HPC-MultiGPU-MatrixMult

Requirements

• Python3
• NVIDIA GPU
• CUDA Toolkit
• OpenMPI
• NVIDIA Nsight Systems (optional, for profiling)
• Make

Compiling

To compile, clone the repository and use Make:

git clone https://github.com/Redy1908/HPC-MultiGPU-MatrixMult.git
cd HPC-MultiGPU-MatrixMult
make

The executable will be ./bin/main.out.

Running Locally

To generate the results file after compiling, run the executable from the project's root directory:

mkdir -p csv profiling bin plots
mpirun -np <n_process> bin/main.out <matrix_size> <tile_width> <grid_width> <grid_height> <test_name>

To generate the plots from the resulting files, install the required Python packages and run the plots.py script from the project's root directory:

python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
python3 scripts/plots.py

Outputs

Standard output

Plots will be generated in the plots/ directory.

Running on IBiSco

From the project's root directory, prepare and launch the jobs:

./scripts/run.sh

Outputs

Log files will be created in the logs/ directory:

• Standard output: logs/output.log
• Errors: logs/error.log

Plots files will be generated in the plots/ directory.

Profiling files generated by NVIDIA Nsight Systems will be saved in the profiling/ directory:

• 1 file for each MPI process