ozaki op• Data kinds: matrix × matrix → matrix
• Call: import fullseye as fs; fs.ledger.matmul_reproducible(A, B, tol: 'float' = 1.1102230246251565e-16) (to call the implementation directly, import ozakimm; ozakimm.matmul_reproducible(A, B, tol: 'float' = 1.1102230246251565e-16); from the registry, opsdrive.get("matmul_reproducible"))
Bitwise-reproducible A @ B with FP64 accuracy (ValueError if this cannot be guaranteed = fail-closed).
The Ozaki-I automatic selection of matmul_ozaki with on_insufficient="raise" fixed. The error per element is at most tol · (|A||B|)_ij + the rounding of the final assembly (a few ulp). Returns the same bits regardless of the order of the inner index (order of points) or the number of BLAS threads. Intended for point-cloud contractions such as 3×N @ N×3 or 6×N @ N×6 (for products with large outputs it is 15–83 times slower than DGEMM).
• Sample-data catalog (download URLs / licences) — 2-D uses skimage.data (BSD/public domain) plus synthetic images; 3-D lists download URLs for real data sources (Stanford, PDS, …).
• Operator provenance and references — the sources of the research/methods this op family came from.
• The canonical algorithm (author, year) and its uses are named in the family usage guide above.
• poc_reproducible_icp — py -3.11 examples/poc_reproducible_icp.py
matrix as input)world_camera · lidar_scan · relative_motion · foe_from_motion · flow_from_depth_motion · ttc_truth · world_materials · mesh_signed_volume
ozaki)matmul_ozaki · ozaki_error_bound · cross_covariance_reproducible · kabsch_reproducible · fp64_emulation_probe
*Provenance: ozakimm.py — DRIVE operator registry. This per-op note is generated by tools/opdocs.py md (do not hand-edit).*
© 2026 Kazufumi Furuse — Fullseye operator documentation. Licensed under Apache-2.0.