One Solution for Deterministic Scheduling on GPU for Automotive Algorithms

In order for higher levels of autonomous driving to be achieved, the demanding environmental perception and decision-making algorithms are developed. Meeting real-time system requirements is a mandatory task for all of these algorithms to be properly utilized. In order to meet these requirements, GPUs are widely used, and one of the biggest challenges is providing the deterministic scheduling of different tasks on GPU based on their priority classes. The complete determinism which ensures the absolute control of processes on GPU has still not been achieved. In this paper, we propose and evaluate one method that allows scheduling of tasks with different types of time constraints.