[llvm] [AMDGPU] Speedup canBeVOPD() (PR #200058)

[Stanislav Mekhanoshin via llvm-commits]

rampitec wrote:

> This looks clean. LGTM. I agree it would be interesting to see the perf difference

Actually I think https://github.com/llvm/llvm-project/pull/199072 by @IgWod is better. It produces shorter tables, one entry per opcode. This version produces an entry per permutation. It still would be better to produce a single table for both X and Y but I do not think it is currently possible.

https://github.com/llvm/llvm-project/pull/200058