Getting Started

Running the benchmark

The main benchmark compares several matrix-multiplication implementations, including pure Python, Numba, blocked versions, and numpy.dot.

Run it from the project root:

python benchmarks/matmul_numba_bench.py

You can optionally pass the matrix size as the first argument. If omitted, the script defaults to 512:

python benchmarks/matmul_numba_bench.py 256

The benchmark prints a performance table and opens a plot window at the end. If you are on a headless machine, you may need to configure a non-interactive Matplotlib backend before running it.

Running the example demo

The demo shows how to use the swiftpack decorator with a scheduler strategy from swiftpack.scheduler:

python demos/matmul_demo.py

If the demo succeeds, it prints:

Result verified successfully against NumPy!