Changelog

Performance

• Fix O(N²) codegen offset accumulation and regex bottleneck — significant compilation speedup for large kernels

Bug Fixes

• Fix MMA NodeIndex in value numbering pass
• Always commit warpgroup before WGMMA store — correctness fix for warp-specialized kernels

SDK

• SDK infrastructure updates

Documentation

• Add coding style guide for contributors

Highlights

• Open-source release of the CroqTile compiler and runtime
• Core TileFlow programming model with full GPU kernel compilation pipeline
• Choreo DSL with zero-cost abstractions over TMA, DMA, and MMA operations

Infrastructure

• Public repository setup with submodules and build tooling
• LIT test framework integration for compiler validation