Home

Welcome to SPHINX

SPHINX is a research project focused on transferring machine learning-based optimization predictors across MLIR dialects by representing programs as ProGraML graphs.

From Programs to Vectors: A Blog Series

A Systematic Way to Represent Code with Applications to Optimization and Understanding

We are releasing a three-part series documenting the core concepts of the project, covering the translation from MLIR to ProGraML and its applications in machine learning-based optimization.

Chapter	Topic	Release Date
Chapter 1	Understanding ProGraML	Dec 5, 2025
Chapter 2	From MLIR to ProGraML	TBA
Chapter 3	Applications: Liveness Analysis	TBA

Resources

Project Links

• Repository: lac-dcc/sphinx

Technologies & References

• ProGraML: Program Graphs for Machine Learning (Cummins et al.)
• MLIR: Multi-Level Intermediate Representation
• LLVM: The Compiler Infrastructure