Compute 1-D Gaussian kernel weights for a single non-uniform coordinate. The coordinate `x` must already be shifted to `[0, 2π)`.

Generate the public dispatcher function for sizes 2, 4, 8 (all ISAs). Priority on `x86_64` (f64): AVX-512F > AVX2+FMA > AVX > SSE2 > scalar. Priority

Parse `gen_rader_codelet!(N)` input and dispatch. # Errors Returns `syn::Error` if the input is not a valid integer literal or the prime is not in {1

Generate window coefficients for the given window type and length. For `n == 1` the function always returns `[1.0]` regardless of window type. All tr

Compute the geometric mean of a non-empty slice of positive values.

Compute gradient of a scalar loss with respect to FFT input. Given ∂L/∂(FFT(x)), computes ∂L/∂x.

Generate Hann (raised cosine) window. `w[n] = 0.5 * (1 - cos(2πn/(N-1)))`

Compute the instantaneous phase of a real signal. Returns `atan2(im, re)` for each sample of the analytic signal, yielding values in the range `[-π,

Convenience function for 1D Complex-to-Real FFT with normalization. Takes N/2+1 complex values (conjugate symmetric) and produces N real values. This

Convenience function for batched 1D Complex-to-Real FFT with normalization. Performs `howmany` independent C2R FFTs, each producing `n` real values.

Check if n only has prime factors <= max_factor.

Magnitude spectrum: |X\[k\]| for each bin.

Generate a synthetic input signal of length `n`.

Build a minimal valid wisdom string with `n` entries starting at `base_hash`.

Maximum absolute error across all bins (f64).

Compute log-mel spectrogram of a signal. Applies STFT with a Hann window, computes the power spectrogram, applies the mel filterbank matrix, and take

Modular inverse of `a` modulo `m` via the extended Euclidean algorithm. Returns `None` if `gcd(a, m) != 1` (i.e., the inverse does not exist).

Naive O(n²) DFT reference for f32.

Naive DFT of real input — reference for verification.

3D NUFFT Type 1: Non-uniform to uniform. Given `M` non-uniform sample points `(xj, yj, zj) ∈ [-π, π)³` with complex strengths `cj`, computes the 3-D

Compute Type 1 NUFFT (non-uniform → uniform). # Arguments `points` - Non-uniform sample locations in [-π, π] `values` - Complex values at the sample

Compute Type 2 NUFFT (uniform → non-uniform). # Arguments `coeffs` - Uniform Fourier coefficients `points` - Non-uniform evaluation points in [-π, π

Get number of buckets.

Calculate optimal number of buckets. The number of buckets B should be O(k) for optimal complexity.

Parse a `--key value` pair from the argument list, returning the value parsed as `usize`. Returns `None` if the key is absent or unparseable.

Compute polynomial power using repeated squaring. Computes p(x)^n efficiently.

Compute power spectrogram. # Arguments `spectrogram` - Complex spectrogram # Returns Power of each bin `(|X[k]|²)`.

Push multiple samples.

Read the last `n` samples into a slice (in order). Returns the number of samples actually read.

Real-valued twiddle pre-processing for C2R. Inverse of [`real_twiddle_post`]. Given the R2C half-spectrum `X[0..=N/2]`, produces packed-complex data

Reset to initial (empty) state, keeping the same window size.

Rewrite an expression, replacing shared subexpressions with `Temp` refs.

Compute sparse FFT with automatic sparsity detection (adaptive k). Instead of requiring the caller to specify `k` (expected number of non-zero freque

Compute sparse FFT with automatic sparsity detection and custom energy ratio. This is the configurable version of [`sparse_fft_auto`]. The `energy_ra

Analyze bucket occupancy across all FFAST stages and suggest an adjusted sparsity estimate. Computes the fraction of non-zero buckets (those whose en

Get the detection threshold.

Radix-2 DIT twiddle butterfly, forward. Layout: twiddles[(j-1)*stride + s] for j=1, i.e., twiddles[s].

Scalar fallback: `data[i] *= twiddles[i]` with no SIMD.

Apply twiddle factors to `data` in-place: `data[i] *= twiddles[i]`. Dispatches to the best available SIMD backend at runtime: AVX2+FMA → SSE2 → NEON

Map a `u64` PRNG output to `f64` in `[-1.0, 1.0]` without precision-loss casts. Constructs an IEEE 754 f64 in `[1.0, 2.0)` from the top 52 mantissa b

Import wisdom from string. Accepts both the legacy `(oxifft-wisdom-1.0 …)` header (format version 0) and the current `(oxifft-wisdom …)` header with

Get access to the global wisdom cache for reading.

Advance the buffer by `n` samples (discarding oldest).

Apply filter to signal in frequency domain. # Arguments `signal_fft` - Signal in frequency domain # Returns Filtered signal in frequency domain.

Apply 1D FFT along a specific dimension.

Apply twiddle factors to the matrix after column FFTs. For element (i, j) in the N1×N2 matrix, multiply by W_N^(i*j) where W_N = e^(sign * 2πi / N).