[llvm] [AMDGPU] Eliminate unnecessary packing in wider f16 vectors for sdwa/opsel-able instruction (PR #137137)

[Pierre van Houtryve via llvm-commits]

https://github.com/Pierre-vh requested changes to this pull request.

Precommit test is merged into this patch so it kind of defeats the purpose of precommiting the tests 
I'd also like to see MIR tests that test the specific patterns you're looking for, along with some edge cases that can break the code to test for infinite loops, crashes and so on.

https://github.com/llvm/llvm-project/pull/137137