X-Git-Url: https://git.sesse.net/?p=stockfish;a=blobdiff_plain;f=src%2Fnnue%2Flayers%2Faffine_transform.h;h=20ec2f1234befe5a67f6ab2d3c70c42fabac08c4;hp=ecc3008a6cc8a728927011ba4b6154a232e76994;hb=875183b310a8249922c2155e82cb4cecfae2097e;hpb=a6e89293df5af35931b61d86b6de3872a981c100;ds=sidebyside

diff --git a/src/nnue/layers/affine_transform.h b/src/nnue/layers/affine_transform.h
index ecc3008a..20ec2f12 100644
--- a/src/nnue/layers/affine_transform.h
+++ b/src/nnue/layers/affine_transform.h
@@ -104,8 +104,7 @@ namespace Eval::NNUE::Layers {
         __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);
         }
@@ -119,9 +118,7 @@ namespace Eval::NNUE::Layers {
             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);
@@ -134,8 +131,7 @@ namespace Eval::NNUE::Layers {
         __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);
         }
@@ -149,8 +145,7 @@ namespace Eval::NNUE::Layers {
         __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);
         }