AMPLE: Automatic MaPping of aLgorithms for Embedded systems1

Presented here is AMPLE, a platform-based design methodology and its realization in a soft-ware tool for automatic mapping and performance optimization of algorithms to embedded system architectures. Source code of a high level programming language is initially analyzed through the use of a set of profiling tools. Analysis results provide insight on both the compu-tational intensive tasks as well as information on variable usage/sharing requirements. An ar-chitectural configuration is then selected based on a typical CPU-FPGA interconnection scheme. The code is divided into parts which will execute on a microprocessor and parts which will be handled by special purpose hardware (on the FPGA of the platform). Shared memory areas defined through the use of common variables are also mapped to the memory components of the platform. Nested loops are of special interest since they are usually the most computationally intensive tasks and are automatically parallelized and implemented in hardware in the platform’s FPGA. The presented methodology can effectively reduce design time and thus minimize time-to-market requirements as well as provide a considerable per-formance improvement in the resulting system