[FFmpeg-user] Speed differences, was: The Magic Microsecond

[Michael Bradshaw]

On Mon, Jul 9, 2012 at 8:32 AM, Jan Ehrhardt <phpdev at ehrhardt.nl> wrote:
> Michael Bradshaw in gmane.comp.video.ffmpeg.user (Sun, 8 Jul 2012 17:14:57 -0600):
>>Not that I know of, but if you're willing, you can do a git bisect
>>between two versions (pick the two revisions that are closest to each
>>other but still show the significant speed difference) and you can
>>find out which patch caused the slowdown.
>
> The delay was introduced in between 20120622-git-c17808c and
> 20120627-git-b536e2f. Four lines in the output differ
> substantially:
>
> ffmpeg-20120622-git-c17808c -i %input% -vf scale=384:288 -aspect 4:3
>         -sws_flags bicubic -vcodec wmv2 -b:v 320k -acodec wmav2 -ab 64k
>         -ar 44100 -y output.wmv
>
> [buffer @ 022399e0] w:352 h:288 pixfmt:yuv420p tb:1/90000 fr:25/1
>         sar:12/11 sws_param:flags=2
> [ffmpeg_buffersink @ 02239c60] No opaque field provided
> [Parsed_scale_0 @ 0426d1e0] w:352 h:288 fmt:yuv420p sar:12/11 -> w:384
>         h:288 fmt:yuv420p sar:1/1 flags:0x4
> [aformat @ 02310ca0] auto-inserting filter 'auto-inserted resampler 0'
>         between the filter 'Parsed_anull_0' and the filter 'aformat'
>
> fps: 562
>
> versus
>
> ffmpeg-20120627-git-b536e2f -i %input% -vf scale=384:288 -aspect 4:3
>         -sws_flags bicubic -vcodec wmv2 -b:v 320k -acodec wmav2 -ab 64k
>         -ar 44100 -y output.wmv
>
> [graph 0 input from stream 0:0 @ 03704c60] w:352 h:288 pixfmt:yuv420p
>         tb:1/90000 fr:25/1 sar:12/11 sws_param:flags=2
> [output stream 0:0 @ 0036d840] No opaque field provided
> [Parsed_scale_0 @ 03704b40] w:352 h:288 fmt:yuv420p sar:12/11 -> w:384
>         h:288 fmt:yuv420p sar:1/1 flags:0x4
> [graph 1 input from stream 0:1 @ 03791ec0] tb:1/90000 samplefmt:s16
>         samplerate:48000 chlayout:0x3
>
> fps: 450
>
> There were a lot of patches between 20120622 and 20120627 ...

There's probably a couple hundred commits in those 5 days, which would
require about 8 iterations with git bisect. I'd bisect it myself but
I'm currently on OS X and haven't taken the time to see if the same
problem exists for me on my Windows partition.

I'll try to see if I have the same problem, and if I do I'll git
bisect it (if you haven't already), but I make no promises about when
that will be (may not be for quite a while).

Thanks for the persistent investigation.

--Michael