I have a data-parallel code which runs nicely with -fastsse, nicely
on two processors with -Mconcur and WITHOUT -fastsse, and which gives
NaNs when Mconcur and fastsse flags are combined.
Trying though the various switches in fastsse, it turns out to be the
-Mvect=sse that objects to Mconcur.
Is this expected behaviour?
Ecole Centrale de Lyon
It’s definitely not expected. “-Mconcur” and “-Mvect=sse” often parallelize or vectorize the same loops but shouldn’t conflict. Can you please send a report to email@example.com and include the source?