Re: Changing octave to exploit multi-core hardware

[Jaroslav Hajek]

On Tue, Mar 25, 2008 at 3:45 PM, John W. Eaton <address@hidden> wrote:
> On 25-Mar-2008, Quentin Spencer wrote:
>
>  | What about multithreading the mapper functions? I guess we're relying on
>  | external libraries to actually perform the computations of many of the
>  | functions, but I would assume (correct me if I'm wrong) that octave does
>  | the looping through the individual elements of an array. It would seem
>  | that it would be very straightforward to make the computation of
>  | something like cos([1:1000]) faster by just splitting up portions of
>  | large arrays and sending them to separate processors.
>
>  OK.  In the current sources (not 3.0 or the release-3-0-x branch) the
>  loop for all single-argument mapper functions is
>
>   template <class U, class F>
>   Array<U>
>   map (F fcn) const
>   {
>     octave_idx_type len = length ();
>
>     const T *m = data ();
>
>     Array<U> result (dims ());
>     U *p = result.fortran_vec ();
>
>     for (octave_idx_type i = 0; i < len; i++)
>       {
>         OCTAVE_QUIT;
>
>         p[i] = fcn (m[i]);
>       }
>
>     return result;
>   }
>
>  (this is in liboctave/Array.h).
>
>  What would need to happen to send this computation to multiple
>  processors?  Is it necessary to add special code to enable parallel
>  execution of this loop on systems that can do that?  If so I'm not
>  sure desireable to do that for this loop and every other one like it
>  where we could maybe benefit from embarrassingly simple parallelism.
>  Cluttering Octave's sources with special code for this just seems
>  silly to me.  Shouldn't that sort of detail be handled by the compiler
>  and and OS kernel automatically?
>
>  jwe
>

A single OpenMP directive could do that, or even autoparallelization
with some compilers,
e.g. Intel C++. But there are two obstacles: first, the OCTAVE_QUIT
macro (which breaks independence of cycles) and second, there is no
guarantee that F has no side effects.


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