DSL-based acceleration of automotive environment perception and mapping algorithms for embedded CPUs, GPUs, and FPGAs

Abstract

The availability and sophistication of Advanced Driver Assistance System (ADASs) are becoming increasingly important for customers when making purchasing decisions and thus also for the manufacturers of such systems. The increased demands on functionality have also increased the demands in computing power and today’s standard processors in automotive Electronic Control Unit (ECUs) struggle to provide enough computing power for those tasks. Here, heterogeneous systems, for example consisting of Central Processing Unit (CPUs), embedded Graphics Processing Unit (GPUs), and Field-Programmable Gate Array (FPGAs) provide a remedy. These heterogeneous systems, however, increase the development effort and the development costs enormously. In this paper, we analyze the extent to which it is possible to automatically generate code with the help of a Domain-Specific Language (DSL) for typical algorithms in the field of environment perception and environment mapping. We show that with the Heterogeneous Image Processing Acceleration (Hipacc) framework it is possible to generate program code for CPUs, GPUs, and FPGAs. After that, we compare for selected algorithms the execution times of the automatically generated code with hand-written variants from the literature.