vp9/x86: add ff_vp9_loop_filter_[vh]_16_16_ssse3().
authorClément Bœsch <u@pkh.me>
Sun, 5 Jan 2014 20:00:40 +0000 (21:00 +0100)
committerClément Bœsch <u@pkh.me>
Sun, 12 Jan 2014 19:20:24 +0000 (20:20 +0100)
commitaf68bd1c06ff9ea7379a95d39d5fde5fa1938cd3
tree239dff8f40b247f6f7b8eca8f3ae7cacd74415ed
parente11ceea68ff365e59e0fa8dea0c28182dc6f4287
vp9/x86: add ff_vp9_loop_filter_[vh]_16_16_ssse3().

16662 decicycles in loop_filter_h_16_16_c, 8387355 runs, 1253 skips
17510 decicycles in loop_filter_v_16_16_c, 8387516 runs, 1092 skips

4941 decicycles in ff_vp9_loop_filter_h_16_16_ssse3, 8387887 runs, 721 skips
3899 decicycles in ff_vp9_loop_filter_v_16_16_ssse3, 8387980 runs, 628 skips

Overall decode time goes from:
  ./ffmpeg -v 0 -nostats -threads 1 -i ~/samples/vp9/ped1080p.webm -f null -  8.10s user 0.02s system 99% cpu 8.126 total
to:
  ./ffmpeg -v 0 -nostats -threads 1 -i ~/samples/vp9/ped1080p.webm -f null -  6.15s user 0.04s system 99% cpu 6.199 total

(46 to 61 fps)
libavcodec/x86/Makefile
libavcodec/x86/vp9dsp_init.c
libavcodec/x86/vp9lpf.asm [new file with mode: 0644]