Hackathon: Legacy Code Meets Tensor Compute

This hackathon is in-person at Daisytuner Headquarters, Darmstadt, Germany

View details and join hackathon --> https://compilers-and-processors.devpost.com/

Hosted by Daisytuner

In this hackathon, you'll take on the challenge of compiling legacy code to run efficiently on the Tenstorrent Wormhole, a modern tensor processors.
But you're not starting from scratch—we'll be building on top of dace/sdfglib, implementing stateful dataflow multigraphs (Ben-Nun et al., SC'19), and bringing in ideas from Tensorize (Brauckmann et al., CGO’25) to optimize code for AI-native execution models.
This is your chance to explore data-centric compilation and modern tensor chips.

### Requirements

Participants will:

• Analyze and transform legacy codes using stateful dataflow multigraphs.
  - Extend dace/sdfglib to support the lifting of tensor operations from loops, inspired by Tensorize.
  - Target a state-of-the-art tensor processor, a Tenstorrent Wormhole
  - Develop IR passes, scheduling strategies, or new representations that help legacy code embrace tensorized compute.