OSDN Git Service

H.264: split luma dc idct out and implement MMX/SSE2 versions
authorJason Garrett-Glaser <darkshikari@gmail.com>
Fri, 14 Jan 2011 21:34:25 +0000 (21:34 +0000)
committerJason Garrett-Glaser <darkshikari@gmail.com>
Fri, 14 Jan 2011 21:34:25 +0000 (21:34 +0000)
commit19fb234e4af1ff9f58ff2fdd604ac6f6bb87ad6b
tree220be84d79d9c771c1afeab43fdd2aaa82fea01d
parent6c18f1cda2e2b2471ebf75d30d552cb0cb61b6ad
H.264: split luma dc idct out and implement MMX/SSE2 versions
About 2.5x the speed.

NOTE: the way that the asm code handles large qmuls is a bit suboptimal.
If x264-style dequant was used (separate shift and qmul values), it might
be possible to get some extra speed.

Originally committed as revision 26336 to svn://svn.ffmpeg.org/ffmpeg/trunk
12 files changed:
libavcodec/dsputil.h
libavcodec/h264.c
libavcodec/h264.h
libavcodec/h264_cabac.c
libavcodec/h264_cavlc.c
libavcodec/h264dsp.c
libavcodec/h264dsp.h
libavcodec/h264idct.c
libavcodec/svq3.c
libavcodec/x86/dsputil_mmx.c
libavcodec/x86/h264_idct.asm
libavcodec/x86/h264dsp_mmx.c