Provide vectorized NNUE code for SSE2 and MMX targets

[stockfish] / src / nnue / nnue_common.h

diff --git a/src/nnue/nnue_common.h b/src/nnue/nnue_common.h
index e7ce84f7b9f420eaaa3dbadb4302eeb242315c4f..cb1251c58aa37923a23e87fb4e6dbaa668d24b8d 100644 (file)
--- a/src/nnue/nnue_common.h
+++ b/src/nnue/nnue_common.h
@@ -33,6 +33,9 @@
 #elif defined(USE_SSE2)
 #include <emmintrin.h>
 
+#elif defined(USE_MMX)
+#include <mmintrin.h>
+
 #elif defined(USE_NEON)
 #include <arm_neon.h>
 #endif
@@ -52,9 +55,11 @@
 
 #if defined(USE_AVX512)
 #if defined(__GNUC__ ) && (__GNUC__ < 9)
-#define _mm512_loadA_si512  _mm512_loadu_si512
+#define _mm512_loadA_si512   _mm512_loadu_si512
+#define _mm512_storeA_si512  _mm512_storeu_si512
 #else
-#define _mm512_loadA_si512  _mm512_load_si512
+#define _mm512_loadA_si512   _mm512_load_si512
+#define _mm512_storeA_si512  _mm512_store_si512
 #endif
 #endif
 
@@ -77,6 +82,9 @@ namespace Eval::NNUE {
   #elif defined(USE_SSE2)
   constexpr std::size_t kSimdWidth = 16;
 
+  #elif defined(USE_MMX)
+  constexpr std::size_t kSimdWidth = 8;
+
   #elif defined(USE_NEON)
   constexpr std::size_t kSimdWidth = 16;
   #endif