On 2009-01-13 10:27:10 +0100, Robert Fraser <fraseroftheni...@gmail.com> said:
Andrei Alexandrescu wrote:
Robert Fraser wrote:
Andrei Alexandrescu wrote:
Dunno. According to SPJ, automatically parallelizing map was a failed
experiment in Haskell. Explicit parallelizing a la pmap seems to be the
way to go.
Source? I think as processors grow in number, automatic paralellization
will become increasingly important, so I'm wondering why it was a
"failed experiment"?
Private conversation.
Andrei
Did he mention what the implementation was like and/or what the problem
was (i.e. too much overhead, etc.)?
Probably something of the following:
1) In general when you do map you cannot exclude that the user expects
sequentiality, or at least not parallelism.
2) if you use forward iterators (and lazy lists are such) are *very*
sequential.
By the way now that you are going more in the lazy direction you might
want to consider again the problem with pure function not accepting
lazy structures (invariant structures have to be fully evaluated).
I had brought this up quite some time ago...
Fawzi
