Function dwt_simd_f64 

pub fn dwt_simd_f64(
    signal: &Signal<f64>,
    wavelet: &dyn Wavelet,
    boundary: BoundaryMode,
) -> Result<DWTResult<f64>>

SIMD DWT for f64 signals.

Specialized fast path is Haar × Periodic × even-N. Anything outside that envelope (non-Periodic boundary, odd-N Periodic, or non-Haar filters) routes through dwt_generic directly — calling dwt_optimized here would error for non-Periodic / odd-N since the optimized kernels are gated to even-N Periodic, and re-dispatching through the public dwt could recurse back into SIMD for filter_len == 4 + signal_len ≥ 1000. Going straight to the canonical generic path avoids both pitfalls.

The Haar fast path runs through [dwt_simd_kernel_f64], which is portable across architectures via pulp runtime dispatch.