lapidary's Introduction

Lapidary: CGRA Simulator for Multi-Tenant Workload

What is Lapidary?

Lapidary aims to be the CGRA architecture simulator for multi-tenant workloads. Lapidary allows developers to implement their own hardware and scheduler and quickly calculate the performance.

Quick Installation

Usage

python run.py

Contributors

Taeyoung Kong

Citations

@inproceedings{lapidary_wddsa2022,
  author    = {Taeyoung Kong and
               Kalhan Koul and
               Priyanka Raina and
               Mark Horowitz and
               Christopher Torng},
  title     = {Hardware Abstractions and Hardware Mechanisms to Support Multi-Task Execution on Coarse-Grained Reconfigurable Arrays},
  booktitle = {The 1st Workshop on Democratizing Domain-Specific Accelerators, {WDDSA}},
  year      = {2022},
}

lapidary's People

Contributors

Watchers

lapidary's Issues

Fix utilization calculation

If task is mapped to PRR, it assumes PRR is utilized even though it's not doing anything.

Fix the delay requirement for stream distribution

Fix.

Separate App scheduler and Task scheduler

Instead of having task scheduler, we can have app scheduler with infinite queue size.

Improve workload description

Change app to module to make it consistent with CGRA.
Also, for each module, represent II, latency, and input/output size to calculate the total runtime.

Recommend Projects