Not just close to native speed, this will get you the exact same C code for the inner loop.