pub fn inner_product_f16<Idx: Copy + Ord>(
x_idx: &[Idx],
x_val: &[Half],
y_idx: &[Idx],
y_val: &[Half],
) -> Result<f32, LengthMismatch>Expand description
Σ x·y over matching indices for f16 operands; a disjoint-range fast-out skips widening
when the operands cannot intersect.