Module gpu_kernels 

Modules

cubic_bspline_moments

Block 5 — Cubic B-spline cell-moment tables on a fixed knot grid.

cubic_cell

GPU substrate for de-nested cubic-cell derivative moments.

row_hessian_ops

Block 9 Phase 2/3 — device kernels that consume the row-primary Hessian cache (the per-row r × r blocks materialised by [crate::bms::BernoulliMarginalSlopeFamily::build_row_primary_hessian_cache] and stored in [crate::bms::RowPrimaryEvalCache]) and emit either:

survival_rowjet

Survival marginal-slope rigid per-row NLL jet on the GPU (#932 → A100 cutover).