[Haskell-cafe] GSoC and Data Parallel Haskell

Dmitry Dzhus dima at dzhus.org
Thu Mar 22 02:14:18 CET 2012


22.03.2012, 03:24, "Geoffrey Mainland" <mainland at apeiron.net>:
> Taking advantage of GHC's SIMD support from within DPH is in the works.
>
> I'd love to have some good target applications to help drive the work.
> An application with DPH, plain C, C+SSE intrinsics, and, say, OpenCL
> versions would be even better, but that's probably wishful thinking.
> Dmitry, is your thesis about gas dynamics (I suspect it is) and you just
> want to try using Haskell, or is it about using Haskell to implement
> scientific code?

Yes, thesis is about gas dynamics and Haskell is used to implement the
well-established DSMC method wih certain application. I plan a chapter
on how Haskell (and, hopefully, DPH, since current implementation is a
bit naive and too slow, I'm heavily hacking on it) showed itself for
this task. At some point I even planned to write the practical part of
thesis with Literate Haskell, but coursework paper I wrote in last
semester (which formed a preliminary basis for the thesis) with LH
confused people.

Barnes-Hut algorithm is more «appropriate» to demonstrate NDP (and
looks like it's been used for that since early papers on Nesl) due to
its recursive nature, leading to nested subdivision of space; many
other simulation methods usually start with more regular and flat
space division (such as in DSMC, or in Lattice Boltzmann method
(that's a continuum dynamics approximation as opposed to dilute gases
in DSMC, papers on CUDA-powered LB sprout here and there today)), but
I think nesting can be seen here too for example in gradual grid
refinement at areas of simulation domain which require extra interest.



More information about the Haskell-Cafe mailing list