Function _mm256_madd_epi16 

pub fn _mm256_madd_epi16(a: __m256i, b: __m256i) -> __m256i

Available on x86 and target feature avx2 only.

Multiplies packed signed 16-bit integers in a and b, producing intermediate signed 32-bit integers. Horizontally add adjacent pairs of intermediate 32-bit integers.

Intel’s documentation