Bandedness Propagation Analysis of Tensor Compilers This repository implements the Bandedness Propagation Analysis (BPA) to propagate bandwidth information across computational graphs. The analysis is implemented in MLIR using the `linalg` dialect and a custom `DIA` dialect to compress banded matrices. The analysis is implemented as an MLIR pass that annotates operators with bandwidth information, which is then used by a custom code generator to produce banded-aware loop nests that only compute within the necessary bands proved by the analysis.