problems and/or doubt on parallelisation

[davide.tiana]

Dear Yamboers

I've just started using yambo. I run the different examples to get some insight on the code and how to use it.
I have problem in running YAMBO on parallel mpi.
I tried to have it on 2 nodes, 32 cores. I had yambo run properly but when in the output it is written:

* CPU-Threads :32(CPU)-1(threads)
* CPU-Threads :X_all_q(environment)-4 4 2 1(CPUs)-q k c v(ROLEs)
* CPU-Threads :SE(environment)-2 4 4(CPUs)-q qp b(ROLEs)
* MPI CPU : 1
* THREADS (max): 1
* THREADS TOT(max): 32
* I/O NODES : 1
* Fragmented WFs :no

to me it seems like a openMP parallelisation (which being on 2 nodes would be impossible). Also running as the provided example, 8 cores, but on 32 requested processors I obtained the same output and 32 files in the LOG
So I am missing something and I would like to understand where I am wrong.

Cheers
Davide

[andrea.ferretti]

Hi Davide,

since the code is produces 32 log files and states it is running on 32 cpus, it is actually running MPI-parallel on 32 tasks...
You are right though there is a problem in the report. I am trying to reproduce it. Once done I'll let you know.
Andrea

BTW: in your submission script you are not specifying the env variable controlling the number of threads (OMP_NUM_THREADS).
Since in the input file you leave the number of threads unspecified (e.g. X_Threads= 0), the actual value should be set to
OMP_NUM_THREADS. Leaving both unspecified is a bit dangerous (some environment ts set OMP_NUM_THREADS to the max number of cores as a default)
In your case you end up having n_threads = 1, so I think the environment behaves correctly, but you may want to check this explicitly

[davide.tiana]

Hi Andrea

many thanks for answering and checking.

concerning the threads I agree, it should be #SBATCH --cpus-per-task 1 but I've just started using slurm so not 100% sure.
I will add it to be sure