The documents distributed by this server have been provided by the contributing authors as a means to ensure timely dissemination of scholarly and technical work on a noncommercial basis. Copyright and all rights therein are maintained by the authors or by other copyright holders, notwithstanding that they have offered their works here electronically. 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.
Publications of SPCL
|A. Ivanov, T. Schneider, L. Benini, T. Hoefler:|
|RIVETS: An Efficient Training and Inference Library for RISC-V with Snitch Extensions|
(In RISC-V Summit Europe, Jun. 2023)
AbstractThe openness and customizability of RISC-V makes it a compelling platform for executing deep learning applications. We present a library of efficient deep learning kernels for RISC-V hardware, addressing the challenge of achieving optimal performance in both training and inference. The library adopts the Snitch extensions to RISC-V, adheres to the OneDNN interface, and offers portable baseline implementations as well as platform-specific optimizations. Our optimizations that leverage Snitch extensions allow us to achieve up to 0.87 flops per clock cycle. RIVETS is a valuable tool for deep learning practitioners and researchers using RISC-V, providing portability, compatibility with other frameworks, and a baseline for performance comparison.