We will be presenting a tutorial titled xDSL: a Python-based compiler ecosystem.
Abstract
The xDSL framework is a Python-based compiler infrastructure within the MLIR ecosystem. It aims to facilitate and speed up the prototyping of compiler concepts within MLIR without the hassle of C++-based (or other compile-to-assembly) project infrastructures. This tutorial introduces xDSL’s core concepts and practical implementation techniques through hands-on examples. In more detail, we will provide participants with the necessary information on 1) how to define, extend and combine IR dialects, 2) leverage the existing optimization capabilities to define transformation passes that operate domain-specific and across dialects, 3) lower dialects and interface with existing MLIR/LLVM workflows. By the end, attendees will understand how to develop efficient, maintainable DSL compilers using xDSL’s modular approach, and be equipped to integrate xDSL into their development workflows.
Schedule (Tentative)
| Title |
|---|
| Welcome |
| Introduction to xDSL |
| Implementing a Toy language |
| Efficient code generation for highly custom accelerators |
| Other tools and uses (GUI, end-to-end ML compilation) |
| Conclusion and Future |
Organizers
Tobias Grosser Chris Vasiladiotis Alexandre Lopoukhine
Primary Contact
Chris Vasiladiotis