On Sep 24, 2007, at 4:09 PM, Brad Lewis wrote:

> I'd like to see the data flow programming model as part of OpenMP  
> and supported directly by C and Fortran compilers.   Since it's  
> not, I think CORD provides a good framework for doing data flow  
> programming without making a radical change to some obscure  
> language or machine.
>

Let me rephrase. Data parallelism is a major re-whacking of the  
programming model. As shown by Verilog, you can make the language  
look sequential ... and graft on totally alien semantics ... but is  
it a good idea?

If you don't totally have parallel semantics, you aren't going to get  
all of the benefit. If you do, it's not *really* Fortran or C  
(Verilog for those who have never looked at it, looks like C, but  
being a Hardware Description Language is inherently parallel ...  
except where special syntax causes things to be sequential ... it's a  
rate outliter).



Keith H. Bierman    keith.bierman at Sun.COM   |  khbkhb at gmail.com
Strategic Engagement Team                   | AIM: kbiermank
<speaking for myself, not Sun*> Copyright 2007





Reply via email to