========================= Release-candidate sitting ========================= A path under ``benchmarks/`` on this page names a record of the d-SEAMS 2.0 benchmark campaign. The records and the Snakemake rules that write them are in the private paper repository and the Hugging Face dataset ``d-SEAMS/dseams2-benchmark-records`` (``docs/records/README.md`` names the revision); the scripts under ``scripts/`` that time a run stay here. The numbers below are ``benchmarks/rc3-sitting.csv``, written by ``scripts/h2h_rc3_sitting.py``, in the benchmark records dataset. The tables below are rendered from that file, with each stored median and interquartile range checked against the samples. Methods ------- One warm-up is discarded. The column ``n`` is the number of timed calls that remain. In this file it is 6 on every row. ``samples_ms`` lists those calls, in milliseconds, to three decimal places. The median of an even count is the mean of the two central samples after sorting. The interquartile range is the type-7 gap: the linear quantile at position :math:`q(n-1)`, with :math:`q = 0.75` minus :math:`q = 0.25`, which is ``quantile_linear`` in ``scripts/h2h_protocol.py``. ``median_ms`` and ``iqr_ms`` are that pair, written to three decimal places. ``threads`` is the width requested for that row. The worker is started with ``OMP_NUM_THREADS`` set to that width and ``OMP_DYNAMIC=FALSE``. One thread is pinned with ``taskset -c 0``. Eight threads are pinned with ``taskset -c 0-7``. Each guest code then receives the same width through its own setter (``set_num_threads``, ``OVITO_THREAD_COUNT``, ``num_threads`` on the Rust ring call, ``n_threads`` for vesin). ``cores`` is ``Cpus_allowed_list`` from ``/proc/self/status``, read in the worker. It is the cpu affinity of the timed process, not a model name. The host name is not in the file. A one-thread guest that does not change its mask reports ``0``. d-SEAMS calls ``seams::widenAffinity`` before the timed work, which puts back every cpu the kernel still allows, so some one-thread d-SEAMS rows report ``0-7`` while OpenMP itself was one thread. Every eight-thread row reports ``0-7``. Ideal ice Ic at 110592 and 884736 molecules is the neighbour, :math:`q_4`, :math:`q_6`, CHILL+, ring, and PTM block. Cages are the GenIce cells, where cages exist. King's shortest-path count is included only at sizes whose first call returned within 10 minutes. The algorithm pages are `primitive rings `_, `TRACE and signature cages `_, `the symmetry classifier `_, and `carried rings `_. The signature-search widths, and the kernel teams from ``848426f9``, are `thread controls `_. .. include:: performance-tables.rst