[FFmpeg-devel] Experimental MSVC port

Mike Melanson mike
Wed Mar 26 23:48:21 CET 2008


Uoti Urpala wrote:
> On Wed, 2008-03-26 at 14:42 -0700, Mike Melanson wrote:
>> Actually, I think I'm seeing some more significant performance gain in
>> icc vs gcc. I have a few data points collected via FATE. But it also
>> depends on how we're measuring performance.
> 
> Can you give some example? Like what codec and how much faster?

Here's a very long test in FATE:

  http://fate.multimedia.cx/index.php?test_spec=32

An H.264 sample from the conformance suite. I'm having a little
difficulty massaging the data for it right now because I'm not a very
good DBA. icc consistently performs better than any of the gcc's on the
same platform but if I compute the average over the test's entire
history, the difference is not that impressive.

The profiling method isn't perfect.

> What would you make a graph of?

Code size over time. Test performance over time (or code revisions) for
individual tests. Number of warnings over time. Performance comparisons
between different revisions. Just look at the data that FATE is
collection and there are quite a few possibilities.

-- 
	-Mike Melanson




More information about the ffmpeg-devel mailing list