On Thu, Oct 29, 2015 at 07:49:44PM -0400, Ganesh Ajjanagadde wrote: [...] > although you did explain to me where optimizations could be useful at > an algorithmic level privately, I take this as an opportunity to > request for general comments on places that people think could benefit > from performance at a C/algorithmic level (NOT asm/simd stuff for me > personally): yet another ping for Clement.
Mmh... without SIMD let me think... You may look at the TODO in libavfilter/f_ebur128.c. There is a PRE-filter and a RLB-filter to merge into one to reduce computation time. This is an already solved mathematical problem. You may want to have a look, it's most likely a good way to speed up that filter algorithmically. Then maybe you can look into adding threading support in some codecs or filters. No other idea that do not involve SIMD right now. [...] -- Clément B.
Description: PGP signature
_______________________________________________ ffmpeg-devel mailing list firstname.lastname@example.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel