Annotation-Driven CUDA to Tenstorrent Migration

Artifact for a paper on an MLIR-based compiler pipeline that migrates a statically shaped affine subset of CUDA to Tenstorrent Blackhole. Processor assignment, guard realization and buffer streaming are given as declarative annotations the compiler checks and lowers through tt-mlir. Holds the intermediate representations at every stage, the compiled programs for Gaussian elimination, a 2D Jacobi stencil and SYRK on a 10x10 core grid, and scripts that reproduce each timing and accuracy number, including the A100 comparison.