Hi,
when thinking about bug #1668 I've have wondered if it was possible to
implement a keyword to prevent parallelising some numpy functions. For
example I've encountered it a couple of times that I was running a loop
with an openmp directive but the loop contains functioncall to another
function that uses e.g. np.sum (or similar). The way I understand it if I
compile this with openmp support pythran will also try to parallise the
np.sum, is that correct? If that is the case, would it be possible to
implement a "keyword/directive" that prevents this? something like e.g.
#openmp none or #no parallel or similar? I have it quite often that I run
of these functions in a tight-loop and parallelising actually causes a
slow-down.
Cheers
Jochen
