#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 = 5;
-
+// The main transposition table
+TranspositionTable TT;
////
//// Functions
// 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
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;
}
{
if (!tte->key() || tte->key() == posKey32) // empty or overwrite old
{
- // Do not overwrite when new type is VALUE_TYPE_EVAL
- if (tte->key() && t == VALUE_TYPE_EVAL)
+ // Do not overwrite when new type is VALUE_TYPE_EV_LO
+ if (tte->key() && t == VALUE_TYPE_EV_LO)
return;
if (m == MOVE_NONE)
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)
}
-/// 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