The MLIR Sparsifier team is deprecating the PyTACO test suite (which was actually a full python framework with tests residing in test/Integration/Dialect/SparseTensor/taco/). The framework was one of the very first end-to-end examples for MLIR, and provided a fun Pythonesque-way to define sparse operations using tensor index expressions, as illustrated below for SpMV:
import mlir_pytaco_api as pt
# Define formats for storing the sparse matrix and dense vectors.
csr = pt.format([dense, compressed])
dv = pt.format([dense])
...
A = pt.read("matrix_market_example.mtx", csr)
...
# Define the SpMV computation
y[i] = A[i, j] * x[j] + z[i]
However, the underlying implementation was becoming a bit of a maintenance nightmare due to many recent improvements in the sparsifier that were hard to back-port. In addition, the framework was meant purely for testing functionality, not for performance testing, and we wanted to avoid confusion on how well a final system could and should perform.
Therefore, we decided to simply remove the suite for now. But… not to worry, something much better is coming!