Workshops Archive

Accelerating Fast Fourier Transforms with the RISC-V Vector Extension


Workshop: International Workshop on RISC-V for HPC (RISCVHPC)

Authors: Daniel Seibel, Kaveh Haghighi Mood, Stepan Nassyr, Prateek Chawla, Jayesh Badwaik, and Andreas Herten (Juelich Supercomputing Centre, Forschungszentrum Jülich)

Abstract: The RISC-V Vector extension (RVV) enables greater flexibility for Single Instruction, Multiple Data (SIMD) and vector processing compared to established extensions like AVX for x86. For Fast Fourier transforms (FFTs) in particular, variable vector lengths, segmented loads and stores, and permutation instructions allow for novel vectorization strategies previously not feasible with traditional SIMD. In this work, different vectorization schemes for FFTs are derived by systematically matching unique RVV features to complex data flows and arithmetic operations. The resulting FFT algorithms are then evaluated on a wide range of RVV 1.0 hardware, including the SpacemiT K1 and K3, the Sophon SG2044, and the SiFive X280. The results demonstrate that achieving high performance across diverse platforms requires four distinct vectorization paths, as the optimal strategy depends heavily on the hardware. Ultimately, these results highlight the necessity of dynamic, run-time kernel selection to sustain optimal performance in FFT across the heterogeneous RISC-V ecosystem.


Back to International Workshop on RISC-V for HPC (RISCVHPC) Archive Listing

Back to Full Workshop Archive Listing