Scalasca
Scalasca is a performance analysis tool that supports large-scale systems,
including IBM Blue Gene and CrayXT and small systems.
The Scalasca provides information about the communication and synchronization among
the processors. This information will help to do the performance analysis, optimization,
and tunning of scientificcodes. Scalasca supports OpenMP, MPI, and hybrid programming model,
and a analysis can be done by using the GUI which can be seen in below figure.
Environmental models for Scalasca on ULHPC¶
module load purge
module load swenv/default-env/v1.1-20180716-production
module load toolchain/foss/2018a
module load perf/Scalasca/2.3.1-foss-2018a
module load perf/Score-P/3.1-foss-2018a
Interactive Mode¶
Work flow:
# instrument
$ scorep mpicxx example.cc
# analyze
scalasca -analyze mpirun -n 28 ./a.out
# examine
$ scalasca -examine -s scorep_a_28_sum
INFO: Post-processing runtime summarization report...
INFO: Score report written to ./scorep_a_28_sum/scorep.score
# graphical visualization
$ scalasca -examine result_folder
Batch mode¶
Shared memory programming (OpenMP)¶
#!/bin/bash -l
#SBATCH -J Scalasca
###SBATCH -A <project_name>
#SBATCH -N 1
#SBATCH -c 16
#SBATCH --time=00:10:00
#SBATCH -p batch
module load purge
module load swenv/default-env/v1.1-20180716-production
module load toolchain/foss/2018a
module load perf/Scalasca/2.3.1-foss-2018a
module load perf/Score-P/3.1-foss-2018a
export OMP_NUM_THREADS=16
# analyze
scalasca -analyze ./a.out
# examine
$ scalasca -examine -s scorep_a_28_sum
INFO: Post-processing runtime summarization report...
INFO: Score report written to ./scorep_a_28_sum/scorep.score
# graphical visualization
$ scalasca -examine result_folder
Distributed memory programming (MPI)¶
#!/bin/bash -l
#SBATCH -J Scalasca
###SBATCH -A <project_name>
#SBATCH -N 2
#SBATCH --ntasks-per-node=28
#SBATCH --time=00:10:00
#SBATCH -p batch
module load purge
module load swenv/default-env/v1.1-20180716-production
module load toolchain/foss/2018a
module load perf/Scalasca/2.3.1-foss-2018a
module load perf/Score-P/3.1-foss-2018a
scalasca -analyze srun -n ${SLURM_NTASKS} ./a.out
Tip
If you find some issues with the instructions above, please report it to us using support ticket.
Last update: November 13, 2024