jurahul wrote: Planning to do some profiling with the new llvm-mc --runs option to see which of the above options has better decode times. I am assuming we want to err on the side of faster decoder over smaller code size. https://github.com/llvm/llvm-project/pull/146593