Hi Xuelei, This version looks much better and has addressed most issues, so is OK. It's still beaten by __memset_emag for cases > 128 bytes, so you might want to look into improving the loop further (including making sure it is aligned to a fetch boundary). Cheers, Wilco