You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Currently numerical/grid/grid.f90:grid_check() does some basic sanity checks. However a lot of bad cases can get through, causing a non-obvious error like putting NaNs in plasma variables or MUMPS failing to solve. Normally, the Python front-end will only give "good" number of MPI images. However if a user manually specifies the number of MPI images, grid_check() often doesn't catch the problem, and so we get NaN check errors or MUMPS memory errors.
Examples triggering error
there are many more cases that will fail, this is just one example of each.
We continue to have issues with the mpi image count checking and init. As of the latest commit specifying a manual grid is the only way to be sure that it will run correctly... This appears to be relatively recent as my students are using versions several weeks old and have not complained about this bug at all.
Activity
[-]Should add Fortran MPI image count check[/-][+]enhance Fortran MPI image count check[/+]mattzett commentedon Jun 11, 2020
We continue to have issues with the mpi image count checking and init. As of the latest commit specifying a manual grid is the only way to be sure that it will run correctly... This appears to be relatively recent as my students are using versions several weeks old and have not complained about this bug at all.
[-]enhance Fortran MPI image count check[/-][+]error in Fortran MPI image count check[/+]scivision commentedon Aug 19, 2020
This has been fixed by checks so that these configurations promptly error like: