[ADMB Users] Parallel processing with ADMB

[JohnS]

Chris,
I'm not familiar with CUDA, but I can tell you that writing parallel
code for automatic differentiation on a shared memory machine is very
challenging. One of the immediate goals of the ADMB Project is to
completely open the source code  in mid 2009. If you can wait a few
months you would be very welcome to try to compile the source code for
this target. Parallel processing is definitely on of the improvements
envisaged for future enhancements to ADMB. You might consider
participating in this process.

You should be aware the the current ADMB source code contains well
over 100,000 lines of C++ code, so that writing and testing something
equivalent from scratch would take some time.
Regards,
John Sibert

On Jan 26, 9:05 am, developers-boun... at admb-project.org wrote:
> [ Attached Message ]From:"Grandin, Chris" <Chris.Gran... at dfo-mpo.gc.ca>To:<develop... at admb-project.org>Date:Thu, 22 Jan 2009 10:22:43 -0800Local:Thurs, Jan 22 2009 8:22 amSubject:ADMB port to Cuda
>
> Hi,
>
> I have deigned and am in the process of writing a Management Strategy
> Evaluator using a massively parallel personal supercomputer with 960
> processing cores. I need to modify and recompile the ADMB source for
> these processors, for which the code (written in CUDA-C) must be
> compiled using the nvcc compiler. I understand the reluctance to release
> source code but I must ask before I embark on the long journey of
> rewriting my own version from scratch. My project will bring ADMB into
> the realm of parallel computing and I would of course make this port
> (library) available to all without cost.
>
> Thanks in advance
>
> Chris Grandin
> Hake assessment
> Fisheries and Oceans Canada
> Pacific Biological Station
> C-356, 3190 Hammond Bay Road
> Nanaimo, BC, Canada, V9T 6N7
> chris.gran... at dfo-mpo.gc.ca
>
> _______________________________________________
> Users mailing list
> Us... at admb-project.orghttp://lists.admb-project.org/mailman/listinfo/users