2010/4/15 "Miguel A. Vázquez" <mavazq...@cttc.es>:
> Hello all,
>
> I am trying to implement some code in OpenMP under Octave, but I have
> some problems.
>
> #include <omp.h>
> #include <octave/oct.h>
>
> DEFUN_DLD (pmean, args, , "Calculates then mean in parallel")
> {
> int nargin = args.length ();
>
> Matrix m (1,1);
>
>
> int tid,nthreads;
> omp_set_num_threads(2);
> #pragma omp parallel private( nthreads, tid)
> tid = omp_get_thread_num();
> nthreads = omp_get_num_threads();
> octave_stdout << "Thread "<< tid << " of " << nthreads << "
> threads \n.";
> if (nargin != 1)
> print_usage ();
> else
> {
> NDArray A = args(0).array_value ();
>
>
>
> #pragma omp parallel for reduction(+:m)
> for (octave_idx_type i = 0; i < A.nelem (); i++){
> m = m + A.elem (i)/A.nelem ();
> }
>
> return octave_value (m);
>
> }
>
> }
>
> This code is supposed to calculate a mean of an Octave matrix in
> parallel: some iterations of the for will be done by different threads.
>
> I set the number of threads to 2 (omp_set_num_threads(2)). I compiled
> and linked it nicely but, unfortunatelly, when I executed it:
>
> octave:1> pmean([10 0 5])
> Thread 0 of 1 threads
> .ans = 5
>
> One thread is executed instead. I don't really know how to solve this
> problem, even don't know where I have to look for. Does anyone of you
> have treated this problem before?
>
> Thank you in advance,
>
> Miguel
>
I think you lack some basic OpenMP knowledge. In C++, omp parallel
applies only to the next statement or block, so the way you've written
it it doesn't make much sense. Use omp critical for I/O to make it
atomic.
You didn't initialize m; besides, you must use a primitive type for
reduction clause. Also be aware that each DEFUN must return a value or
throw an exception, just like every C++ function.
This version works for me:
DEFUN_DLD (pmean, args, , "Calculates then mean in parallel")
{
int nargin = args.length ();
octave_value retval;
omp_set_num_threads(2);
#pragma omp parallel
{
int tid,nthreads;
tid = omp_get_thread_num();
nthreads = omp_get_num_threads();
// WARNING: I/O not thread safe
#pragma omp critical
octave_stdout << "Thread "<< tid << " of " << nthreads << " threads. \n";
}
if (nargin != 1)
print_usage ();
else
{
const NDArray A = args(0).array_value ();
double m = 0;
#pragma omp parallel for reduction(+:m)
for (octave_idx_type i = 0; i < A.nelem (); i++)
m += A.elem (i) / A.nelem ();
retval = m;
}
return retval;
}
hth
--
RNDr. Jaroslav Hajek, PhD
computing expert & GNU Octave developer
Aeronautical Research and Test Institute (VZLU)
Prague, Czech Republic
url: www.highegg.matfyz.cz
------------------------------------------------------------------------------
Download Intel® Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
_______________________________________________
Octave-dev mailing list
Octave-dev@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/octave-dev
