On 03/19/12 10:33, H.J. Lu wrote: > Here is a patch to use __builtin_bswap[32|64] in i386/x86_64 bits/byteswap.h > for GCC >= 4.2. I didn't use __builtin_bswap32 for __bswap_16 since it is > less efficient: For size, yes. For speed, is the partial register stall on rorw larger than a single extra shift insn? r~