[Speex-dev] Development on DSP 6711

[Jean-Marc Valin]

Hi Collin,

Sorry, I forgot that the C67xx is float. In this case, I think the
floating point would probably be faster than the fixed-point version (as
it is on x86). As for handwritten assembly, I think the main functions
that would benefit from that are in filters.c, vq.c and ltp.c (total
around 5-7 functions).

	Jean-Marc

P.S. My first name is Jean-Marc

On jeu, 2004-09-16 at 10:33 -0300, Colin MacKenzie wrote:
> Jean: The 6711 is actually a floating point processor so it can do the
> floating point without emulation. I am working on the 6416 which does
> not have it. Working with fixed point may still speed things up
> possibly.
> 
> Muhammad: It seems like you might have to start getting into linear
> assembly to speed things up. If you are serious to getting speex
> working on TI DSP we should work together on the routines. 
> 
> C
> 
> Jean-Marc Valin wrote:
> > Hi Muhammad,
> > 
> > Did you define FIXED_POINT when compiling the code? Otherwise, you'll be
> > using floating point emulation which is usually very slow.
> > 
> > 	Jean-Marc
> > 
> > On jeu, 2004-09-09 at 10:39 +0500, Muhammad Saad Saleem wrote:
> >   
> > > Dear developers,
> > >          I am a research associate in Al-Khwarzimi Institute of Computer
> > >          Science. I am working on TI 6711 Evaluation Board. I want to
> > >          implement audio compression algorithm on this board. I have
> > >          already implemented sub-band compression on this board. But the
> > >          resultant quality is really poor. Therefore, I want to use
> > >          speex open source library for compression. The main problem
> > >          which I am facing is that when I compress a 160 samples block,
> > >          it takes almost 40 million cycles on contrary to my sub-band
> > >          algorithm which only takes 3 million cycles. So my input buffer
> > >          overflow as the data is slowly compressed. If you have
> > >          experienced any such problem, then please let me know that what
> > >          did you do to avoid it. I will be grateful to you.
> > > 
> > > -----
> > > Best Regards
> > > Muhammad Saad Saleem
> > > Research Associate
> > > University of Engineering and Technology, Lahore
> > > http://www.uet.edu.pk/saad
> > >     
> > 
> >   
> _______________________________________________
> Speex-dev mailing list
> Speex-dev at xiph.org
> http://lists.xiph.org/mailman/listinfo/speex-dev