[FFmpeg-devel] [PATCH] FFV1 speed tweaks

Jason Garrett-Glaser darkshikari
Sat Aug 9 03:34:01 CEST 2008

> The change to cur_sample and the sign handling seem independant. They should
> be in seperate and benchmarked patches.
> I have no choice but to reject such combined patches (this isnt the first ...),
> and iam affraid this could make you angry which is not what i want at all.
> Its a matter of less than 5 min to split this and running a benchmark 3 times
> which each should barely take a minute with a short file.
> Combining several changes and benchmarking them together leads to chunks of
> code being commited that are slower then the original ...

Oh, I can split it.  I already benchmarked each change as I made it
though; on my system each change independently improved performance.

> Also as the sign part ended up worse on ARM, maybe setting arch/cpu/tune flags
> correctly would help x86 by making it use cmov without changing the code.

Nope, GCC is retarded as usual.  Setting march had no effect.

Dark Shikari

More information about the ffmpeg-devel mailing list