A Task Compilation Framework Design for a Heterogeneous Course-Grained Reconfigurable Processor

For the heterogeneous course-grained reconfigurable processor-GReP, we proposes an extension language called GR-C, which is based on ANSI C and friendly to programmer, and a full automatic task compilation framework is designed. The task compiler frontend takes the GR-C as input and the parser transforms the GR-C into an intermediate representation (IR) with static single assignment form. The task compiler backend takes the IR as input and generates executable files and configuration information after the steps of software-hardware partition, code optimization, task scheduling and mapping and so on. The experimental results of typical computing-intensive algorithms like IDCT and FIR show that our task compiler efficiently exploits the parallel-computing ability of the heterogeneous course-grained reconfigurable processor, which proves the rationality and feasibility of our GR-C language and automatic compilation framework.