if (GOMP_has_masked_thread_num(thread_num)) block
GOMP_has_masked_thread_num encapsulates the following check, so that OMPT
can distinguish the compiler-generated code from user-level thread-number
queries:
omp_get_thread_num() == thread_num
Hereby, thread_num has the value of the argument to the filter
clause or zero if not specified.
Alternately, we generate two copies of the parallel subfunction and only include this in the version run by the thread_num thread. Surely this is not worthwhile though...