https://github.com/MrSidims commented: The approach is LGTM, thanks! Probably a test is missing for either shader with vector length > 4 or kernel with length > 16 https://github.com/llvm/llvm-project/pull/169665