I am porting some OpenMP code to a 4P AMD Opteron box which we bought the PGI compilers for. The code runs fine on our Sun box. When I call omp_set_num_threads( int ) routine, it causes an error. It turns out that if I set the number of threads to anything other than 1, it fails (which is obviously not useful)
omp_get_num_procs returns 4, but omp_get_max_threads returns 1. I tried setting OMP_NUM_THREADS, and I get similar results. Is there something I need to do on the compile line other than -mp? Is there something I should set with the OS? (SuSE 9.x)?