[FFmpeg-devel] [RFC] Runtime-inited versus Hardcoded tables.

Rich Felker dalias
Thu Jan 31 17:31:17 CET 2008


On Thu, Jan 31, 2008 at 05:09:48PM +0100, Michael Niedermayer wrote:
> On Thu, Jan 31, 2008 at 05:57:14PM +0200, Uoti Urpala wrote:
> > 
> > On Thu, 2008-01-31 at 11:01 -0500, Rich Felker wrote:
> > > On Thu, Jan 31, 2008 at 03:34:15PM +0200, Uoti Urpala wrote:
> > > > without much speed loss (yes the cmp function can be optimized
> > > 
> > > Once again I'm glad Uoti has no authority in this project...
> > 
> > Why? Do you have reason to believe that what I said would not be
> > technically correct? What kind of speed loss do you claim to be
> > inevitable if you drop motion_est.o size to half with a reasonable
> > amount of work? I wouldn't be surprised if it got noticeably faster, the
> > current implementation wastes quite a lot of cache.
> 
> A patch which makes it smaller at the same or better speed
> or one which makes it faster at the same size, is welcome.

Indeed, I agree totally. What's not welcome is the attitude that a few
percent speed loss does not matter. If Uoti really claims he can
improve speed due to better cache utilization, a patch (with
benchmarks) would be welcome!

Rich




More information about the ffmpeg-devel mailing list