On Sat, Aug 15, 2015 at 07:32:13PM +0200, Nicolas George wrote: > L'octidi 28 thermidor, an CCXXIII, Clement Boesch a écrit : > > --- > > Seems a bit tricky to do it in asetnsamples so... Might be relevant to do > > the > > same in aevalsrc as well. > > How does it affect speed? One of the two purposes of sine is to be fast (the > other is to be bit-exact), and I am a bit afraid that the expression > evaluation is slow, even for constants. >
The expression is evaluated once per frame, so very low overhead. But for the record, I tried the following: perf record ./ffmpeg_g -v quiet -f lavfi -i sine=samples_per_frame=16 -t 1000 -f null - (so a very small frame making it trigger the eval very often). Here is the top ten result in perf report: 8.82% ffmpeg_g libc-2.22.so [.] _int_free 7.44% ffmpeg_g libc-2.22.so [.] vfprintf 6.42% ffmpeg_g ffmpeg_g [.] av_rescale_rnd 5.52% ffmpeg_g libc-2.22.so [.] _int_malloc 4.82% ffmpeg_g ffmpeg_g [.] av_frame_move_ref 4.52% ffmpeg_g ffmpeg_g [.] av_frame_unref 4.12% ffmpeg_g libc-2.22.so [.] _mid_memalign 3.42% ffmpeg_g libc-2.22.so [.] _int_memalign 2.97% ffmpeg_g libc-2.22.so [.] __strcmp_sse2_unaligned 2.65% ffmpeg_g libc-2.22.so [.] malloc_consolidate Eval appears way later: 0.10% ffmpeg_g ffmpeg_g [.] av_expr_eval [...] -- Clément B.
pgpfZ_jGsf9xl.pgp
Description: PGP signature
_______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
