Module ops 

Block-rate audio ops at f32. See crate::ops64 for the f64 mirror.

Two implementations per op when wide-backend is enabled: a scalar fallback (the *_scalar variant, also the body when the feature is off) and a vector path that processes f32x8 chunks. The scalar variants are kept pub so the Criterion benches can compare pre-/post-vectorization on exactly the same operation.

Functions

abs_max_block

max(buf[i].abs()). Returns 0.0 for an empty slice; returns f32::NAN on first NaN (so meters can flag a runaway plugin instead of silently reporting in-range peaks).

copy_block

out[i] = src[i]. Equivalent to copy_from_slice but exposed in the same surface as the other ops for code that wires up its inner loops from this module.

gain_block

buf[i] *= gain for every element. Vectorized when wide-backend is on, otherwise identical to the scalar path.

gain_block_scalar

buf[i] *= gain for every element. Scalar.

mac_block

out[i] += src[i] * scale. Vectorized when wide-backend is on.

mac_block_scalar

Multiply-accumulate: out[i] += src[i] * scale. Scalar.

mix_block

out[i] = a[i] * gain_a + b[i] * gain_b. Vectorized when wide-backend is on.

mix_block_scalar

out[i] = a[i] * gain_a + b[i] * gain_b. Scalar.

mul_block

out[i] = a[i] * b[i]. Vectorized when wide-backend is on.

mul_block_scalar

out[i] = a[i] * b[i] (length: min(out, a, b)). Scalar.

scale_block

out[i] = src[i] * scale. Vectorized when wide-backend is on.

scale_block_scalar

out[i] = src[i] * scale (length: min(out, src)). Scalar. The non-aliased counterpart to gain_block - fills the “copy then gain” hole in the API so a single line covers what would otherwise need two calls and a discarded intermediate.

zero_block

out[i] = 0.0 for all i.