On 2014-02-05 15:00:57 +0100, Christophe Gisquet wrote: > Hi, > > 2014-02-05 Martin Storsjö <mar...@martin.st>: > > Based on a patch by Ronald S. Bultje. > > --- > > Updated as suggested by Janne, to not require pushing anything to > > the stack. > > That's not really an issue worthy of additional work, but with such > patches, it may be interesting to know the impact of the fix > speed-wise.
the penalty of the fix is not huge but it's meaningless. It doesn't make sense to benchmark incorrect code. I can i"solve" every problem in a small and constant amount of time incorrectly. > For instance, this second patch looks like the right thing to do. But > besides that, is there an actual/objective reason, e.g. number of > cycles? (I guess they can vary significantly depending on > architecture, cache/memory speed etc, but still...). Number of cpu cycles/overall run time, and while those indeed vary for different CPU the relative difference between two patches on the same CPU differs not that much across systems. Janne _______________________________________________ libav-devel mailing list libav-devel@libav.org https://lists.libav.org/mailman/listinfo/libav-devel