...
| Code Block |
|---|
[@blog1 test02]$ salloc -A <project-name> -t 30:00 -N 2 --tasks-per-node=3 -c 1 # Note that the application is picking up the 6 requested tasks: (2 nodes x 3 tasks per node) [@m001 test02]$ export OMP_NUM_THREADS=1 [@m001 test02]$ srun brains param_con ========================BRAINS========================== Starts to run... 6 cores used. continuum data points: 125 con scale: 7.465401e-01 1.339513e+00 RM rcloud_min_max_set: 0.000000 1000000.000000 Tcon_min_max: -10.000000 10.000000 # Print parameter name in .//data/para_names_con.txt #======================================================= # Starting CDNest. # Use 6 cores. ... |
...
Observation
We do not see the Use ? cores changing when the cpus-per-task is updated. This value appears to only be related to nodes multiplied by the tasks-per-node. This does make us question whether setting OMP_NUM_THREADS to be equal to cpus-per-task has any effect on the underlying computation. We would ask if any users know whether OpenMP is utilized within the code (it might be within the underlying library dependencies gsl and lapack) to let us know, and we will update this page accordingly.
Examples
As of 20221212, during testing, ARCC has noticed that the required structure for examples/data is slightly different than in the documentation and within the GitHub repository. We followed the following structure:
...