// Make and search the move
StateInfo st;
pos.do_move(move, st, dcCandidates);
+ TT.prefetch(pos.get_key());
if (moveCount == 1) // The first move in list is the PV
value = -search_pv(pos, ss, -beta, -alpha, newDepth, ply+1, threadID);
StateInfo st;
pos.do_null_move(st);
+ TT.prefetch(pos.get_key());
+
int R = (depth >= 5 * OnePly ? 4 : 3); // Null move dynamic reduction
Value nullValue = -search(pos, ss, -(beta-1), depth-R*OnePly, ply+1, false, threadID);
// Make and search the move
StateInfo st;
pos.do_move(move, st, dcCandidates);
+ TT.prefetch(pos.get_key());
// Try to reduce non-pv search depth by one ply if move seems not problematic,
// if the move fails high will be re-searched at full depth.
// Make and search the move.
StateInfo st;
pos.do_move(move, st, dcCandidates);
+ TT.prefetch(pos.get_key());
Value value = -qsearch(pos, ss, -beta, -alpha, depth-OnePly, ply+1, threadID);
pos.undo_move(move);
#include <cassert>
#include <cmath>
#include <cstring>
+#include <xmmintrin.h>
#include "movegen.h"
#include "tt.h"
return NULL;
}
+/// 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.
+
+void TranspositionTable::prefetch(const Key posKey) const {
+
+ _mm_prefetch((char*)first_entry(posKey), _MM_HINT_T0);
+}
/// TranspositionTable::first_entry returns a pointer to the first
/// entry of a cluster given a position. The low 32 bits of the key
void clear();
void store(const Key posKey, Value v, ValueType type, Depth d, Move m);
TTEntry* retrieve(const Key posKey) const;
+ void prefetch(const Key posKey) const;
void new_search();
void insert_pv(const Position& pos, Move pv[]);
void extract_pv(const Position& pos, Move pv[]);