Optimising multi-loop programs for heterogeneous computing systems

This paper presents a method for optimising parallelisation and scheduling of task graphs containing representation of loops for implementation in heterogeneous computing systems with both software and hardware processors. The method integrates loop unrolling with task scheduling and determines the extent to which each loop should be unrolled to maximise performance, while meeting size constraints. A performance-driven strategy is proposed to find the best unrolling factor for each loop, such that the closer the match of run-time conditions and compile-time parameters, the higher the performance. Experimental results obtained using a speech recognition system show the proposed method outperforms an approach without unrolling by 2.1 times, and using the processing time of a 2.6 GHz microprocessor as a reference, a speed up of 10 times can be achieved when compile-time and run-time parameters are matched, while the performance drops gradually when they are different.