-
-
Notifications
You must be signed in to change notification settings - Fork 23
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Regression in mumps 5.7.3=*_3 build causing hang? #125
Comments
It anyone is experiencing something similar, a possible workaround is set |
The failure is a bit hard to replicate as it happens only for the tests of the python bindings, not for the similar tests for C++ functionality. |
some packages (e.g. openblas) have OMP as a deprioritized feature. Maybe we should do that? From my performance testing for #124, OMP has substantial value for highly parallel single-process solves (2x speedup with 8 threads), but a small cost for 1 or 2-threaded solves, and the results for mumps-seq x N omp threads are always slower than mumps-mpi x N processes. I kind of assumed OMP would be a simple win, but it doesn't seem to be. |
What are the dimensions/solving time of the problem we are testing? We use ipopt for problems that are solved in ~50/100 ms, and for those the openmp thing is tipically just overhead, hence we tipically set OMP_NUM_THREADS=1 for actual use even when everything is working. |
The problems I'm testing are on the order of 1 second for factorization, 5-10 seconds start to finish. I get a 3-4x speedup using mkl + 16 OMP threads. But mumps-mpi is faster. |
Solution to issue cannot be found in the documentation.
Issue
I could investigate deep down the hole, but it seems that the latest mumps build is creating some deadlong/hang when used inside ipopt in some situation ami-iit/bipedal-locomotion-framework#882 (comment) .
Installed packages
Environment info
The text was updated successfully, but these errors were encountered: