[<<][staapl][>>][..]
Wed Jul 29 09:26:11 CEST 2009

Extreme Forth

Reminds me of the dataflow vs. `thread oriented' programming idea
mentioned before.  In order to make DSP work for Forth, some kind of
local naming needs to be used to 1) make the random-access work. and
2) introduce concurrency in a serial language.

I'd say there is really only one way to properly program DSP code, and
that is using a dataflow language (DFL).

The point is that using Forth as a _frontend_ syntax isn't such a bad
idea.  The syntax allows _programmers_ to perform extreme factoring.

DFL code is tedious to write due to explicit naming of outputs.  See
the following table:

                named-IN  named-OUT
    - forth         /        /
    - expression    x        /
    - dataflow      x        x

So, if you add some kind of primitive to forth that represent (local)
DFL nodes as names (one could `constant' semantics for inputs and use
the `->' operator as node single assignment)

        A B + -> C

Looking at the C18 described in [1] the local names are actually
_streams_ or more concretely, autoincremented memory access.  The rest
of the article talks about distributing an app over multiple cores.

[1] http://www.ddj.com/hpc-high-performance-computing/210603608



[Reply][About]
[<<][staapl][>>][..]