[erlang-questions] Parallel Shootout & a style question

Kevin Scaldeferri <>
Thu Sep 4 17:28:22 CEST 2008


On Sep 4, 2008, at 8:09 AM, Gleb Peregud wrote:

> On Thu, Sep 4, 2008 at 5:02 PM, Kevin Scaldeferri < 
> > wrote:
>> On Sep 4, 2008, at 3:45 AM, Gleb Peregud wrote:
>>> Automatic "parallelization" of lists:* will not be
>>> efficient, at least without extensive compile-time (or even run- 
>>> time)
>>> analysis of the code (if this is possible/feasible at all).
>>
>> The work on data-parallelism in Haskell seems to suggest this is both
>> possible and feasible.  There, the programmer need only mark a list  
>> for
>> parallelization; the run-time then decides if it should be  
>> parallelized at
>> all, how many thread to use, etc.
>>
>
> I see. That's very nice to hear! It would be awesome if Erlang was
> capable of this :)
>
> Btw, are lists in haskell tagged with their length? If no how the
> division into chunks is done without prior knowledge of length? Is it
> calculated beforehand? Isn't it too slow, since it is O(n)?


Here's a starting point to learn more: http://www.haskell.org/haskellwiki/GHC/Data_Parallel_Haskell

However, skimming over that page a couple of the links, I'm getting  
the impression that I may have overestimated how far along this work  
is based on a couple talks I heard SPJ give.  In particular, if I read  
things correctly, at the moment the user must segment the arrays  
manually; although there's clearly intent to improve on this situation  
(so that the [: :] syntax is actually useful).


-kevin



More information about the erlang-questions mailing list