On 21 December 2010 18:45, Stephan Aßmus <superstippi@xxxxxx> wrote: > Am 21.12.2010 08:25, schrieb dlmcpaul@xxxxxxxxx: >> >> Author: dlmcpaul >> Date: 2010-12-21 08:25:40 +0100 (Tue, 21 Dec 2010) >> New Revision: 39913 >> Changeset: http://dev.haiku-os.org/changeset/39913 >> >> Added: >> haiku/trunk/src/add-ons/media/plugins/ffmpeg/yuvrgb_sse.nasm >> haiku/trunk/src/add-ons/media/plugins/ffmpeg/yuvrgb_sse2.nasm >> haiku/trunk/src/add-ons/media/plugins/ffmpeg/yuvrgb_ssse3.nasm >> Removed: >> haiku/trunk/src/add-ons/media/plugins/ffmpeg/yuvrgb.nasm >> Modified: >> haiku/trunk/src/add-ons/media/plugins/ffmpeg/Jamfile >> haiku/trunk/src/add-ons/media/plugins/ffmpeg/gfx_conv_mmx.cpp >> haiku/trunk/src/add-ons/media/plugins/ffmpeg/gfx_conv_mmx.h >> haiku/trunk/src/add-ons/media/plugins/ffmpeg/gfx_util.cpp >> Log: >> correct yuv422 planar conversion. Seperate sse, sse2 and ssse3 asm code. >> Add packed convertor > > Cool stuff, thanks for looking into it! One remark from the ticket is still > not implemented, though, which is that the SIMD versions still perform > alignment checks and fallback back to the plain C implementation. If those > checks are moved higher up, we could fall back to the swscale code path, > which should be a lot faster than our C version. Sure. -- Cheers David