In ABACUS's screen output, the labels PROCESSORS and THREADS are ambiguous and cannot clearly reflect the actual parallel configuration.
For instance, I executed the task with 28 MPI processes and 1 OpenMP thread per process using the command:
mpirun -np 28 ./abacus
However, the output shows misleading information as below:
---------------------------------------------------------
Self-consistent calculations for electrons
---------------------------------------------------------
SPIN KPOINTS PROCESSORS THREADS
1 1 28 28
---------------------------------------------------------
Existing problems
- The term
PROCESSORS is not precise. It refers to MPI tasks in practice, rather than physical CPU processors. Using PROCESSES or MPI PROCS would be more appropriate.
- The term
THREADS is confusing. It displays the total threads instead of threads per MPI rank, which may cause misunderstanding of the running setup.
In ABACUS's screen output, the labels
PROCESSORSandTHREADSare ambiguous and cannot clearly reflect the actual parallel configuration.For instance, I executed the task with 28 MPI processes and 1 OpenMP thread per process using the command:
mpirun -np 28 ./abacusHowever, the output shows misleading information as below:
Existing problems
PROCESSORSis not precise. It refers to MPI tasks in practice, rather than physical CPU processors. UsingPROCESSESorMPI PROCSwould be more appropriate.THREADSis confusing. It displays the total threads instead of threads per MPI rank, which may cause misunderstanding of the running setup.