[llvm] [AMDGPU][GlobalISel] Reduce instruction count for store in global addrspace for non-powers-of-2 (PR #206043)

[Krzysztof Drewniak via llvm-commits]

================

----------------
krzysz00 wrote:

1. This looks like an awkward special case - we should also see better handling for, say, {5,6,7} x i8, 5 x i16, 7 x half, etc 

https://github.com/llvm/llvm-project/pull/206043