e-kud wrote: Checked on the set of benchmarks that I usually use for measuring the coverage of GlobalISel. No regression was found at least significant to influence more than 0.01% (out of ~4mln functions) https://github.com/llvm/llvm-project/pull/197374