Implementação de uma transformada rápida de Fourier para computação reconfigurável de alto desempenho

Abstract

Reconfigurable computing is an approach to system design where the application defines the processor structure. This approach is useful for high performance computing, allowing the creation of hybrid systems as a combination of hardware and software, presenting a higher performance than software-only implementations. This work has the goal of implementing a fast Fourier transform for possible use in reconfigurable computing. This implementation can be used to increase the performance of applications that make use of such numerical method. This work is important for exploring this new high performance paradigm and, at the same time, make applications use more efficiently the hardware resources available in a Cray XD1 supercomputer, equipped with FPGA (Field-programmable Gate Array) reconfigurable processors. The fast Fourier transform algorithm was chosen because it is a computational kernel used in several meteorological applications. This implementations was simulated and presented a satisfactory number of clock cycles to execute this function.