[Ffmpeg-devel] benchmark of different CABAC routines

[Michael Niedermayer]

Hi

On Tue, Oct 10, 2006 at 02:24:07PM +0200, Guillaume Poirier wrote:
> Hi,
> 
> Michael Niedermayer wrote:
> > On Tue, Oct 10, 2006 at 01:43:36PM +0200, Guillaume POIRIER wrote:
> >>On 10/10/06, Michael Niedermayer <michaelni at gmx.at> wrote:
> >>>asm -march=athlon -mcpu=athlon -mtune=athlon
> >>>4223 dezicycles in decode_residual, 2095208 runs, 1944 skipsbits/s dup=0 
> >>>drop=0
> >>
> >>When you say "branchless", which one are you using specificly? I
> >>assume it's the one on by default on SVN code, isn't it?
> >>Or is it the asm code with cmov in it?
> > 
> > 
> > branchless is the stuff which is used if BRANCHLESS_CABAC_DECODER is defined
> 
> 
> Dam, am I tired and sleepy. I meant to say: when you say "asm":
> There are a lot of asm routines available (one of them being on by
> default)... so is the one you used the default one or the one with cmov?

renorm_cabac_decoder_once used the default (=c code) in all cases and anyway
only the not branchless c decoder should be affected by 
renorm_cabac_decoder_once() optimizations, the others will call
renorm_cabac_decoder_once only in get_cabac_terminate() which isnt
called that often (and not at all in the code between START/STOP_TIMER IIRC)

[...]

-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

In the past you could go to a library and read, borrow or copy any book
Today you'd get arrested for mere telling someone where the library is