FFT Compiler: From Math to Efficient Hardware

This paper presents a high-level compiler that generates hardware implementations of the discrete Fourier transform (DFT) from mathematical specifications. The matrix formula input language captures not only the DFT calculation but also the implementation options at the algorithmic and architectural levels. By selecting the appropriate formula, the resulting hardware implementations (described in a synthesizable Verilog description) can achieve a wide range of tradeoffs between implementation cost and performance. The compiler is also parameterized for a set of technology-specific optimizations, to allow it to target specific implementation platforms. This paper gives a brief overview of the system and presents synthesis results.