On Thu, Feb 18, 2016 at 11:12:19AM -0300, James Almer wrote: > On 2/18/2016 10:42 AM, Clément Bœsch wrote: > > On Thu, Feb 18, 2016 at 02:40:06PM +0100, Clément Bœsch wrote: > >> On Thu, Feb 18, 2016 at 01:00:10PM +0000, Carl Eugen Hoyos wrote: > >>> Clément Bœsch <u <at> pkh.me> writes: > >>> > >>>> It's not tested on real hardware (qemu only), so I have > >>>> no idea about performance. > >>> > >>> I can test this on real hardware if you can provide > >>> (exact) command lines. > >>> > >> > >> for ifmt in nv12 nv21 yuv420p yuv422p; do > >> for ofmt in argb rgba abgr bgra; do > >> ./ffmpeg -f lavfi -i testsrc2=d=10 -vf format=$ifmt,format=$ofmt > >> -f null - > >> done > >> done > > > > This will unfortunately involve more converts (testsrc fmt -> ifmt) so it > > won't really give you much hint. You probably want to test with sth like > > av_gettime() before and after the > > ff_##ifmt##_to_##ofmt##_neon_##precision() calls in > > libswscale/aarch64/swscale_unscaled.c > > The START_TIMER and STOP_TIMER() macros are made specifically for this > scenario. >
The function is at frame level, so it's likely not called enough times. -- Clément B.
signature.asc
Description: PGP signature
_______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel