Changelog

Features

• TileFlow Core: Complete implementation of the TileFlow programming paradigm
• Symbolic Dimensions: Dynamic shape support with compile-time validation
• Compile-Time Safety: Tiling mismatch detection, shape propagation checks
• CUTE Backend: Code generation targeting NVIDIA CUTLASS/CUTE
• Visualization: Tiling behavior visualization for debugging

Improvements

• 40% less code compared to equivalent CUDA implementations
• Zero-cost abstractions over hardware-specific DMA operations
• Automatic memory barrier insertion