On Wed, Aug 29, 2001 at 11:15:50AM +0200, Arpi wrote:
> Hi,
>
> global:
> - CPU detection
surely
> video:
> - YUV<->RGB conversion stuff (YV12-> any bpp, YV12->YUY2, YUY2->YV12,
> any bpp -> any bpp, especially 15->16, 24->32, and 24/32 bpp -> YUY2,YV12)
> - software scaling routines (YV12,YUY2,RGB)
> - stride copy (copy same bpp image but stride is different)
> - image blitting (like Nick's fastmemcpy)
these are in order of preference for me - at the first stage
we want to create good API
> - iDCT routine (maybe DCT too for compression)
> - Motion comp stuff (it's a lot, but can grabbed from ffmpeg)
these will follow later...
>
> audio:
> - up/downsampling
of course - just not sure what is the best routine here - I've
something in aviplayer already
> - copy with format conversion (8<->16bpp, LE<->BE)
> - maybe the DCT/FFT stuff used by ac3/mp3 codecs
>
> I think it's enough for first step.
> And you don't have to code all, most of these are already done in
> different projects, like sdl, ffmpeg, libmpeg2, libvo, mp3lib...
I know - I've just picked something from libmpeg for testing
> > Well - linux doesn't have something like mediaLib on Sun and Sun will
> > surely not help us :) to port their library to i386 platform - I'd like
> They done. Solaris x86 also have mlib, using MMX stuff!
> But they didn't ported to linux yet... and it isn't opensource.
> But maybe it's worth to ask them first, SUN seems to be linux-friendly.
Well I've been speeking once with some main Java3D developers about two
years ago - basicaly - linux is their concurenting OS - so their support
to linux will be rather limited.
I've sent them already at least two letter - they have never answered.
Another point is - it wouldn't be under GPL/LGPL licence anyway and
we want to have some control over the source - right ? :)
> > Library will have its own separate configure script
> > and should be usable elsewhere just by copying the subtree.
> > But that's just the plan...
>
> I still prefer statically linkable version!
> (for speed and 100% compatibility)
For MMX,SSE routines - you may push all the registers on the stack so
you are not limited by -fPIC and you may use all the registers as you
wish - the only problem would be in C part - but I think almost
everyone has CPU with at least MMX support so I wouldn't see this
as huge problem.
But I would rather prefer to have one such library in the system
instead of that each program will link the library staticaly - if
there will not be measurable speed difference.
It has advantage that as soon as some function would be implemented
by some better code - all the programs will be able to use it
immediately without recompilation.
--
Debian GNU/Linux maintainer - www.debian.{org,cz}
Zdenek Kabelac http://i.am/kabi/ kabi@{i.am, debian.org, fi.muni.cz}
Resistance is futile. You all will be packaged
_______________________________________________
Avifile mailing list
[EMAIL PROTECTED]
http://prak.org/mailman/listinfo/avifile
