[FFmpeg-cvslog] r8992 - trunk/libavcodec/i386/motion_est_mmx.c
Fri May 11 05:42:19 CEST 2007
On Fri, 11 May 2007, lorenm wrote:
> sse2 version of fullpel sad. 16% faster than mmx2 on core2, but slower on k8.
Oops, this crashes in svq1. How is one supposed to deal with codecs that
encode 16x16 blocks with stride not divisible by 16? I could increase the
alignment in get_buffer, but it seems like wasted memory when svq1 is the
only codec that uses 16x16 for chroma.
More information about the ffmpeg-cvslog