[FFmpeg-devel] [PATCH][VAAPI][2/6] Add common data structures and helpers (take 3)

[Diego Biurrun]

On Mon, Mar 09, 2009 at 01:28:49PM +0100, Reimar D?ffinger wrote:
> On Mon, Mar 09, 2009 at 12:33:35PM +0100, Diego Biurrun wrote:
> > On Mon, Mar 09, 2009 at 05:24:14AM +0100, Michael Niedermayer wrote:
> > > On Mon, Mar 09, 2009 at 01:48:33AM +0100, Diego Biurrun wrote:
> > > > On Sun, Mar 08, 2009 at 05:51:43PM +0100, Michael Niedermayer wrote:
> > > > > 
> > > > > glibc and "best" in the same paragraph makes me want to puke
> > > > > 
> > > > > anyway, actual numbers: (done 3x to show that they are stable)
> > > > > 
> > > > > k7 : cpu clocks=170059006 = 98361us  (1016.663fps)  1525.0MB/s
> > > > > mmx: cpu clocks=293085663 = 169516us  (589.915fps)  884.9MB/s
> > > > > sse: cpu clocks=170377116 = 98544us  (1014.775fps)  1522.2MB/s
> > > > > c: cpu clocks=195054405 = 112817us  (886.391fps)  1329.6MB/s
> > > > > k7 : cpu clocks=170553318 = 98645us  (1013.736fps)  1520.6MB/s
> > > > > mmx: cpu clocks=291090475 = 168362us  (593.958fps)  890.9MB/s
> > > > > sse: cpu clocks=173589832 = 100402us  (995.996fps)  1494.0MB/s
> > > > > c: cpu clocks=193683165 = 112023us  (892.674fps)  1339.0MB/s
> > > > > k7 : cpu clocks=171431091 = 99154us  (1008.532fps)  1512.8MB/s
> > > > > mmx: cpu clocks=298011922 = 172365us  (580.164fps)  870.2MB/s
> > > > > sse: cpu clocks=173702776 = 100468us  (995.342fps)  1493.0MB/s
> > > > > c: cpu clocks=194631164 = 112571us  (888.328fps)  1332.5MB/s
> > > > > 
> > > > > (core 2 based cpu in case you wonder)
> > > > > 
> > > > > dirty patch needed to make above run: 
> > > > > (patch/commit to fix it cleanly in mplayer svn is welcome especially
> > > > >  by thouse who decided to #include config.h all over the place)
> > > > 
> > > > I'd rather call it broken than dirty :)
> > > > 
> > > > The build should be fixed now and work without such kludges.
> > > 
> > > maybe it should but it does not.
> > > it just generates 8 binaries that each call libc memcpy() and result
> > > thus in rather boring and misleading information.
> > 
> > I'll be more precise: The build system part is fixed now.  The tool is
> > still broken as is noted at the top, but that was Reimar's fault ;-p
> 
> Hm? I think my "fault" was only to make fast_memcpy explicit, so you'd
> have to replace memcpy with fast_memcpy, but that is not enough to
> actually compile different versions.
> So far my best idea is just to put the HAVE_... defines in config.h
> under #ifndef so they can be overridden from the commandline.

I committed something that works, even though it duplicates some lines
of code.  I'll see if I improve it later on, but for such a small and
one-off tool, it should not be a big deal.

Diego