This repository is the result of doing the High-Performance Computing (HPC) course at the University. It aims to optimize the code and use HPC techniques in order to achieve the highes performance computation on the cluster.

The repository is structured in the following way:

• 01_Matrix_Multiplication_Optimization - bare optimisation of the matrix multiplication. Includes cache awareness, 1D matrix transforming, optimised loop ordering and loop vectorization.
• 02_OpenMP - parallelise the code from 01_Matrix_Multiplication_Optimization using the OpenMP directives. #pragma parallel for and #pragma taskloop comparison.
• 03_MPI_OpenMP_CUDA - distribute the matrix multiplication operation across multiple devices and use accelerator programming to achieve the higher performance. Use OpenMP on CPU to parallelize the matrix operations. There present 2 version of performing MPI communication: using MPI collectives and MPI one-sided communication with data stored in MPI_Win.

Each directory includes the benchmarks and a short report over the optimisation results.