[llvm] [AMDGPU] Implement LSR cost model for GFX9+ (PR #184138)

[via llvm-commits]

michaelselehov wrote:

I made some performance testing with rocRAND, llama and CK. rocRAND shows 93 improvements above 5% with zero regressions above 5%. Most noticeable improvements come from threefry kernels up to 32%.
llama and CK - no visible changes (GEMM kernels are different and aren't affected by this change).

@arsenm @rampitec @jayfoad please take a look.

https://github.com/llvm/llvm-project/pull/184138