Add support for VNNI

[stockfish] / src / misc.cpp

diff --git a/src/misc.cpp b/src/misc.cpp
index 401a65058fc8ddc058fa64223148a7b861b5600c..ab52d30bb15bd449d4d6bcdff5681c7ced055efc 100644 (file)
--- a/src/misc.cpp
+++ b/src/misc.cpp
@@ -51,6 +51,11 @@ typedef bool(*fun3_t)(HANDLE, CONST GROUP_AFFINITY*, PGROUP_AFFINITY);
 #include <sys/mman.h>
 #endif
 
+#if defined(__APPLE__) || defined(__ANDROID__) || defined(__OpenBSD__) || (defined(__GLIBCXX__) && !defined(_GLIBCXX_HAVE_ALIGNED_ALLOC) && !defined(_WIN32))
+#define POSIXALIGNEDALLOC
+#include <stdlib.h>
+#endif
+
 #include "misc.h"
 #include "thread.h"
 
@@ -214,6 +219,9 @@ const std::string compiler_info() {
 
   compiler += "\nCompilation settings include: ";
   compiler += (Is64Bit ? " 64bit" : " 32bit");
+  #if defined(USE_VNNI)
+    compiler += " VNNI";
+  #endif
   #if defined(USE_AVX512)
     compiler += " AVX512";
   #endif
@@ -318,9 +326,12 @@ void prefetch(void* addr) {
 ///
 
 void* std_aligned_alloc(size_t alignment, size_t size) {
-#if (defined(__APPLE__) && defined(_LIBCPP_HAS_C11_FEATURES)) || defined(__ANDROID__) || defined(__OpenBSD__) || (defined(__GLIBCXX__) && !defined(_GLIBCXX_HAVE_ALIGNED_ALLOC) && !defined(_WIN32))
-  return aligned_alloc(alignment, size);
-#elif (defined(_WIN32) || (defined(__APPLE__) && !defined(_LIBCPP_HAS_C11_FEATURES)))
+#if defined(POSIXALIGNEDALLOC)
+  void *pointer;
+  if(posix_memalign(&pointer, alignment, size) == 0)
+      return pointer;
+  return nullptr;
+#elif defined(_WIN32)
   return _mm_malloc(size, alignment);
 #else
   return std::aligned_alloc(alignment, size);
@@ -328,9 +339,9 @@ void* std_aligned_alloc(size_t alignment, size_t size) {
 }
 
 void std_aligned_free(void* ptr) {
-#if (defined(__APPLE__) && defined(_LIBCPP_HAS_C11_FEATURES)) || defined(__ANDROID__) || defined(__OpenBSD__) || (defined(__GLIBCXX__) && !defined(_GLIBCXX_HAVE_ALIGNED_ALLOC) && !defined(_WIN32))
+#if defined(POSIXALIGNEDALLOC)
   free(ptr);
-#elif (defined(_WIN32) || (defined(__APPLE__) && !defined(_LIBCPP_HAS_C11_FEATURES)))
+#elif defined(_WIN32)
   _mm_free(ptr);
 #else
   free(ptr);