Static task scheduling for configurable multiprocessors

Abstract

Our task scheduling pass implemented in the COINS compiler uses a Binary Linear Programming model for scheduling a program into a multiprocessor system-on-chip where each processor can be accelerated with instruction set extensions. We compare our work to state of the art approaches and estimate an average speedup of 4.01 in application execution time compared to a sequential approach. We estimate on average a 1.43 times speedup over the use of multiprocessor scheduling without instruction set extensions.