF file   29: libcublas_static.29.sm_75.cubinELF file   37: libcublas_static.37.sm_75.cubinELF file   45: libcublas_static.45.sm_75.cubinELF file   53: libcublas_static.53.sm_75.cubinELF file   61: libcublas_static.61.sm_75.cubin

I'm not entirely sure why the minor CC versions in-between (when present) might matter to be retained.

The concern does not apply to the other two supported real archs handled vianvprune assm_80 is already provided, whilesm_90 does not target anything newer (since it's theonly arch for that CC major):