That’s a cool data point, thanks! Though mind that it is a superoptimizer on LLVM bitcode, not on machine code itself, avoiding all the combinatorial explosion of register allocations and probably a bunch of more, but I don’t know enough about the program.