Re: Death of DSP support?

[munck@mindspring.com (Robert Munck)]

On Tue, 26 Aug 1997 07:33:01 -0700, Steven O'Neill
<oneills@top.monad.net> wrote:

>...It appears that, in buying Tartan, TI has seriously curtailed any
>possibility of Ada use in the DSP domain.

I know very little about general DSP programming, but did
a lot of work for NRL in underwater audio (i.e. Sonar) 
signal processing.  We showed pretty conclusively that the
"natural" programming language for that restricted domain
is data flow diagrams.  All of the work done in textual
languages that we saw was a disaster.

This seemed to be true because the great majority of
computation done in a DSP was performed in highly-
specialized SIMD array processors.  These were very
tightly microcoded (horizontal microcode) to do a
small number of SP primitives -- Walsh filters, FFT,
bandwidth, etc.  The control programs, written in Ada
and running in 68K micros, did little more than 
schedule the movement of data between the SIMD machines
and large buffer memories. Rather than hardcode this
logic, we developed a system to interpret encoded DFWs.
Coding was done on a MAC using a CAD/flowcharting tool.

AT&T implemented this basic architecture in a wonderful
processor called the ECOS EMSP.  However, IBM went
completely bananas attacking it (at the Congressional
lobbyist level) after we reproduced in 6 coder-months
a system for the AN/UYS-1 that they had spent 200
coder-years developing.

I don't know if this approach would work as well in
the general case of DSP.

Bob Munck
Mill Creek Systems LC