|automatic exploitation of task parallelism from sequential email@example.com (Dave Lloyd) (1995-11-23)|
|Re: automatic exploitation of task parallelism from sequential firstname.lastname@example.org (1995-12-01)|
|Organization:||Dept of Computing and Electrical Engineering, Heriot-Watt University|
|Date:||Fri, 1 Dec 1995 12:17:44 GMT|
Dave Lloyd <email@example.com>:
>> Is it true that most researchers in parallel computing, consider
>> automatic exploitation of parallelism from sequential programs, like
>> C, too complicated ?
Imperative languages like C have side effects which makes elucidating
parallelism somewhat tricky. Its much easier for pure functional languages
which lack assignment: program components can't interact through
changes to shared variables so you can find potential parallelism
everywhere. Alas, most of that that parallelism is too fine grain for
efficient exploitation. However, there are nice correspondences between
higher order functions and algorithmic skeletons which provide hints as
to sites of exploitable parallelism. For example, see Murray Cole's book.
You might also look at David Busvine's and Tore Bratvold's work on
parallelising compilers to turn sequential SML programs into occam2
using a skeleton based approach. You can find details on:
Of course, algorithmic skeletons can be used for parallelisng imperative
languages as for example in Susan Pelagatti's work on the Pisa Parallel
Return to the
Search the comp.compilers archives again.