Re: [Haskell-cafe] my gsoc project topic

[Max Bolingbroke]

On 29 March 2010 14:01, Csaba Hruska  wrote:
> Hi!
> - Another LLVM related idea is to create a framework to support writing llvm
> passes in haskell, this should be based on existing llvm haskell binding.

This is a cool idea!

For bonus points, you could provide a binding to the new version of
the Hoopl library that Simon PJ and co are working on. This is a
Haskell library that makes it astonishingly easy to write
dataflow-based optimisations. It is designed for use with C-- in GHC,
but is actually agnostic about the type of graphs it optimises, so you
should be able to plug it into LLVM's data types quite easily!

For double bonus points, you could then use the LLVM Hoopl backend to
write a LLVM pass that fixes up any problems we observe in the output
of the LLVM code generator for programs originating from GHC :-)

You can get an old version of the paper:
http://www.cs.tufts.edu/~nr/pubs/dfopt-abstract.html at not - but note
that a Brand New and much improved version is being worked on for ICFP
- hopefully the code will be public soon.

Cheers,
Max
___
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe

Re: [Haskell-cafe] my gsoc project topic

[Edward Kmett]

On Mon, Mar 29, 2010 at 9:01 AM, Csaba Hruska wrote:

> I'm wondering about the topic of my gsoc project. [..]
>
 I'm interested in these areas:
> - improving cabal (more sophisticated dependency, multi version, compiler
> support, etc) [..]
>

I like the idea of improving cabal, and one can definitely see the community
benefit of work on such a pervasively used tool.


> - Another idea is to improve GHC's new LLVM backend. [..]
>

I like this idea, but like you said, a number of people seem to be
interested in working on the LLVM backend this summer, so it may be a
crowded space and hard to coordinate. But anything that could expose more
information about the shape of thunks and slot usage to LLVM could help a
great deal with alias analysis, which is usually in LLVM the biggest barrier
to reordering and rewriting.


> - Another LLVM related idea is to create a framework to support writing
> llvm passes in haskell, this should be based on existing llvm haskell
> binding. [..]
>

I think this would be an amazing value-add for the LLVM backend, and might
open the door to a whole new class of Haskell optimizations.


> - I'm also interested in computer graphics. While this kind of project
> could look really good, the community could not benefit from it. [..]
>

While I personally enjoy working on 3d graphics in general, it is hard to
sell the community benefit.

 -Edward Kmett
___
Haskell-Cafe mailing list
Haskell-Cafe@haskell.org
http://www.haskell.org/mailman/listinfo/haskell-cafe