prod_axis_f32

void prod_axis_f32(const float * in, float * out, int outer, int reduce_dim, int inner)

source edit

Reduces a single axis by multiplicative product into a packed (outer, inner) output.

Identity element is 1; an empty reduce dimension yields 1 per cell. Uses the generic axis_reduce template — there is no vDSP equivalent. Note that floating-point products are not associative, so the result is deterministic only for a fixed loop order (the kernel iterates r from 0 to reduce_dim - 1).

Math

$\text{out}[o, i] = \prod_{r=0}^{\text{reduce\_dim}-1} \text{in}[o, r, i].$

Parameters

inconst float*

Source buffer of shape (outer, reduce_dim, inner).

outfloat*

Destination buffer of shape (outer, inner).

outer, reduce_dim, innerstd::size_t

Layout extents; see sum_axis_f32.

Notes

Overflow and underflow are unchecked. Callers that require log-sum-exp style stability must compute the product in log space themselves.