benches

SRC benchmarks

This directory collects the experiments used to measure the performance of the different SRC variants. They are grouped by the kind of workload exercised:

• primitives/ — Synthetic micro-benchmarks of the four core SRC primitives (MPO–MPO, MPO–MPS, …). Used to assess speedup and accuracy of individual primitives versus quimb references on a fixed problem size. See primitives/README.md.
• stack/ — One-shot SRC over stacks of trains against sequential pairwise application, by stack depth: accuracy against dense references and wall time on small chains, and quench dynamics of 50 and 100 qubits on the GPU. See stack/README.md.
• large/ — Out-of-core SRC of N . V . M . U with a large M read from disk, on one GPU or several: plan, wall time, memory peaks, stall time and parallel efficiency. See large/README.md.

The scripts need the bench dependency group (included in dev):

uv sync --group bench

HPC clusters

The recorded results come from the Booster partition of Leonardo at CINECA. The HPC clusters section of the documentation covers the setup and has an example job script; pass it the benchmark command, e.g. python benches/large/bench_large.py run ....