Expand description
Bernoulli marginal-slope GPU kernels.
These modules contain BMS-specific row math. Generic CUDA runtime, memory,
policy, and diagnostic plumbing stays under gam_gpu.
Modules§
- device_
pcg - flex
- Bernoulli marginal-slope FLEX GPU policy and backend probe.
- row
- Stage 2 of the BMS FLEX row kernel — per-row math that turns per-cell
derivative moments (built by Stage 1 in
src/gpu/cubic_cell/mod.rs) into a row gradient and row-primaryr × rHessian.