Abstract

This work explores interactive synthesis of compiler schedules for hardware accelerators. Using Programming by Navigation and the FuseFlow sparse machine learning compiler, the prototype guides users through valid scheduling choices while preserving access to every valid schedule supported by its library. The approach combines programmer control with guarantees that intermediate choices can lead to a valid schedule.

Article

pdf

BibTeX

@article{williams-evan2026,
  title={Interactive Compiler Scheduling with Strong Guarantees},
  author={Evan Williams and Justin Lubin and Rubens Lacouture and Olivia Hsu},
  journal={Workshop on Languages, Tools, and Techniques for Accelerator Design (LATTE) co-located with ASPLOS},
  year={2026},
  month={March}
}