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.

L. Tang, S. Chen, K. Harisrikanth, G. Xu, K. Mai and Franz Franchetti (Proc. High Performance Extreme Computing (HPEC), 2022)
A High Throughput Hardware Accelerator for FFTW Codelets: A First Look
Preprint (9.7 MB)
Bibtex

The Fast Fourier Transform (FFT) is a critical computation for numerous applications in science and engineering. Its implementation has been widely studied and optimized on various computing platforms, with the FFTW library becoming the standard interface in HPC. In this work, we propose hardware acceleration of the FFTW library by putting a software codelet into hardware. The hardware is exposed to the user through an FFTW-compatible software library while actual computation takes place behind the scenes on a custom accelerator. To demonstrate a first look at this idea, we design a high throughput accelerator for FFTW twiddle codelets. The FFT hardware is automatically generated using SPIRAL and a test chip is fabricated in a TSMC 28nm process. We provide measured results of the test chip and discuss many opportunities for future work.

Keywords:
FFT, FFTW, Acceleration, Fast Fourier Transform