pub fn measure_resident_solve_throughput(
shape: EncodeShape,
reps: usize,
) -> ResidentSolveThroughputExpand description
Measure the device-resident penalized-normal-equations solve throughput for
one shape: upload X once, then time reps solves that cross only w
(H2D), rhs (H2D, fixed), and β (D2H) — the production IRLS inner step.
reps is the number of timed solves; w is perturbed per rep so each solve
is genuine work, mirroring an IRLS weight update. Returns a
ResidentSolveThroughput whose engaged flag is the false-routing guard:
on a CPU-only host (or if the device declines) it is false and the rate is
0.0.