RSVPTM : An Automotive Vector Processor
Silviu M. S. A. Chiricescu, Michael A. Schuette, R. Essick, B. Lucas, K. Moat, J. Norris · 2004
Abstruct-A myriad of sensors (i.e., video, radar, laser, ultrasound, etc.) continuously monitoring the environment will be incorporated in future automobiles. The algorithms processing the data captured by these sensors are streaming in nature and require high levels of processing power. Due to the characteristics of the automotive market, this processing power has to be delivered under very low energy and cost budgets. The Reconfigurable Streaming Vector Processing (RSVP~~ is a vector coprocessor architecture which accelerates streaming data processing. This paper presents the RSVP architecture, programming model, and a first implementation. Our results show significant speedups on data streaming functions. Running compiled code, RSVP outperforms an ARM9 host processor on average by a factor of 31 on a set of kernels. From a performance/$ and performancdmw perspective, RSVP compares favorably with leading DSP architectures. The time to market is substantially reduced due to ease of programmability, elimination of hand-tuned assembly code, and support for software re-use through binary compatibility across multiple implementations. I. INTRODUCTION Future automobiles will be equipped with a large number of sensors (i.e., video, radar, laser, ultrasound, etc.) which will continuously monitor the vehicle’s surroundings. The data captured by these sensors will be analyzed near the sensors. The relevant information will be extracted and interpreted for the driver. The algorithms processing these huge amounts of data require a high level of processing power and are highly streaming in nature. Moreover, due to packaging constraints and the desire to place the processor in proximity to the sensors, energy consumption must be held within a strict budget. Due to the high volumeflow margin nature of automotive electronics and the large number of sensors envisioned, low cost is of paramount importance. This paper presents the Reconfigurable Streaming Vector Processor (RSVP ) architecture which is able to deliver high vector processing performance with low power and at a low incremental cost, making it well suited to these future automotive applications. RSVP is able to do this because it is programmed using a Data Flow Graph (DFG) language, well-suited to describing vector operations, and has an architecture targeted at DFG execution. This allows it to achieve RSVPTM is a trademark of Motorola Inc. Other product or service names are the property of their respective owners. extreme parallelism and yet make very effective use of costly memory bandwidth, often while operating its data path as a fixed pipeline like an ASIC. Moreover, this extreme parallelism can be achieved with compiled code whose performance can scale up with improvements in the hardware without software modification, saving on software development costs. The rest of the paper is organized as follows. The next section describes the motivation of this work, followed by a description of the architecture and programming model. Next, the first implementation of the architecture is presented. Finally, benchmark results for RSVP, -946, and leading DSPs on common DSP kernels are shown.