__m256i sum = _mm256_setzero_si256();
const auto row = reinterpret_cast<const __m256i*>(&weights_[offset]);
for (IndexType j = 0; j < kNumChunks; ++j) {
__m256i sum = _mm256_setzero_si256();
const auto row = reinterpret_cast<const __m256i*>(&weights_[offset]);
for (IndexType j = 0; j < kNumChunks; ++j) {
__m256i product = _mm256_maddubs_epi16(_mm256_loadA_si256(&input_vector[j]), _mm256_load_si256(&row[j]));
product = _mm256_madd_epi16(product, kOnes);
sum = _mm256_add_epi32(sum, product);
__m256i product = _mm256_maddubs_epi16(_mm256_loadA_si256(&input_vector[j]), _mm256_load_si256(&row[j]));
product = _mm256_madd_epi16(product, kOnes);
sum = _mm256_add_epi32(sum, product);
}
__m128i sum128 = _mm_add_epi32(_mm256_castsi256_si128(sum), _mm256_extracti128_si256(sum, 1));
sum128 = _mm_add_epi32(sum128, _mm_shuffle_epi32(sum128, _MM_PERM_BADC));
}
__m128i sum128 = _mm_add_epi32(_mm256_castsi256_si128(sum), _mm256_extracti128_si256(sum, 1));
sum128 = _mm_add_epi32(sum128, _mm_shuffle_epi32(sum128, _MM_PERM_BADC));