Enum InnerSolveMode 

pub enum InnerSolveMode {
    DeviceResident,
    DeviceReupload,
    CpuReference,
}

Per-iterate solve backend for the resident inner Newton loop.

All three modes run the IDENTICAL host control flow (run_inner_loop): residual-gradient assembly, LM trust-region accept/reject, ridge schedule. They differ ONLY in how the per-iterate arrow step is computed, which is exactly the residency lever #1017 measures:

• InnerSolveMode::DeviceResident — the Phase-3 fix: factor the constant Hessian blocks ONCE into a crate::gpu_kernels::arrow_schur::ResidentArrowFrameHandle and, every iterate, upload only the O(n·d + k) gradient and read back only δ. No per-solve D/B re-upload, no per-solve POTRF.
• InnerSolveMode::DeviceReupload — the BEFORE path: call solve_arrow_newton_step per iterate, which re-packs and re-uploads D/B/g and re-runs the per-row POTRF + border Schur factor every call. This is the residency baseline the bench divides against.
• InnerSolveMode::CpuReference — the dense f64 oracle (re-factors per iterate on the host), used for the correctness parity check.

Variants

DeviceResident

DeviceReupload

CpuReference

Trait Implementations

impl Clone for InnerSolveMode

fn clone(&self) -> InnerSolveMode

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for InnerSolveMode

impl Debug for InnerSolveMode

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

Formats the value using the given formatter.

impl Eq for InnerSolveMode

impl PartialEq for InnerSolveMode

fn eq(&self, other: &InnerSolveMode) -> bool

Tests for self and other values to be equal, and is used by ==.

fn ne(&self, other: &Rhs) -> bool

Tests for !=. The default implementation is almost always sufficient, and should not be overridden without very good reason.

impl StructuralPartialEq for InnerSolveMode