Kyle wrote:
> Doesn't the application have to be somehow aware that it is working in a 
 > cell environment, similar to SMP?

 From everything I've read, the "cell environment" is a SIMD 
environment. (That's "Single Instruction Multiple Data", which is the 
kind of thing the MMX extensions do, only on a large scale.)  I don't 
imagine it'll be any time soon before a language like C or C++ (or 
Fortran or Ada or ...) will be able to easily take advantage of this 
without special programming. Something like SQL or APL or Prolog or 
something where you don't really have a high-resolution "program 
counter" concept built into the language would work better; that is, 
something where each instruction in the language could take several 
compute seconds could be rearranged more easily by the compiler to take 
advantage of parallelism.  I imagine bits of POV-Ray could be 
special-cased to take advantage of the Cell stuff, like matrix 
multiplies or something, but I don't think you'll get any sort of large 
speed-up just by recompiling the same C code that runs efficiently on a 
SISD machine.  Just my opinion.

-- 
   Darren New / San Diego, CA, USA (PST)
     The samba was clearly inspired
     by the margarita.

Post a reply to this message