Tutorial Table of Contents
- Overview
- What are FireSim, Chipyard and Gemmini?
- Tutorial Schedule
- Attendee Logistics/Requirements
- Remote attendees
- Registration
- Sponsors
Overview
We’re running a hands-on full-day tutorial on FireSim, Chipyard, Gemmini and AuRORA at MICRO 2024!
We’ll be providing access to AWS EC2 F1 instances to in-person attendees free-of-charge to interactively follow the tutorial, thanks to the generosity of AWS and Xilinx! Remote attendees, please see the remote attendees section below.
Attendees will be able to customize an industry and silicon-proven RISC-V microprocessor design, run their own high-performance FPGA-accelerated simulations of their design in the cloud, and learn how to push their design to silicon, guided by the FireSim and Chipyard developers. See the schedule below for more details.
What are FireSim, Chipyard and Gemmini?
Chipyard is a one-stop shop for generating complex RISC-V SoCs, including in-order and out-of-order processors, uncore components, vector co-processors, and other kinds of accelerators. Users can customize any component of the system and push it through automated ASIC flows (e.g. Hammer), software simulation (e.g. Verilator and VCS), and FPGA-accelerated simulation flows (e.g. FireSim) to enable agile end-to-end computer architecture research with a single re-usable toolchain. Furthermore, recent development extended Chipyard to Chiplet-yard, our new framework for systematic generation, evaluation, and verification of disaggregated chiplets in an aggregated multi-chiplet system, enabling research into heterogeneous chiplet platforms.
FireSim is an open-source FPGA-accelerated simulation framework that can simulate designs built in Chipyard and deploy them to cloud FPGAs, running complex software stacks (e.g. Linux + applications) at 100s of MHz. FireSim simulations exactly and deterministically model Chipyard designs, matching cycle-by-cycle bit-by-bit behavior of the design as if it were taped out in silicon. I/Os like DRAM, UART, and Ethernet are also modeled cycle-accurately, allowing users to model complex systems, including large clusters, beyond the capabilities of test-chips. Recent development includes adding FireAxe, enabling large-scale RTL design simulation by partitioning the designs across multiple FPGAs.
Gemmini is an open platform for full-system, full-stack DNN accelerator evaluation. Gemmini allows users to generate a variety of different DNN hardware accelerators, with different underlying system, SoC, and programming stack components. Gemmini enables users to explore and evaluate a variety of different DNN accelerator and system configurations, exposing how these different parameters interact to impact end-to-end performance and efficiency. Accelerators are integrated with AuRORA methodology which uses ReRoCC (remote RoCC interface) that disaggregates accelerators from the core for scalable integration, which is integrated into Chipyard. Gemmini has been presented at DAC 2021, where it won the Best Paper award, and AuRORA has been presented at MICRO 2023 and was chosen as IEEE Micro Top Pick (2023).
Together, Chipyard, FireSim, and Gemmini bridge the gap between open-hardware and ML accelerator, architecture, and systems research, automating many common tasks of ML, architecture, systems, and VLSI researchers in a single, easy-to-use platform.
Tutorial Schedule
| Time (CDT) | Session Name | Speaker | Slides |
|---|---|---|---|
| 8:20am | Introduction/Overview, Amazon EC2 Instance Setup, Logistics | Seah Kim | |
| 8:40am | Chipyard Basics | Jerry Zhao | |
| 9:00am | Building Custom RISC-V SoCs in Chipyard | Jerry Zhao | |
| 9:30am | Hammer VLSI Flow, Chips, and Chiplet-yard | Vikram Jain | |
| 10:00am | Coffee Break | ||
| 10:30am | Gemmini Introduction: Architecture and Programming | Seah Kim | |
| 11:10am | AuRORA Introduction | Seah Kim | |
| 11:30am | Gemmini SoC Integration with AuRORA | Seah Kim | |
| 12:00pm | Lunch | ||
| 1:00am | FireSim Introduction | Abraham Gonzalez | |
| 1:20am | FireSim Update and FireAxe | Joonho Hwangbo | |
| 1:50pm | Building Hardware Designs in FireSim | Joonho Hwangbo | |
| 2:20pm | Running FireAxe | Joonho Hwangbo | |
| 3:00pm | Coffee Break | ||
| 3:30pm | Building Software ML Workloads with FireMarshal | Abraham Gonzalez | |
| 4:00pm | Running a FireSim Simulation: End-to-End ML Workload | Vikram Jain | |
| 4:30pm | Debugging and Profiling FireSim-Simulated Designs | Abraham Gonzalez | |
| 5:00pm | Conclusion | Abraham Gonzalez | |
| 5:15pm | End of Tutorial |
Speakers/Organizers: Seah Kim, Abraham Gonzalez, Jerry Zhao, Joonho Hwangbo, Vikram Jain
Attendee Logistics/Requirements
No prior experience with FireSim/Chipyard/Gemmini/RISC-V/Chisel is necessary. To follow along with the tutorial on the EC2 instances we provide, attendees will need to bring a laptop with an ssh client installed. Users may want to consider installing mosh (https://mosh.org/), a reliable ssh-client replacement. Our EC2 instances will also support connecting via mosh.
Remote attendees
Live remote attendance is not available for this event. However, we will be recording the sessions and will publish them here, along with a script that allows you to recreate the tutorial AWS AMI on your own account.
Registration
To attend the tutorial in-person, you must register for MICRO 2024. Please make sure you select the FireSim/Chipyard/Gemmini tutorial on the registration form, so we can provision a sufficient number of EC2 instances for attendees.
Stay tuned!
We will continue to update this page as the tutorial is finalized. Join the FireSim mailing list and follow the FireSim Twitter account to stay up-to-date as we finalize the tutorial!
Sponsors
Thanks to the following sponsors for their generous support in running this tutorial:
