__m512i sum = _mm512_setzero_si512();
const auto row = reinterpret_cast<const __m512i*>(&weights_[offset]);
for (IndexType j = 0; j < kNumChunks; ++j) {
- __m512i product = _mm512_maddubs_epi16(
- _mm512_load_si512(&input_vector[j]), _mm512_load_si512(&row[j]));
+ __m512i product = _mm512_maddubs_epi16(_mm512_loadA_si512(&input_vector[j]), _mm512_load_si512(&row[j]));
product = _mm512_madd_epi16(product, kOnes);
sum = _mm512_add_epi32(sum, product);
}
const auto iv_256 = reinterpret_cast<const __m256i*>(input);
const auto row_256 = reinterpret_cast<const __m256i*>(&weights_[offset]);
int j = kNumChunks * 2;
-
- __m256i sum256 = _mm256_maddubs_epi16(
- _mm256_load_si256(&iv_256[j]), _mm256_load_si256(&row_256[j]));
+ __m256i sum256 = _mm256_maddubs_epi16(_mm256_loadA_si256(&iv_256[j]), _mm256_load_si256(&row_256[j]));
sum256 = _mm256_madd_epi16(sum256, _mm256_set1_epi16(1));
sum256 = _mm256_hadd_epi32(sum256, sum256);
sum256 = _mm256_hadd_epi32(sum256, sum256);
__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_load_si256(&input_vector[j]), _mm256_load_si256(&row[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);
}
__m128i sum = _mm_cvtsi32_si128(biases_[i]);
const auto row = reinterpret_cast<const __m128i*>(&weights_[offset]);
for (IndexType j = 0; j < kNumChunks; ++j) {
- __m128i product = _mm_maddubs_epi16(
- _mm_load_si128(&input_vector[j]), _mm_load_si128(&row[j]));
+ __m128i product = _mm_maddubs_epi16(_mm_load_si128(&input_vector[j]), _mm_load_si128(&row[j]));
product = _mm_madd_epi16(product, kOnes);
sum = _mm_add_epi32(sum, product);
}