[llvm] [AMDGPU] Only try DecoderTables for the current subtarget. NFCI. (PR #82992)

[Ivan Kosarev via llvm-commits]

https://github.com/kosarev approved this pull request.

LGTM. We would not need to do the checks manually if the disassembler TableGen backend a) tests predicates as early as possible and b) allows differently predicated / mutually exclusive instructions in the same decoding table. This would also greatly reduce the size of our decoding tables, which currently contain a predicate node at the end of every single decoding path.

https://github.com/llvm/llvm-project/pull/82992