On Thu, Aug 15, 2013 at 03:35:32PM -0700, Jason Garrett-Glaser wrote: > On Thu, Aug 15, 2013 at 3:29 PM, Diego Biurrun <[email protected]> wrote: > > > Older versions of clang choke if that function is forcibly inlined. > > Furthermore, inlining the function gives no performance benefit at > > least with gcc 4.4 and 4.6. > > --- > > > > I ran benchmarks on my trusty old K6-III. > > Have you considered benchmarking on a modern CPU, perhaps using > START_TIMER, or at least a higher-bitrate video for which CABAC is a larger > portion of decoding time?
I don't have access to another x86_32 machine right now. I could run some benchmarks on a reasonably current x86_64 next week. Suggestions for test samples and more benchmarks welcome. > I recall that when this was originally implemented, it was benchmarked and > confirmed to be faster; I don't think a benchmark on a 15-year old CPU is > exactly a reliable test here. But for which gcc version was this confirmed to be faster? Note that I did not consider my set of benchmarks exhaustive, just a sensible start. Diego _______________________________________________ libav-devel mailing list [email protected] https://lists.libav.org/mailman/listinfo/libav-devel
