OPENMP for loop with reduction

[arocketguy]

I am trying to get an openmp for loop with reduction to compile and run. Based on the documentation and examples I would have thought something like:

#pragma omp parallel for reduction(+:y) ADOLC_OPENMP
for ( i = 0 ; i < n ; i++ ) {

 y += x[i]*x[i];

 y+= cos(x[i]);

}

would work, but the g++ compiler complains:
error: user defined reduction not found for ‘y’

so it's not figuring out how to reduce y...

playing around with things:

#pragma omp ADOLC_OPENMP parallel for reduction(+:y)
for ( i = 0 ; i < n ; i++ ) {

 y += x[i]*x[i];

 y+= cos(x[i]);

}

does compile and run (correctly it seems), but is only using a single processor.

If I just ignore the reduction (yes, it will be wrong) and go back to the original syntax:

#pragma omp parallel for ADOLC_OPENMP
for ( i = 0 ; i < n ; i++ ) {

 y += x[i]*x[i];

 y+= cos(x[i]);

}

will compile (no error: user defined reduction not found for ‘y’ ), and it does use all the threads given it...but of course it gives the incorrect answer.

Is ADOL-C forcing the reduction down to 1 thread? And am I employing the syntax correctly?