Let TT clusters (16*4=64 bytes) to hold on a singe cache line.
This avoids the need for the double prefetch.
Original patches by Lucas and Jean-Francois that has also tested
on his AMD FX:
BIG HASHTABLE
./stockfish bench 1024 1 18 > /dev/null
Before:
1437642 nps
1426519 nps
1438493 nps
After:
1474482 nps
1476375 nps
1475877 nps
SMALL HASHTABLE
./stockfish bench 128 1 18 > /dev/null
Before:
1435207 nps
1435586 nps
1433741 nps
After:
1479143 nps
1471042 nps
1472286 nps
No functional change.
# if defined(__INTEL_COMPILER) || defined(_MSC_VER)
_mm_prefetch(addr, _MM_HINT_T0);
- _mm_prefetch(addr+64, _MM_HINT_T0); // 64 bytes ahead
# else
__builtin_prefetch(addr);
- __builtin_prefetch(addr+64);
# endif
}
if (hashMask == size - ClusterSize)
return;
- hashMask = size - ClusterSize;
- delete [] table;
- table = new (std::nothrow) TTEntry[size];
-
- if (!table)
+ free(mem);
+ mem = malloc(size * sizeof(TTEntry) + (CACHE_LINE_SIZE - 1));
+ if (!mem)
{
std::cerr << "Failed to allocate " << mbSize
<< "MB for transposition table." << std::endl;
exit(EXIT_FAILURE);
}
- clear(); // Operator new is not guaranteed to initialize memory to zero
+ table = (TTEntry*)((size_t(mem) + CACHE_LINE_SIZE - 1) & ~(CACHE_LINE_SIZE - 1));
+ hashMask = size - ClusterSize;
+ clear(); // Newly allocated block of memory is not initialized
}
static const unsigned ClusterSize = 4; // A cluster is 64 Bytes
public:
- ~TranspositionTable() { delete [] table; }
+ ~TranspositionTable() { free(mem); }
void new_search() { generation++; }
TTEntry* probe(const Key key) const;
private:
uint32_t hashMask;
TTEntry* table;
+ void* mem;
uint8_t generation; // Size must be not bigger then TTEntry::generation8
};
# include <xmmintrin.h> // Intel and Microsoft header for _mm_prefetch()
# endif
+#define CACHE_LINE_SIZE 64
#if defined(_MSC_VER) || defined(__INTEL_COMPILER)
-# define CACHE_LINE_ALIGNMENT __declspec(align(64))
+# define CACHE_LINE_ALIGNMENT __declspec(align(CACHE_LINE_SIZE))
#else
-# define CACHE_LINE_ALIGNMENT __attribute__ ((aligned(64)))
+# define CACHE_LINE_ALIGNMENT __attribute__ ((aligned(CACHE_LINE_SIZE)))
#endif
#if defined(_MSC_VER)