David's PBX designs use a blackfin. OSLEC runs on them real-time. I 
don't know what optimization it required.

There are lots of us who would do the integer port were it not for worry 
about breaking code we don't understand. If you have time enough to get 
it wrong and start again, go for it.

     Thanks

     Bruce

On 10/10/2011 02:53 PM, Steve Strobel wrote:
> I am interested in running codec2 on a Blackfin (BF537) under uClinux.
>   The BF537 is a 16/32-bit processor running at 500 MHz but does not
> have a FPU, so my guess is that like the 400 MHz FriendlyArm discussed
> here a month or so ago, it probably won't handle encoding or decoding
> even a single stream of codec2 in real time without some extra work,
> and I would really like to use at least two simultaneous streams.
>
> An easy starting point would appear to be enabling the Blackfin gcc
> ”-mfast-fp” option for "fast floating point" that uses relaxed rules
> for NaN and is "several times faster than [the floating point routines
> in]...libgcc" [1] and [2]. I don't know whether that will just make it
> comparable to the speed of the FriendlyArm or whether it might just be
> fast enough; trying it is probably the only way to find out.
>
> If that isn't enough, I would consider porting an integer FFT as
> suggested by Bruce Robertson a month or two ago [3].  Would converting
> Codec2 to use an integer FFT require in-depth knowledge of the codec,
> or would it just be a matter of properly scaling the inputs and
> outputs of the FFT routine?  There is a 16-bit FFT in the Blackfin DSP
> library [4];  is that enough resolution or would codec2 need 32 bit
> results?  If so, it might be possible to use the techniques described
> in EE-186, Extended-Precision Fixed-Point Arithmetic on the Blackfin
> Processor Platform [5] to make a fast Blackfin-specific 32-bit FFT.
> Or a portable integer FFT in C might be fast enough and would benefit
> other no-FPU platforms.
>
> Any suggestions would be very welcome.  Or news of a fixed-point port
> in progress :) [6]
>
> Steve
>
>
> [1] - 
> http://docs.blackfin.uclinux.org/doku.php?id=optimization#gcc_optimization_for_blackfin
> [2] - 
> http://www.analog.com/static/imported-files/application_notes/EE.185.Rev.4.08.07.pdf
> [3] - "You will find the non-FPU operation of this code much more
> efficient if you can port an integer fft. I believe we are still using
> the kissfft library, and it was a integer version of this that I
> experimented with making use of."  August 28 on this mailing list.
> [4] - 
> http://docs.blackfin.uclinux.org/doku.php?id=toolchain:libbfdsp#transformational_functions
> [5] - I couldn't find a copy on analog.com, but did at
> http://www.datasheetarchive.com/EE-186-datasheet.html
> [6] -<http://www.rowetel.com/blog/?page_id=452>  shows a "fixed point
> port" as milestone #3
>
>
> --
> Steve Strobel
> Link Communications, Inc.
> 1035 Cerise Rd
> Billings, MT 59101-7378
> (406) 245-5002 ext 102
> (406) 245-4889 (fax)
> WWW: http://www.link-comm.com
> MailTo:[email protected]
>
> ------------------------------------------------------------------------------
> All the data continuously generated in your IT infrastructure contains a
> definitive record of customers, application performance, security
> threats, fraudulent activity and more. Splunk takes this data and makes
> sense of it. Business sense. IT sense. Common sense.
> http://p.sf.net/sfu/splunk-d2d-oct
> _______________________________________________
> Freetel-codec2 mailing list
> [email protected]
> https://lists.sourceforge.net/lists/listinfo/freetel-codec2


------------------------------------------------------------------------------
All the data continuously generated in your IT infrastructure contains a
definitive record of customers, application performance, security
threats, fraudulent activity and more. Splunk takes this data and makes
sense of it. Business sense. IT sense. Common sense.
http://p.sf.net/sfu/splunk-d2d-oct
_______________________________________________
Freetel-codec2 mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freetel-codec2

Reply via email to