Copyrights to these papers may be held by the publishers. The download files are preprints. It is understood that all persons copying this information will adhere to the terms and constraints invoked by each author's copyright. These works may not be reposted without the explicit permission of the copyright holder.

G. Xu, Tze-Meng Low, James C. Hoe and Franz Franchetti (High Performance Extreme Computing Conference (HPEC), 2017)
Optimizing FFT Resource Efficiency of FPGA using High-Level Synthesis
Comment: poster with abstract
Preprint (572 KB)
Bibtex

The practical use of High-level Synthesis (HLS) for FPGAs may compromise resource efficiency, which is a common design goal in FPGA programming. Specifically, when designing the iterative datapath for computing a FFT, combining naive HLS code with compiler optimizations can fail to saturate the BRAM ports and incur significant penalty in clock frequency. To solve this problem, we suggest that the target datapath should be explicit in HLS code. Using this solution, our FFT cores reduce the FFT latency by N cycles for the radix-2 N-point FFT comparing with previous HLS-based work and the Xilinx LogiCORE FFT. Meanwhile, our designs achieve comparable peak frequency and consume similar resources to Xilinx.

Keywords:
Fast Fourier Transform, FPGA, Synthesis, Optimizing