Complete spline evaluation reuse comparison¶
| Scope | Previous | Shared evaluations | Observed ratio |
|---|---|---|---|
| Complete worker | 92.704 s | 86.305 s | 1.074x |
| Four CPU fits, including covariance | 20.395 s | 13.370 s | 1.525x |
Two alternating fresh-process pairs replay the original public GLIM Livox bag's first six seconds with four map rounds. This is a descriptive observation on an unreserved host. The CPU fitter now shares bounded evaluation work within each fit; both variants run the same adopted CUDA scan-odometry kernels. Whole and fit ratios have different scopes and are not multiplied.
The previous revision is 28acb5819bc254f1f8c9889c0bd4b03f2deb3c3f and the current revision
is 6182c4793120a178ce1a5e204a8a0db3833d7ea2. The input bag SHA-256 is
f57a2e85339316d4235ab693ab2876aac5e113625c7f8b0f7873b62f92e09643. No raw scans or compiler caches are redistributed.
Every worker uses one BLAS thread and a separate initially empty CuPy cache. First GPU initialization and compilation belong to whole time. The unchanged dense policy uses 0.03 m voxels, three local-map scans, a 1 nm coordinate grid and returned-motion grids of 1 micrometre translation and 0.1 microradian rotation-vector components. Those grids remain off by default.
Every complete physical/evidence field and execution report is exactly equal in both pairs after excluding provenance timestamps, commands and source revisions. All four complete fit records in each pair match exactly after excluding wall time: calibration, covariance, trajectory, costs, status, iterations and dimensions. The protocol explicitly requires this stronger equality alongside the existing physical checks. Final verdicts remain INCONCLUSIVE; this clip supplies no independent extrinsic truth.
Whole timing includes startup/imports, decoding, every initial/gyro/feedback/final odometry pass, association, fitting/covariance, noise updates, evidence, signed controls, hashing, saves and import/source verification. Fitting remains on CPU. The publication index binds all four full results, original measurements/logs, both source capsules/manifests, launcher, schema and publisher.
Audit without a GPU or bag:
python -m tools.audit_bspline_lidar_imu_benchmark docs/assets/spline_evaluation_reuse