Hi, 2012/1/19 Jason Garrett-Glaser <[email protected]>: > Is the SSE2 version faster than using 64-bit bswap on x86_64? If not, > it shouldn't be used on 64-bit.
If you mean something like the attached patch (MD5 validated), then it's almost a tie: on a longer sequence to reduce error measures and a 64bits system, I get the equivalent of 62k cycles on average compare to sse2's 61k. In that test, the C version was properly compiled with bswap r??. The fastest run seemed also to the advantage of the SSE2 version. Anyway, here's an updated patch taking into account Loren's comments. Christophe
bswap64.diff
Description: Binary data
0004-dsputil-provide-SIMD-versions-of-bswap_buf.patch
Description: Binary data
_______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
