Re: paralell cellfun 'parcellfun'

[Jaroslav Hajek]

On Wed, Mar 18, 2009 at 12:42 PM, WMennerich <address@hidden> wrote:
>
>
> Jaroslav Hajek-2 wrote:
>>
>> If a simple SMP parallelism would suffice your needs, I have recently
>> contributed the parcellfun (a parallel cellfun) function into
>> OctaveForge/main/general.
>> It is simple to use and quite efficient on systems implementing
>> efficient process forking, such as GNU/Linux.
>>
>
> Thats exactly that what I was also searching for. Already the cellfun /
> arrayfun constructs can make your sources more readable (but often they are
> not faster than a for-loop).
> Does this function work with the 'standart-Octave' or are some
> multicore-subsystems needed?

It should work in a normal Octave installation. I have only tested it
with the development version; but it's likely it will work with 3.0.x
as well. Note that it uses process forking, which is likely to be
inefficient on Windows/Cygwin (windows provides no native support for
the operation). On GNU/Linux, it scales great.

> Does also a 'pararrayfun' exist?

No, not yet. This carries some problems with optimally distributing
the work. But eventually, I intend to implement pararrayfun using
parcellfun.

> Maybe this function(s) should be core function(s) of octave since then you
> can use multiple cores in nearly each multithread situation (E.g.
> simulations with different parameter sets).
>

I'm not that sure, given its limitations. But if people will want that
to happen, I have no problem with it.

-- 
RNDr. Jaroslav Hajek
computing expert & GNU Octave developer
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz