Re: [Swarm-Modelling] Large scale ABM Simulation on the GPU

[Marcus G. Daniels]

Miles Parker wrote:
> Right, that's OpenCL.
CUDA is a branch with a new GPU back-end  of a branch (PathScale 
compiler) of the Open64 compiler, but just not open.   As far as Apple 
is concerned, OpenCL seems to mean elaborating the clang front end to 
LLVM and/or the gcc-4.2 branch for LLVM.    My guess is that any 
hypothetical new work for new LLVM backends for ATI hardware, like 
NVidia did for their hardware with PathScale/CUDA, will end up not a 
part of the public LLVM tree.   I speculate OpenCL is smoke and mirrors 
to facilitate proprietary extensions that otherwise would have been open 
source in LLVM proper.

Meanwhile, Cell support has been in the (public) GCC and the Linux 
kernel sources for years..