Pack 3 TT entries in 32 bytes cluster

Idea from Ron Britvich

Code reworked by Marco Costalba and Joona Kiiski

Bench: 8095369

Resolves #3
Resolves #10

src/tt.cpp

@@ -28,20 +28,21 @@ TranspositionTable TT; // Our global transposition table
 
 /// TranspositionTable::resize() sets the size of the transposition table,
 /// measured in megabytes. Transposition table consists of a power of 2 number
-/// of clusters and each cluster consists of ClusterSize number of TTEntry.
+/// of clusters and each cluster consists of TTClusterSize number of TTEntry.
 
 void TranspositionTable::resize(uint64_t mbSize) {
 
-  assert(msb((mbSize << 20) / sizeof(TTEntry)) < 32);
+  assert(msb((mbSize * 1024 * 1024) / sizeof(TTCluster)) < 32);
 
-  uint32_t size = ClusterSize << msb((mbSize << 20) / sizeof(TTEntry[ClusterSize]));
+  uint32_t newClusterCount = 1 << msb((mbSize * 1024 * 1024) / sizeof(TTCluster));
 
-  if (hashMask == size - ClusterSize)
+  if (newClusterCount == clusterCount)
       return;
 
-  hashMask = size - ClusterSize;
+  clusterCount = newClusterCount;
+
   free(mem);
-  mem = calloc(size * sizeof(TTEntry) + CACHE_LINE_SIZE - 1, 1);
+  mem = calloc(clusterCount * sizeof(TTCluster) + CACHE_LINE_SIZE - 1, 1);
 
   if (!mem)
   {
@@ -50,7 +51,7 @@ void TranspositionTable::resize(uint64_t mbSize) {
       exit(EXIT_FAILURE);
   }
 
-  table = (TTEntry*)((uintptr_t(mem) + CACHE_LINE_SIZE - 1) & ~(CACHE_LINE_SIZE - 1));
+  table = (TTCluster*)((uintptr_t(mem) + CACHE_LINE_SIZE - 1) & ~(CACHE_LINE_SIZE - 1));
 }
 
 
@@ -60,7 +61,7 @@ void TranspositionTable::resize(uint64_t mbSize) {
 
 void TranspositionTable::clear() {
 
-  std::memset(table, 0, (hashMask + ClusterSize) * sizeof(TTEntry));
+  std::memset(table, 0, clusterCount * sizeof(TTCluster));
 }
 
 
@@ -71,12 +72,12 @@ void TranspositionTable::clear() {
 const TTEntry* TranspositionTable::probe(const Key key) const {
 
   TTEntry* tte = first_entry(key);
-  uint32_t key32 = key >> 32;
+  uint16_t key16 = key >> 48;
 
-  for (unsigned i = 0; i < ClusterSize; ++i, ++tte)
-      if (tte->key32 == key32)
+  for (unsigned i = 0; i < TTClusterSize; ++i, ++tte)
+      if (tte->key16 == key16)
       {
-          tte->generation8 = generation; // Refresh
+          tte->genBound8 = generation | tte->bound(); // Refresh
           return tte;
       }
 
@@ -95,13 +96,13 @@ const TTEntry* TranspositionTable::probe(const Key key) const {
 void TranspositionTable::store(const Key key, Value v, Bound b, Depth d, Move m, Value statV) {
 
   TTEntry *tte, *replace;
-  uint32_t key32 = key >> 32; // Use the high 32 bits as key inside the cluster
+  uint16_t key16 = key >> 48; // Use the high 16 bits as key inside the cluster
 
   tte = replace = first_entry(key);
 
-  for (unsigned i = 0; i < ClusterSize; ++i, ++tte)
+  for (unsigned i = 0; i < TTClusterSize; ++i, ++tte)
   {
-      if (!tte->key32 || tte->key32 == key32) // Empty or overwrite old
+      if (!tte->key16 || tte->key16 == key16) // Empty or overwrite old
       {
           if (!m)
               m = tte->move(); // Preserve any existing ttMove
@@ -111,11 +112,11 @@ void TranspositionTable::store(const Key key, Value v, Bound b, Depth d, Move m,
       }
 
       // Implement replace strategy
-      if (  (    tte->generation8 == generation || tte->bound() == BOUND_EXACT)
-          - (replace->generation8 == generation)
-          - (tte->depth16 < replace->depth16) < 0)
+      if (  ((    tte->genBound8 & 0xFC) == generation || tte->bound() == BOUND_EXACT)
+          - ((replace->genBound8 & 0xFC) == generation)
+          - (tte->depth8 < replace->depth8) < 0)
           replace = tte;
   }
 
-  replace->save(key32, v, b, d, m, generation, statV);
+  replace->save(key16, v, b, d, m, generation, statV);
 }

src/tt.h

@@ -23,58 +23,69 @@
 #include "misc.h"
 #include "types.h"
 
-/// The TTEntry is the 14 bytes transposition table entry, defined as below:
+/// The TTEntry is the 10 bytes transposition table entry, defined as below:
 ///
-/// key        32 bit
+/// key        16 bit
 /// move       16 bit
-/// bound type  8 bit
-/// generation  8 bit
 /// value      16 bit
-/// depth      16 bit
 /// eval value 16 bit
+/// generation  6 bit
+/// bound type  2 bit
+/// depth       8 bit
 
 struct TTEntry {
 
   Move  move()  const      { return (Move )move16; }
-  Bound bound() const      { return (Bound)bound8; }
   Value value() const      { return (Value)value16; }
-  Depth depth() const      { return (Depth)depth16; }
   Value eval_value() const { return (Value)evalValue; }
+  Depth depth() const      { return (Depth)(depth8) + DEPTH_NONE; }
+  Bound bound() const      { return (Bound)(genBound8 & 0x3); }
 
 private:
   friend class TranspositionTable;
 
-  void save(uint32_t k, Value v, Bound b, Depth d, Move m, uint8_t g, Value ev) {
+  void save(uint16_t k, Value v, Bound b, Depth d, Move m, uint8_t g, Value ev) {
 
-    key32       = (uint32_t)k;
-    move16      = (uint16_t)m;
-    bound8      = (uint8_t)b;
-    generation8 = (uint8_t)g;
-    value16     = (int16_t)v;
-    depth16     = (int16_t)d;
-    evalValue   = (int16_t)ev;
+    key16     = (uint16_t)k;
+    move16    = (uint16_t)m;
+    value16   = (int16_t)v;
+    evalValue = (int16_t)ev;
+    depth8    = (uint8_t)(d - DEPTH_NONE);
+    genBound8 = g | (uint8_t)b;
   }
 
-  uint32_t key32;
+  uint16_t key16;
   uint16_t move16;
-  uint8_t bound8, generation8;
-  int16_t value16, depth16, evalValue;
+  int16_t  value16;
+  int16_t  evalValue;
+  uint8_t  genBound8;
+  uint8_t  depth8;
 };
 
+/// TTCluster is a 32 bytes cluster of TT entries consisting of:
+///
+/// 3 x TTEntry (3 x 10 bytes)
+/// padding     (2 bytes)
+
+const unsigned TTClusterSize = 3;
+
+struct TTCluster {
+
+  TTEntry entry[TTClusterSize];
+  char padding[2];
+};
 
 /// A TranspositionTable consists of a power of 2 number of clusters and each
-/// cluster consists of ClusterSize number of TTEntry. Each non-empty entry
+/// cluster consists of TTClusterSize number of TTEntry. Each non-empty entry
 /// contains information of exactly one position. The size of a cluster should
 /// not be bigger than a cache line size. In case it is less, it should be padded
 /// to guarantee always aligned accesses.
 
 class TranspositionTable {
 
-  static const unsigned ClusterSize = 4;
-
 public:
  ~TranspositionTable() { free(mem); }
-  void new_search() { ++generation; }
+  void new_search() { generation += 4; } // Lower 2 bits are used by Bound
 
   const TTEntry* probe(const Key key) const;
   TTEntry* first_entry(const Key key) const;
@@ -83,10 +94,10 @@ public:
   void store(const Key key, Value v, Bound type, Depth d, Move m, Value statV);
 
 private:
-  uint32_t hashMask;
-  TTEntry* table;
+  uint32_t clusterCount;
+  TTCluster* table;
   void* mem;
-  uint8_t generation; // Size must be not bigger than TTEntry::generation8
+  uint8_t generation; // Size must be not bigger than TTEntry::genBound8
 };
 
 extern TranspositionTable TT;
@@ -94,11 +105,11 @@ extern TranspositionTable TT;
 
 /// TranspositionTable::first_entry() returns a pointer to the first entry of
 /// a cluster given a position. The lowest order bits of the key are used to
-/// get the index of the cluster.
+/// get the index of the cluster inside the table.
 
 inline TTEntry* TranspositionTable::first_entry(const Key key) const {
 
-  return table + ((uint32_t)key & hashMask);
+  return &table[(uint32_t)key & (clusterCount - 1)].entry[0];
 }
 
 #endif // #ifndef TT_H_INCLUDED

src/types.h

@@ -218,7 +218,7 @@ enum Depth {
   DEPTH_QS_NO_CHECKS  = -1 * ONE_PLY,
   DEPTH_QS_RECAPTURES = -5 * ONE_PLY,
 
-  DEPTH_NONE = -127 * ONE_PLY
+  DEPTH_NONE = -6 * ONE_PLY
 };
 
 enum Square {