Filippo Barbari

Hello everyone! I’m Filippo, I’m 22 and I’m currently in the middle of my master’s degree in Computer Science and Engineering at the University of Bologna, Italy. I have always been curious about parallel and concurrent programming and, after attending the High-Performance Computing course during the third year of my bachelor’s degree, I discovered my passion. During that course, prof. Moreno Marzolla taught me MPI, OpenMP and CUDA but my preferred one is by far the last one due to it being more down-to-the-metal than the others. That’s also the reason why I developed my bachelor’s thesis project in CUDA.

What are my interests?

I have always been interested in algorithms, especially non-polynomial ones, and the correlated data structures. Lately, as my passion for parallel programming grew, I tried to combine the two and ended up with numerical algorithms.

What is my project about?

I’m really excited to participate in PRACE’s Summer School of HPC 2022. This will be a great experience (to say the least) both for working and for studying. My project is the 2221, “Designing Scientific Applications on GPUs”; I will be working with Mr. Ezhilmathi Krishnasamy at the University of Luxembourg on the LibRSB library. LibRSB makes available some common operations for sparse matrices for the Recursive Sparse Blocks (RSB) format, which is its main feature, and for the Sparse BLAS standard. This library is designed to work on shared memory architectures, but can be compiled serially if needed, so it features OpenMP parallelism. My job will be to make it work work with NVIDIA GPUs, but CUDA is not a full fledged programming language like C or C++, it’s more similar to a framework, so I will actually have to implement many functions of this library in order to make it work with CUDA while maintaining the same C API. In order to assess the performance of my implementations and its scalability, I will be granted access to the IRIS HPC cluster of the University of Luxembourg (you can find its technical specifications here) which features 96 really expensive CUDA-capable GPUs: NVIDIA Tesla V100 SXM2. It will be a pleasure, and surely a unique experience, to develop on this monster of a computer and I will update you with my work in the next post so stay tuned!