>
> A very interesting paper about accelerating agent-based models with graphics
> cards (GPU). I placed the paper at
> http://www.friam.org/DSouza_GPUacceleratedABM.pdf

Pretty neat.    The fact is that if we want to take advantage of these 
kind of architectures in a general purpose way, we need compiler 
technology appropriate for them.   Toolkits may still work if they can 
physically live within the constraints of such hardware, but Java VMs 
certainly cannot (e.g. a 256KB local store on a Cell SPU).    Better yet 
would be to have an ABM language directly integrated with the compiler 
and have, e.g.  agent distribution between different memory areas and 
processors directly managed.  The compiler is in the best position to 
manage hardware.  Certainly better than most modelers.   One problem 
with software layering (ABM toolkits) is that compilers can't see 
through all that to prove those operations that are safe to do in parallel.

LLVM (http://www.llvm.org/) would be one foundation, and a lot of work 
has been done on GCC to make new front ends easier.  

Marcus

============================================================
FRIAM Applied Complexity Group listserv
Meets Fridays 9a-11:30 at cafe at St. John's College
lectures, archives, unsubscribe, maps at http://www.friam.org

Reply via email to