[FFmpeg-devel] Parallelized h264 proof-of-concept
Andreas Öman
andreas
Sun Jun 3 10:13:05 CEST 2007
Hi
Guillaume POIRIER wrote:
> Hi,
>
> What about the speed figures with and without patch?
>
For deblocking type 2 content:
with patch
20235 dezicycles in filter_mb_fast, 60203 runs, 5333 skips
21377 dezicycles in filter_mb_fast, 125710 runs, 5362 skips
20628 dezicycles in filter_mb_fast, 256721 runs, 5423 skips
19867 dezicycles in filter_mb_fast, 518720 runs, 5568 skips
18682 dezicycles in filter_mb_fast, 1042831 runs, 5745 skips
20060 dezicycles in filter_mb_fast, 2090877 runs, 6275 skips
5 runs with 1000 frames, average time for avcodec_decode_video()
(27376 + 27377 + 27338 + 27373 + 27310) / 5 = 27354 ?s
without patch
19410 dezicycles in filter_mb_fast, 60088 runs, 5448 skips
20598 dezicycles in filter_mb_fast, 125593 runs, 5479 skips
19896 dezicycles in filter_mb_fast, 256604 runs, 5540 skips
19133 dezicycles in filter_mb_fast, 518639 runs, 5649 skips
17956 dezicycles in filter_mb_fast, 1042732 runs, 5844 skips
19364 dezicycles in filter_mb_fast, 2090835 runs, 6317 skips
5 runs with 1000 frames, average time for avcodec_decode_video()
(27147 + 27120 + 27197 + 27121 + 27124) / 5 = 27141 ?s
For non-2 type content i don't see any measurable difference
More information about the ffmpeg-devel
mailing list