Struct RemlOuterAdmission 

pub struct RemlOuterAdmission {
    pub n: usize,
    pub p: usize,
    pub num_rho: usize,
    pub family: Option<PirlsLoopFamilyKind>,
    pub curvature: PirlsLoopCurvatureKind,
    pub gpu_available: bool,
}

Inputs to [should_run_reml_outer_on_device]. The admission predicate for routing the outer REML BFGS-over-ρ loop onto a fully device-resident driver (rather than the host orchestrator that hops out per step).

Fields are intentionally lifted from data the CPU REML entry has on hand before it touches the seed generator or the inner P-IRLS loop, so the admission check is allocation-free and can short-circuit before any device call.

Fields

n: usize

Active design rows (post-transform).

p: usize

Active design columns / penalised-Hessian dimension.

num_rho: usize

Number of smoothing parameters ρ the outer BFGS optimises over.

family: Option<PirlsLoopFamilyKind>

Inner family / link pair the device-resident PIRLS loop can evaluate. None means the family does not map onto the six JIT-cached row kernels — the outer loop must stay on the host orchestrator because the inner step would already hop out anyway.

curvature: PirlsLoopCurvatureKind

Curvature surface the inner loop will use; tied to family via pirls_loop_curvature_for.

gpu_available: bool

True when the CUDA runtime is initialised on this host.

Trait Implementations

impl Clone for RemlOuterAdmission

fn clone(&self) -> RemlOuterAdmission

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for RemlOuterAdmission

impl Debug for RemlOuterAdmission

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.