Just-In-Time Scheduling Techniques for Multicore Signal Processing Systems

International audienceThis paper introduces a novel multicore scheduling method that leverages a parameterized dataflow Model of Computation (MoC). This method, which we have named Just-In-Time Multicore Scheduling (JIT-MS), aims to efficiently schedule Parameterized and Interfaced Synchronous DataFlow (PiSDF) graphs on multicore architectures. This method exploits features of PiSDF to find locally static regions that exhibit predictable communications. This paper uses a multicore signal processing benchmark to demonstrate that the JIT-MS scheduler can exploit more parallelism than a conventional multicore task scheduler based on task creation and dispatch. Experimental results of the JIT-MS on an 8-core Texas Instruments Keystone Digital Signal Processor (DSP) are compared with those obtained from the OpenMP implementation provided by Texas Instruments. Results shows latency improvements of up to 26% for multicore signal processing systems