Impact of Using A Modern Compiler and Optimized Workflow on the Performance of a Complex Nuclear Fusion Application for ITER

Harnessing the vast computational power of HPC systems empowers us to expedite and magnify our efforts, effectively addressing intricate challenges spanning various domains. Yet, optimizing scientific applications to fully exploit this potential necessitates substantial expertise and dedication to code optimization. This work presents a pragmatic approach tailored to enhance the performance of PARTICLE-3D (P3D), a nuclear fusion application code crucial for simulating runaway electron dynamics—a critical aspect impacting the operational lifespan of nuclear fusion machines like ITER. Transitioning from a legacy compiler dependency, the authors identified performance bottlenecks within the serial version of the code, particularly pertaining to I/O communication between subroutine calls. By meticulously restructuring the code and refining inter-subroutine communication, the article highlights the process adopted by the authors to unlock embarrassingly parallel computation workflow on the HPC cluster. The optimized version of P3D exhibits a performance improvement of several orders of magnitude compared to its base version. These optimizations not only elevated the performance but also improved the code portability, ensuring seamless integration with diverse HPC architectures equipped with modern compilers. The methodology described here can serve as a valuable case study for early researchers and code developers to enhance the performance of any such code designed for problems characterised by the similar algorithmic topologies. A recording of this session is available on YouTube: https://youtu.be/MjhhCI0GOU8