llvm-project

2e93eb71 - [AMDGPU] Use subtarget feature for flat offset bit width instead of arch checks (#183742)