[llvm] [TLI] Add glibc 2.35 libmvec vector functions to the x86 table (PR #223817)

[via llvm-commits]

anun333 wrote:

@Andarwinux @tonuonu you're asking for opposite things, so I'll leave the branch as it is until a reviewer picks; the full set is one revert away.

Agreed that non-AVX2 targets are already exposed through the existing `_ZGVd` rows: in my tests the mixed-width shape (a loop whose width comes from a narrower type than the call's) fails at `-march=x86-64` with the existing `pow` row, 56/64 wrong on 22.1.8, while a pure-double loop picks `_ZGVbN2v_sin` and is correct. The question is only whether the 12 new functions join that set now or after #162239.

@RKSimon which would you prefer: the full set now, with #162239 tracking the ISA selection, or the `_ZGVb` rows only until it's fixed?


https://github.com/llvm/llvm-project/pull/223817