ozaki op• Data kinds: none → table (an op determined by its arguments alone — it takes no image or data input)
• Call: import fullseye as fs; fs.ledger.fp64_emulation_probe() (to call the implementation directly, import ozakimm; ozakimm.fp64_emulation_probe(); from the registry, opsdrive.get("fp64_emulation_probe"))
Returns a table showing whether cuBLAS FP64 emulation (CUDA 13.0 Update 2 or later) is usable on this machine.
Never raises: if there is no GPU, CUDA or cuBLAS 13, returns available=False with the reason (no heavy dependencies; it only opens the libraries via ctypes and reads their versions). Returns: available (bool), rows (per candidate: library, path, loaded, cublas_version, emulation_api (whether cublasSetEmulationStrategy exists), gpu (whether a handle could be created), reason), how (how to enable it), note.
Enabling is opt-in: math mode CUBLAS_FP64_EMULATED_FIXEDPOINT_MATH, or, without changing code, CUBLAS_EMULATE_DOUBLE_PRECISION=1. The cuBLAS documentation states that the bitwise reproducibility guarantee no longer holds during emulation —— if you need reproducibility, use matmul_reproducible.
• 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
table as input)course_layout · course_occupancy · course_contains · world_build · world_camera · world_move · lidar_scan · rss_longitudinal_same
ozaki)matmul_ozaki · matmul_reproducible · ozaki_error_bound · cross_covariance_reproducible · kabsch_reproducible
*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.