The current CPU-based implementation has performance limitations on GPUs, due to code divergence, virtual function calls, and complexity (deep call stack) leading to using many registers. This makes geometry a major hotspot in GPU-based simulation workflows (AdePT) and calls for a specialized implementation. The meeting main objective is to converge on a promising approach, sketching a plan of work and a roadmap based on a realistic estimate of the work and resources needed.
Choose timezone
Your profile timezone: