Re: [Haskell-cafe] [GSoC] A data parallel physics engine

There's an effort going on to use techniques from Lava (the Haskell- based hardware description language) to target GPUs. Joel Svensson [1] has written his Master's thesis on this and is now working on this for his PhD, so if you ask kindly he might tell you more about this or send you the thesis. [1] .. http://www.chalmers.se/cse/EN/people/svensson-joel On 12 mar 2008, at 22.54, Dan Piponi wrote:

On Wed, Mar 12, 2008 at 2:33 PM, Andrew Coppin wrote:

...

Hanging around here, you really feel like you're at the cutting edge of... something... heh.

Another approach isn't to target a CUDA back end for Haskell but to write an array library that builds computations that can target a CUDA (or other) back end. My first real world job that involved programming was APL [1] based. APL (and its offspring) is a functional-ish programming language that manipulates arrays using a relatively small number of primitives, most of which probably map nicely to CUDA hardware because of the potential for data parallelism. Despite the write-only nature of APL source code, and the negative comments about it by Dijkstra, the expressivity of APL for numerical work is unbelievable. I would love to see some of those ideas somehow brought into Haskell as a library.

[1] http://en.wikipedia.org/wiki/APL_%28programming_language%29 -- Dan _______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe