Schedules Are Solvable Symbols: Tuning-Free Compilation of Tile Programs on Dataflow Architectures
Summary
Loom is a tuning-free symbolic compiler framework for tile-based SPMD programs on spatial dataflow architectures. It treats compilation as a hardware-explicit static optimization problem, exploring mapping and communication candidates while keeping parameters symbolic, and solves scheduling via CP-SAT at compile time. The approach achieves performance parity or superiority over vendor-optimized libraries on Tenstorrent hardware without per-shape profiling, suggesting a retargetable, explainable path away from profiling-based tuning.