GPU Optimizations for HEP Analysis in ROOT

This project focuses on improving the runtime performance of High Energy Physics (HEP) analyses by leveraging GPUs, which are increasingly common coprocessors used to accelerate computationally intense operations. Histogramming is a core operation in HEP analysis, however the current GPU implementation is bounded by memory transfers. Hence, we implemented a generic batch histogramming kernel to fill multiple histograms at the same time. In future work, this implementation can be used to improve data reuse. Additionally, we explored offloading define actions to the GPU, which showed the potential for significant performance gains, up to speedups of $100\times$ compared against equivalent multi threaded CPU implementations.