View Proposal

---

Proposer

Rob Stewart

Title

Static analysis of dataflow program performance

Goal

Integration framework to enable throughput optimisation for embedded systems

Description

Dataflow languages are good models for programming embedded and configurable hardware architectures, because the distributed memory model maps well to these architectures. There's a wide variety of dataflow models, from static dataflow all the way to dynamic dataflow. Static models are easier to reason about e.g. how much data can be processed, but these programming models are restrictive. Dynamic models can express complex algorithms but reasoning about runtime behaviour is much harder. This exposes a sweet spot: can we expose moderately expressive programming models without losing all ability to reason about performance, or to generate efficient hardware. This project will explore the idea of abstracting properties of dataflow programs, to compute what the throughput performance capabilities of such programs is, without needing to run them. The technology to integrate could be (1) HoCL for programming, (2) Kiter to determine throughput performance and (2) PREESM to turn programs into executables. Possibly also using DIF as an interchangeable dataflow model format. https://github.com/jserot/hocl/ https://github.com/bbodin/kiter https://preesm.github.io https://www.researchgate.net/publication/220714226_DIF_An_Interchange_Format_for_Dataflow-Based_Design_Tools

Resources

Background

Url

Difficulty Level

High

Ethical Approval

None

Number Of Students

1

Supervisor

Rob Stewart

Keywords

Degrees