On 30/09/15 09:48, Anton Khirnov wrote: > Quoting Henrik Gramner (2015-09-28 17:41:44) >> On Mon, Sep 28, 2015 at 9:49 AM, Anton Khirnov <[email protected]> wrote: >>> But does it actually improve performance measurably? I'd argue that >>> those functions are used in places where it doesn't really matter. >> >> I was using some perf tools through checkasm when I noticed an awful >> lot of time was spent calling av_isdigit() which was kind of silly, >> and inlining it made it run around 5% faster overall. But yes, it's >> obviously not a performance critical piece of code by any means. I >> haven't really looked at other code that uses any of those functions >> though. >> >>> And since inline public functions tend to generate pain, it's better to >>> avoid them unless there are large practical gains otherwise. >> >> av_toupper() and av_tolower() are similar short functions in the same >> file that are currently inlined though, so one could argue that this >> patch improves consistency if nothing else. > > Well, I was planning to send a patch to uninline them :) > > Anyone else got an opinion on this? >
I'm not against having the inline, they did not change and hopefully they won't change later. lu _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
