[FFmpeg-soc] swscale ideas

Michael Niedermayer michaelni at gmx.at
Fri Jul 11 00:02:41 CEST 2008


On Fri, Jul 11, 2008 at 12:27:32AM +0900, Luca Barbato wrote:
> Michael Niedermayer wrote:
> >> MMX dither code mangles global vars, shouldn't.
> > 
> > ?
> 
> MANGLE() in it's _best_

?


> 
> >> MMX code has variables in the context not ifdeffed, shouldn't.
> > 
> > ?
> 
> Altivec and bfin variables are cleanly marked and got compiled only on 
> the right arch, mmx are considered part of the standard swscale and got 
> in the way always.

These are not mmx variables, they are SIMD variables which should have been
used by altivec.

To clean this up requires fixing altivec to use the common SIMD variables
[and yes it requires a SIMD_SIZE define so their size matches the altivec vs.
mmx register size]


> 
> > Code _changes_ global vars and breaks multi threaded apps. <--- that should
> > be fixed.
> 
> from what I seen it's due the MANGLE() usage, I guess a nice legacy from 
> mplayer assumptions.

no, you are confused. MANGLE is one of several ways to access variables.
The problem is that they are global, not how they are accessed.


[...]
> > I do expect a little more than moving asm code around for the google
> > payment!
> 
> Right now he had:
> 
> - pushed small changes and polishing here and there for svn (chiaki)
> - moved away the code I disliked and cleaned up swscale.c (nodame)
> - botched committing my changes for ppc and bfin. (nodame)
> 
> I'd like you to calm a bit.

Right now we should have
Finished complete cleanup of swscale in main ffmpeg svn and finished
design and implementation of a clean API for the various asm optims.

And should now be working on getting swscale bitexact, the yuv table
code rewritten and ffmpeg switched to swscale by default.

[and above is not even the primary goals, let alone the secondary]

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

I hate to see young programmers poisoned by the kind of thinking
Ulrich Drepper puts forward since it is simply too narrow -- Roman Shaposhnik
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: Digital signature
URL: <http://lists.mplayerhq.hu/pipermail/ffmpeg-soc/attachments/20080711/809ad752/attachment.pgp>


More information about the FFmpeg-soc mailing list