Um fluxo de prototipagem rápida em FPGA para algoritmos de processamento de vídeo.

Abstract

Following the constant increasing complexity of SoC devices, programmable logic devices as FPGA’s now contain millions of logic gates and can operate at speeds close to 600 MHz, enabling their use even for real time video coding and decoding. This increasing of capacity combined with fast prototyping time and relative low-cost for short production cycles makes FPGA’s a very attractive technology for systems validation or even as final market solution. However, the complexity of digital systems seems to increase faster than the advance of design techniques and tools based on RTL, creating a very tortuous path from the system modeling (usually a C++ algorithm) to the final prototype. This document proposes an alternative methodology for accelerating the FPGA prototyping of video processing subsystems for demonstrative purposes. Based on a High-Level Synthesis (HLS) design tool, the alternative design flow establishes the general rules to guide the designer from an algorithm system level C description to its FPGA prototyping, keeping satisfactory results with a minimal effort and development time. A case study using the proposed methodology is presented at the end of this document, discussing the difficulties encountered and showing the practical results obtained during its implementation. As a main contribution of this work we can cite the evaluation of high-level synthesis for the development of video processing systems. The studies and obtained results indicates a remarkable progress of high-level synthesis tools currently available on the market, which already makes them capable of being employed in the development of complex systems, as video processing systems. Additionally, the Catapult, the employed synthesis tool in the proposed flow, is now the most powerful synthesis tool on the market, being used by huge companies such as Panasonic, Siemens, STMicroelectronics, Nokia and Alcatel. Nevertheless, no Catapult licensing is currently assigned to Brazil. Thus, the presented work also contributes providing information and independent evaluations about its utility to the academy.