Struct PcgDiagnostics 

pub struct PcgDiagnostics {
    pub iterations: usize,
    pub matvec_calls: usize,
    pub precond_apply_calls: usize,
    pub ridge_escalations: usize,
    pub final_relative_residual: f64,
    pub stopping_reason: PcgStopReason,
    pub mixed_precision_status: MixedPrecisionStatus,
    pub used_device_arrow: bool,
    pub injected_host_procedural_matvec: bool,
}

Per-solve instrumentation counters returned alongside the PCG solution.

All fields default to zero; callers that do not need diagnostics simply ignore the value. The struct is Copy so passing it through return tuples is zero-overhead.

Fields

iterations: usize

Number of CG iterations executed.

matvec_calls: usize

Total calls to the Schur matvec A·p.

precond_apply_calls: usize

Total calls to the preconditioner M^{-1}·r.

ridge_escalations: usize

Number of times the LM ridge was escalated before a successful factor.

final_relative_residual: f64

Relative residual at termination; 0.0 when the RHS was zero.

stopping_reason: PcgStopReason

Why the loop stopped.

mixed_precision_status: MixedPrecisionStatus

Mixed-precision certificate outcome for this solve.

used_device_arrow: bool

True only when the reduced-Schur solve was actually executed on the device: either the fully device-resident batched Arrow-Schur Direct sequence (try_device_arrow_direct → solve_arrow_newton_step) or the device-resident matrix-free SAE PCG (solve_sae_matrix_free_pcg, which runs the matvec in CUDA kernels over device-resident frames). It is NOT set merely because a GPU runtime exists and a dispatch gate fired (#1209).

injected_host_procedural_matvec: bool

True when a reduced-Schur matvec backend was injected through maybe_inject_gpu_schur_matvec but the matvec itself runs as a host (CPU Rust/Rayon) procedural closure — both the matrix-free build_row_procedural_matvec branch and the cuda::build_schur_matvec_backend branch return host closures that evaluate Σ_i Y_iᵀ(Y_i x) on the CPU, even when a CUDA context was opened to build the per-row factors. This path must NOT report used_device_arrow: the arithmetic is host-side (#1209). Distinct field so perf accounting never mistakes a host procedural matvec for true device execution.

Trait Implementations

impl Clone for PcgDiagnostics

fn clone(&self) -> PcgDiagnostics

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for PcgDiagnostics

impl Debug for PcgDiagnostics

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

Formats the value using the given formatter.

impl Default for PcgDiagnostics

fn default() -> PcgDiagnostics

Returns the “default value” for a type.