Closed jbusecke closed 2 years ago
I just tested this PR locally and got some really encouraging results:
This is a soft scaling experiment on my local machine:
Up to 4 threads the relative execution time remains very close to 1! Afterwards it increases almost linearly, which makes sense since my laptop has 4 cores (there are some low energy cores, but I am unsure if these are recruited for stuff like this).
Ill merge this now and release, so we can hopefully use this for tomorrows session.
I added the
threadsafe
user statement to both signature files. This blocks using the callback function, but since we did not have success with the error messages until now, I hope that this will improve the performance a bit as implied by @rabernat