No abstract

No abstract

Abstract-A Large Scale Printer (LSP) is a CyberPhysical System (CPS) printing thousands of sheets per day with high quality. The print requests arrive at run-time requiring online scheduling. We capture the LSP scheduling problem as online scheduling of reentrant flowshops with sequence dependent setup times and relative due dates with makespan minimization as the scheduling criterion. Exhaustive approaches like Mixed Integer Programming can be used, but they are compute intensive and not suited for online use. We present a novel heuristic for scheduling of LSPs that on average requires 0.3 seconds per sheet to find schedules for industrial test cases. We compare the schedules to lower bounds, to schedules generated by the current scheduler and schedules generated by a modified version of the classical NEH (MNEH) heuristic [1], [2]. On average, the proposed heuristic generates schedules that are 40% shorter than the current scheduler, have an average difference of 25% compared to the estimated lower bounds and generates schedules with less than 67% of the makespan of schedules generated by the MNEH heuristic.

No abstract

The Halide DSL and compiler have enabled high-performance code generation for image processing pipelines targeting heterogeneous architectures through the separation of algorithmic description and optimization schedule. However, automatic schedule generation is currently only possible for multi-core CPU architectures. As a result, expert knowledge is still required when optimizing for platforms with GPU capabilities. In this work, we extend the current Halide Autoscheduler with novel optimization passes to efficiently generate schedules for CUDA-based GPU architectures. We evaluate our proposed method across a variety of applications and show that it can achieve performance competitive with that of manually tuned Halide schedules, or in many cases even better performance. Experimental results show that our schedules are on average 10% faster than manual schedules and over 2× faster than previous autoscheduling attempts.