[Ffmpeg-devel] benchmark of different CABAC routines
Wed Oct 11 21:48:20 CEST 2006
On Wed, 2006-10-11 at 21:07 +0200, Michael Niedermayer wrote:
> well, if you repeated them a few times then what are these numbers?
> average? minimum? furthermore did you repeat one test several times
Minimum. I tested some modifications and repeated the tests, the minimum
times across a few tests looked reasonably consistent (some higher
spikes in individual tests but overall the variation was less than the
differences between cases).
> and then the next or where they interleaved? its not a silly question
> as the skip functionality removes the effects of task switches, without
In this case the times with skip functionality weren't completely
consistent either and differed between runs. I think it might be because
some calls naturally have a significantly higher time than the average
so far (so they're incorrectly skipped as task switches etc), and if the
average so far is high then they're not as easily skipped and thus raise
the average more. I'm not too sure about the reason, there could be
other explanations. Testing a constant skip limit rather than one that
depends on the current average might clarify that.
> and standard deviations or all the individual benchmark scores would be
> nice too, its just that ive no experience with the modified benchmark
> code you use so iam uncertain about its accuracy
I think giving a single number for standard deviation would not be that
useful, but some kind of statistics about the consistency of the results
would be good. Maybe if I do more tests I'll write an automated script
to collect the results...
More information about the ffmpeg-devel