pub fn run_fast_path_cpu(
current: &[[f32; 3]],
history: &[[f32; 3]],
residual_history_in: &[f32],
drift_history_in: &[f32],
width: usize,
height: usize,
lambda: f32,
k: f32,
local_aggregation: bool,
) -> FastPathCpuOutputExpand description
CPU reference implementation of the minimal inline fast-path proxy.
Inputs are flat slices of per-pixel [R, G, B] triples.
All output Vecs are pre-allocated before the inner loop; no allocations
occur inside the per-pixel computation.
This function is deterministic: identical inputs produce identical outputs.