BasisLifecycle

Published configurations

H1 first-binding, identity-working-set and sample-sweep driver.

Measured operation

Measures reference association, identity churn and sample-sweep costs.

The H1 component basis is \(\Phi_{a,c}=\phi_a E_c\). Scalar and vector controls retain the same scalar basis and component convention. Associated Single scopes bind sample zero; Sweep scopes bind every mapped sample. AssociatedFresh measures one first binding before any numerical warmup; its direct checks run afterwards. AssociatedOne repeats one owned formula. AssociatedNine cycles nine distinct formula identities, targeting the current eight-entry reference store. UnprovenNine uses the same owned points but only origin metadata, which must never authorize reference-table substitution. Unproven Single binds the last sample while origin metadata remains index zero; Unproven Sweep likewise retains origin index zero while actual coordinates vary.

Measurement boundary

Mesh, space, formulas and mapped samples are constructed outside timing. Except for AssociatedFresh, two untimed priming cycles establish the identity protocol. Fresh scopes force one iteration and one repetition; a controlled process-first study must additionally select one case per new process. The timed operation includes identity selection, binding and all active-value stores. Identity churn can include allocation, value and derivative tabulation for implementations using the reference store. It is not process-first timing; shared element initialization is already warm. No query-count instrumentation is inserted into timed formula access. Public table-presence queries occur only between priming cycles and report participation rather than requiring a particular implementation to use the cache.

Numerical qualification

Every output is checked exactly against direct scalar basis evaluation. Logical dimensions and reference authority are checked outside timing; Fresh scopes perform all numerical qualification after the first timed bind. Changed output must fail. Only one active-value output buffer is retained.

Recorded results

Each case retains its original inputs, iteration policy and correctness checks. Raw reports preserve counters, repetitions and the Google Benchmark context. CPU, wall and manually supplied times are labeled separately; unlike operations and configurations are not averaged together.

Timing history

RodinH1BasisLifecycleBenchmarks · auto · slow · ranks: 1