[FFmpeg-devel] [PATCH] avfilter/vf_deshake: replace qsort with AV_QSORT

[Ganesh Ajjanagadde]

On Sat, Oct 31, 2015 at 1:27 PM, Ganesh Ajjanagadde <gajjanag at mit.edu> wrote:
> On Sat, Oct 31, 2015 at 12:43 PM, Nicolas George <george at nsup.org> wrote:
>> Le decadi 10 brumaire, an CCXXIV, Ganesh Ajjanagadde a écrit :
>>> Entirely possible. I just did this as a low hanging fruit. I highly
>>> suspect that the algorithm can be improved (see Michael's comment).
>>> The start/stop was just around the qsort statement. Since you asked
>>
>> So the speedup observed may be completely irrelevant.

For the benefit of the thread, here is a benchmark for the find_motion
function to address this point:
new:
2075495840 decicycles in find_motion,       1 runs,      0 skips
1981653450 decicycles in find_motion,       2 runs,      0 skips
1963960877 decicycles in find_motion,       4 runs,      0 skips
1995843222 decicycles in find_motion,       8 runs,      0 skips
1954257600 decicycles in find_motion,      16 runs,      0 skips
1914343743 decicycles in find_motion,      32 runs,      0 skips
1900114312 decicycles in find_motion,      64 runs,      0 skips
1883816712 decicycles in find_motion,     128 runs,      0 skips
1847216223 decicycles in find_motion,     256 runs,      0 skips
1807254028 decicycles in find_motion,     512 runs,      0 skips
1815531917 decicycles in find_motion,    1024 runs,      0 skips

old:
2137526850 decicycles in find_motion,       1 runs,      0 skips
2032845285 decicycles in find_motion,       2 runs,      0 skips
2049874912 decicycles in find_motion,       4 runs,      0 skips
2047886076 decicycles in find_motion,       8 runs,      0 skips
1989838803 decicycles in find_motion,      16 runs,      0 skips
1929553950 decicycles in find_motion,      32 runs,      0 skips
1929938301 decicycles in find_motion,      64 runs,      0 skips
1914471658 decicycles in find_motion,     128 runs,      0 skips
1975107739 decicycles in find_motion,     256 runs,      0 skips
1879646457 decicycles in find_motion,     512 runs,      0 skips
1900282483 decicycles in find_motion,    1024 runs,      0 skips

I leave it to the judgement of others whether or not this is significant enough.

>>
>>> this, I will add it to the commit message.
>>
>> I would have appreciated if you waited for the discussion around the patch
>> to finish before pushing it.
>
> That was an inexcusable misread on my end: I viewed (incorrectly and
> hastily) your message as a general remark. Sorry.
>
>>
>> Regards,
>>
>> --
>>   Nicolas George
>>
>> _______________________________________________
>> ffmpeg-devel mailing list
>> ffmpeg-devel at ffmpeg.org
>> http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>>