[llvm] [AMDGPU][MC][True16] Support V_RCP/SQRT/RSQ/LOG/EXP_F16. (PR #81131)

[Joe Nash via llvm-commits]

https://github.com/Sisyph commented:

Needs a rebase.
Should there be vop3 from vop1 dpp asm tests in this patch?
Should there be vop3 from vop1 non-dpp asm and disasm tests in this patch?

This is more of a suggestion for a future patch, but what is the status of the gfx12 versions of these vop1 _t16 instructions? There are no asm or disasm tests for true16 gfx12 instructions. It looks to me like the vop1 version of these _t16 instructions should work correctly in gfx12 (because the encoding for true16 and fake16 is the same). The VOP3 from VOP1 I would not expect to be encoded correctly looking at how the classes are currently defined. I only ask because if we can add gfx12 tests for the _t16 instructions soon it might help prevent regressions.

https://github.com/llvm/llvm-project/pull/81131