An Open-Source Framework for the Generation of RISC-V Processor + CGRA Accelerator Systems

Abstract

Course-Grained Recongurable Architectures (CGRAs) are programmable hardware devices containing coarse-grained processing elements that can perform ALU-like functions and having word-wide programmable interconnect. CGRAs are promising platforms for the implementation of domain-specific accelerators. In this thesis, we present a framework for building hybrid processor/accelerator systems to realize compute-kernel acceleration.Each system comprises a RISC-V processor and a CGRA. The proposed framework can generate the system automatically integrating CGRAs generated by the open-source , and the RISC-V processor from the PULP project [8]. In an experimental study, several RISC-V+CGRA systems are generated, and we observe an order-of-magnitude speedup vs. software and considerable speedup vs. a vector processor on four. With the hybrid system generation capabilities integrated into the CGRA-ME framework, it now permits hybrid system generation with different types of CGRA architectures, enabling a variety of new research on processor+accelerator architectures and programming models. I would like to express my deep and sincere gratitude to my research supervisor, Professor Jason Anderson, for giving me the opportunity to do research and providing invaluable guidance throughout the project. Without your sincere and selfless support, prompt and useful advice, I would never be where I am now. You have taught me the methodology to carry out the research and to present the work precisely and clearly. It was a great honor and privilege and work with you. I would also like to thank you for your friendship, thoughtfulness, and great sense of humor. Thank you for all that you have done. I would also like to extend my great thanks to people I have worked with during my research -Taka, Vimal, Jin Hee, Julie, Rami, Omar, Linda, Gabriel and Austin. I have learned so much from you all on critical thinking and deep insight. I could always get inspired from your advice and I have always enjoyed talking with you all. A special thanks to Nicholas Giamblanco, who recommended me to Jason's group for Meng project. Thanks for making my research experience possible.I am extremely thankful to all the friends and classmates I have met throughout my time here at the University of Toronto, for adding so much fun in my research life. Special thanks to my roommates, Manling and Qiutong, who were always there when I needed help. Thank you Qiutong for always cooking for me so that I did not starve to death and thank you Manling for always washing the dishes so that I did not need to do it. My life could not be as bright without you.Finally, I want to sincerely thank my family for their support and encouragement.Thank you for always be confident on me and believe that I can do anything. I love you, mom, dad and dear grandma.