lkml.org 
[lkml]   [1999]   [Jul]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: porting linux to DSP
Ingo Oeser wrote:

> On Tue, 27 Jul 1999, Steve Underwood wrote:
>
> > Real time OSes do exist for some DSPs (e.g. SPOX), but they are _very_ much
> > specialist DSP OSes, and they ain't too pretty.
>
> But using a ("normal") workstation (even low end class) as a host
> for a DSP might be a solution. You can run Linux there to do all
> the OS stuff and just do "number-crunching" stuff and may be some
> little scheduling or ISRs at the DSP.
>
> This seems a reasonable solution to me, and there is some work in
> progress to allow linux users to take advantage of these external
> processing services a DSP offers.

Of course. The fact the DSPs do nothing but polynomials well means they are
usually hosted by another processor in some way. In most applications an OS has
no place in a DSP. In some applications, where a number of activities are
interleaved on a single DSP (e.g. most telephony cards process several voice
channels per DSP, and perform several functions concurrently on each channel)
some elementary form of task management may be required. That is where products
like SPOX can be useful.

There is a strong motivation amongst DSP designers to get some microcontroller
functionality into DSP chips. Most DSP applications are embedded (read low
cost), and increasingly low power (read cell-phones). If you can eliminate a $1
microcontroller from a design you might get $3 off the product's selling price,
reduce its size, and possibly extend its battery life a little. Every attempt I
have seen so far has been a failure. The opposite has also been tried - putting
a MAC into a microcontroller. This has also been a failure. A low cost DSP can
still outperform a high end Alpha for most true DSP algorithms. To date, DSPs
are DSPs, microcontrollers are microcontrollers, and the two concepts just don't
mix. I can't imagine any form of Linux that would have any meaningful place on a
current generation DSP. That said, I'm sure someone will eventually get
something like this working out of sheer perversity.

As far as adjunct DSPs are concerned, they are just too expensive for most
things. The DSPs themselves don't cost too much, but the small volume boards on
which they currently reside do. DSPs will never find their way onto the
motherboard - the industry has proved that conclusively. Unless someone can
produce a killer DSP board application that sells in large numbers and brings
down the price, the adjunct DSP is pretty much a non-issue.

Steve



-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:53    [W:0.081 / U:0.260 seconds]
©2003-2014 Jasper Spaans. Advertise on this site