Use pointers instead of array indices in MovePicker

[stockfish] / src / tt.cpp

diff --git a/src/tt.cpp b/src/tt.cpp
index 49dae31d70ffb5b634709e339c7815d8d5e5184d..e140a0ba10ec5fa2183b506b68750a4dcbc39289 100644 (file)
--- a/src/tt.cpp
+++ b/src/tt.cpp
@@ -25,14 +25,13 @@
 #include <cassert>
 #include <cmath>
 #include <cstring>
+#include <xmmintrin.h>
 
 #include "movegen.h"
 #include "tt.h"
 
-
-/// This is the number of TTEntry slots for each position
-static const int ClusterSize = 4;
-
+// The main transposition table
+TranspositionTable TT;
 
 ////
 //// Functions
@@ -62,14 +61,14 @@ void TranspositionTable::set_size(unsigned mbSize) {
 
   // We store a cluster of ClusterSize number of TTEntry for each position
   // and newSize is the maximum number of storable positions.
-  while ((2 * newSize) * ClusterSize * (sizeof(TTEntry)) <= (mbSize << 20))
+  while ((2 * newSize) * sizeof(TTCluster) <= (mbSize << 20))
       newSize *= 2;
 
   if (newSize != size)
   {
       size = newSize;
       delete [] entries;
-      entries = new TTEntry[size * ClusterSize];
+      entries = new TTCluster[size];
       if (!entries)
       {
           std::cerr << "Failed to allocate " << mbSize
@@ -88,7 +87,17 @@ void TranspositionTable::set_size(unsigned mbSize) {
 
 void TranspositionTable::clear() {
 
-  memset(entries, 0, size * ClusterSize * sizeof(TTEntry));
+  memset(entries, 0, size * sizeof(TTCluster));
+}
+
+
+/// TranspositionTable::first_entry returns a pointer to the first
+/// entry of a cluster given a position. The low 32 bits of the key
+/// are used to get the index in the table.
+
+inline TTEntry* TranspositionTable::first_entry(const Key posKey) const {
+
+  return entries[uint32_t(posKey) & (size - 1)].data;
 }
 
 
@@ -144,7 +153,7 @@ void TranspositionTable::store(const Key posKey, Value v, ValueType t, Depth d,
 TTEntry* TranspositionTable::retrieve(const Key posKey) const {
 
   uint32_t posKey32 = posKey >> 32;
-  TTEntry *tte = first_entry(posKey);
+  TTEntry* tte = first_entry(posKey);
 
   for (int i = 0; i < ClusterSize; i++, tte++)
       if (tte->key() == posKey32)
@@ -154,15 +163,27 @@ TTEntry* TranspositionTable::retrieve(const Key posKey) const {
 }
 
 
-/// TranspositionTable::first_entry returns a pointer to the first
-/// entry of a cluster given a position. The low 32 bits of the key
-/// are used to get the index in the table.
+/// TranspositionTable::prefetch looks up the current position in the
+/// transposition table and load it in L1/L2 cache. This is a non
+/// blocking function and do not stalls the CPU waiting for data
+/// to be loaded from RAM, that can be very slow. When we will
+/// subsequently call retrieve() the TT data will be already
+/// quickly accessible in L1/L2 CPU cache.
 
-inline TTEntry* TranspositionTable::first_entry(const Key posKey) const {
+void TranspositionTable::prefetch(const Key posKey) const {
 
-  return entries + ((uint32_t(posKey) & (size - 1)) * ClusterSize);
+#if defined(__INTEL_COMPILER) || defined(__ICL)
+   // This hack prevents prefetches to be optimized away by the
+   // Intel compiler. Both MSVC and gcc seems not affected.
+   __asm__ ("");
+#endif
+
+   char const* addr = (char*)first_entry(posKey);
+  _mm_prefetch(addr, _MM_HINT_T2);
+  _mm_prefetch(addr+64, _MM_HINT_T2); // 64 bytes ahead
 }
 
+
 /// TranspositionTable::new_search() is called at the beginning of every new
 /// search. It increments the "generation" variable, which is used to
 /// distinguish transposition table entries from previous searches from