[llvm] [AMDGPU] Enable vectorization of i8 values. (PR #134934)

[Jeffrey Byrnes via llvm-commits]

https://github.com/jrbyrnes commented:

If we are going to have a custom cost model for memory instructions we should probably add a test to test/Analysis/CostModel/AMDGPU

https://github.com/llvm/llvm-project/pull/134934