On Mon, Nov 23, 2015 at 12:54 PM, Nicolas George <geo...@nsup.org> wrote: > Le tridi 3 frimaire, an CCXXIV, Ganesh Ajjanagadde a écrit : >> It seems like getting rid of the sqrt altogether has a very slight >> positive impact (if any at all). I can post the patch, but would like >> to know what to benchmark. There are numerous choices, e.g >> draw_mandelbrot as a whole, the outer loop, or the inner loop. >> I personally think the inner x loop (lines 268-388) is a good place to >> look at, since the difference is very small anyway, and further >> localization is impossible. > > The smallest unit that is usually (or always) called as a whole. Otherwise, > you may find you did shave 10 cycles on a 30 cycles task that is always > called after a 100M cycles task, and find speed ×1.5 instead of negligible. > > In this case, that would be draw_mandelbrot(), and you can do the benchmark > using just ffmpeg.c and the shell's timing utilities. > > (You made the same mistake for deshake.)
I do recall that, but mandelbrot has issues with such benchmarks due to variable number of iterations as I observed while playing with it and as pointed out by Michael. It was hence that I proposed the inner loop benchmark which although it does not get rid of this issue, should reduce the scope of such things. Anyway, I will try my best possibly incorporating the disabling of the skip stuff as suggested by Michael and post. > > Regards, > > -- > Nicolas George _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
