LJMD_Project - Group 2:

Participants

The commits in the repository correlate to usernames and people as follows:

• debarshibanerjee - Debarshi Banerjee
• Sera91 - Serafina Di Gioia
• AlexTru96 - Alexander Trujillo

The report, with the description of the procedure followed for the code optimization, and the results of the bencmark, can be found in directory Report.

Running the program:

The compile.sh script runs the program by compiling it from scratch in a build directory. It can be edited to run according to user expectations as explained below.

It takes an argument - 1 or 0, which respectively enables or disabled OpenMP support.

USAGE: ./compile.sh 1

The code is always compiled with full MPI support.

The script sets a default number of OpenMP threads using export OMP_NUM_THREADS = 4. This can also be modified locally in one's interactive shell of choice.

This can be modified as needed, of course.

The cmake command in the script can be modified as follows:

• ENABLE_TESTING (yes or no. Default: no)
• ENABLE_SIMPLE_FORCE (yes or no. Default: no)
• ENABLE_OPENMP (1 or 0. Taken from script parameters.)

Testing is disabled in the script by default since it is already automatically carried out by GitHub actions.

The SIMPLE_FORCE directive is for using a force function, where the OpenMP parallelization (with an MPI background) is purely a #pragma omp parallel for and a reduction directive for the main for loop.

The default force function uses the OpenMP custom-reduction approach, in addition to MPI, to create an efficient hybrid function.

The test_verification.sh script runs ctest to verify the tests if ENABLE_TESTING=yes is set.

The result_verification.sh script runs make check in the examples directory for 108 and 2916 size Argon systems and compares it with the reference results.