On Tue, Feb 16, 2010 at 11:37 PM, Isaac Dupree <ml@isaac.cedarswampstudios.org> wrote:
If you finish your CellSPU work, great! GHC 6.16 or so might be able to perform well on CellSPU! If not, then the present situation of using the "unregisterized" (slow) C backend will still be available; we don't lose much by removing the few current "registerized" backends. (Actually it's likely to require some build-system fixes in porting to any new platform, even with the unregisterized backend.) (Also, if it turns out to be easier to make a GHC native-code generator backend than an LLVM backend, then maybe that will be yet another theoretical possibility!)
I might be able to get to it, but the odds are low for two reasons. First, IBM has effectively stopped all development on future Cell hardware (if lack of Cell demos at Supercomputing are a good indication, plus no new IBM Cell hardware for two years and Sony looking for a new processor for the PlayStation 4.) Second, functional programming and hybrid multicore is not well understood. Sure, progress is being made on functional programming on GPUs for specific kinds of problems -- but GPUs are fundamentally a data stream model of programming. Cell is a very different hybrid multicore model that isn't entirely data stream, but has some of its characteristics, with a lot of physical limitations.
A better bet would be to target a tiled array processor, such as the Tilera Tile64 (aka MIT RAW).
-scooter