Yes it can, in which case you embed OpenMP parallelism in low-level library calls, and is another way to accomplish what the article suggests. But it's still a

Yes it can, in which case you embed OpenMP parallelism in low-level library calls, and is another way to accomplish what the article suggests. But it's still a fine-grained parallelism so numpy can match plain Fortran performance in special cases (and not the other way around).