ranapratap55 wrote: > First of all, AMDGPU should not have dedicated target feature check in the first place. Second, this is a generic improvement that should go to the generic handling. Updated. https://github.com/llvm/llvm-project/pull/218358