GPU Porting of the PLUTO Code for Computational Plasma Physics using OpenACC

The simulation of astrophysical plasmas presents significant computational challenges because of the complex and dynamic nature of these systems. The PLUTO code, widely used for such simulations, traditionally relies on CPUbased computations. This thesis explores the porting of the PLUTO code to Graphics Processing Units (GPUs) using OpenACC, a directive-based parallel programming model designed for high-performance computing. The primary objective of this work is to enhance the computational efficiency and scalability of PLUTO, exploiting the parallel processing capabilities of modern GPUs. I began my work with the GPU porting of a reduced version of PLUTO and then incrementally added the missing pieces of the original code. The implementation details are meticulously documented, together with several challenges faced, such as computational bottlenecks and necessary adaptation of the code to GPU computing. Performance evaluation is conducted on different high-performance computing (HPC) clusters in Europe, benchmarking the GPU-accelerated PLUTO, now called gPLUTO, against its CPU counterpart. The results demonstrate significant speed-up and improved efficiency. Importantly, the study also highlights the potential of GPUs to significantly reduce the power consumption of scientific computing tasks, thereby contributing to more sustainable research practices. Specific astrophysical problems have already been studied with gPLUTO and discussed here, providing insight into the advantages of GPU employment for scientific research. The thesis concludes with potential improvements to the code. This work not only contributes to the field of computational astrophysics by enhancing the performance of the PLUTO code but also serves as a valuable example for scientists aiming to harness GPU computing for complex numerical simulations while also promoting energy-efficient computing practices.