Forward-mode automatic differentiation for the RISC-V RV32IMF instruction set.

Abstract

Automatic differentiation (AD) is a methodology to accurately evaluate derivative information. This technique is used in many applications of computational and data science. While existing tools implementing the technology of automatic differentiation are almost exclusively software-based, we propose an alternative approach that is hardware-based. Our proposed method integrates the forward mode of automatic differentiation directly within the microarchitecture of a RISC-V core, enabling the inherently parallel computation of multiple directional derivatives in hardware.