CCA-Simulator

Simulator for Continuum Computer Architecture (CCA) (paper1, and paper2) class of designs.

Summary

The CCA Simulator enables exploring design space of the CCA class of non-Von Neumann intelligent memory systems. These systems are concieved to be highly fine-grain parallel and use event-driven mechanisms to perform computation.

It can be used to design and deploy asynchronous message-driven computations and understand the runtime behavior of AM-CCA configurations. The simulator is high-level enough to be programmed using the diffusive programming model, see our paper3, and yet low-level enough to simulate individual operon movements between CCs. In a single simulation cycle, a message can traverse one hop from one CC to a neighboring CC. We make this assumption since AM-CCA channel links are $256$ bit wide and can easily send the small operons (messages) of our tested applications in a single flit cycle. Simultaneously, a single CC, can perform either of the two operations:

1. a computing instruction, which is contained in the predicate resolution and work in the user application action, or
2. the creation and staging of a new operon.

It means that BFS and SSSP actions take $2$ to $3$ cycles of compute, whereas Page Rank action takes anywhere from $3$ to $70$ cycles of compute. When their diffusions are executed they in turn take cycles proportional to the amount of local edge-list size.

Demo of Message-Driven Execution

Static Graph BFS

Dynamic Graph BFS

Legend

Graph Applications

The Applications directory contains asynchronous message-driven applications written using the CCASimulator. Please browser through each application for build and run instructions.

Tests

The Tests/Run_All_Apps directory contains a convenient script that runs all compiles and runs all the apps. It also performs validation of the computed results.