2008-08-31 23:59:13 -06:00
|
|
|
/*
|
2008-10-19 10:56:28 -06:00
|
|
|
Stockfish, a UCI chess playing engine derived from Glaurung 2.1
|
2021-01-08 09:04:23 -07:00
|
|
|
Copyright (C) 2004-2021 The Stockfish developers (see AUTHORS file)
|
2008-08-31 23:59:13 -06:00
|
|
|
|
2008-10-19 10:56:28 -06:00
|
|
|
Stockfish is free software: you can redistribute it and/or modify
|
2008-08-31 23:59:13 -06:00
|
|
|
it under the terms of the GNU General Public License as published by
|
|
|
|
the Free Software Foundation, either version 3 of the License, or
|
|
|
|
(at your option) any later version.
|
2008-09-06 04:30:07 -06:00
|
|
|
|
2008-10-19 10:56:28 -06:00
|
|
|
Stockfish is distributed in the hope that it will be useful,
|
2008-08-31 23:59:13 -06:00
|
|
|
but WITHOUT ANY WARRANTY; without even the implied warranty of
|
|
|
|
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
|
|
|
|
GNU General Public License for more details.
|
2008-09-06 04:30:07 -06:00
|
|
|
|
2008-08-31 23:59:13 -06:00
|
|
|
You should have received a copy of the GNU General Public License
|
|
|
|
along with this program. If not, see <http://www.gnu.org/licenses/>.
|
|
|
|
*/
|
|
|
|
|
2013-07-23 07:31:57 -06:00
|
|
|
#ifndef TT_H_INCLUDED
|
2008-08-31 23:59:13 -06:00
|
|
|
#define TT_H_INCLUDED
|
|
|
|
|
2011-12-04 02:53:40 -07:00
|
|
|
#include "misc.h"
|
2011-02-23 13:27:22 -07:00
|
|
|
#include "types.h"
|
2011-01-07 02:34:16 -07:00
|
|
|
|
2021-02-26 02:02:13 -07:00
|
|
|
namespace Stockfish {
|
|
|
|
|
2015-01-09 04:35:44 -07:00
|
|
|
/// TTEntry struct is the 10 bytes transposition table entry, defined as below:
|
2009-03-30 01:29:02 -06:00
|
|
|
///
|
2014-06-28 12:05:58 -06:00
|
|
|
/// key 16 bit
|
Allow TT entries with key16==0 to be fetched
Fix the issue where a TT entry with key16==0 would always be reported
as a miss. Instead, we'll use depth8 to detect whether the TT entry is
occupied. In order to do that, we'll change DEPTH_OFFSET to -7
(depth8==0) to distinguish between an unoccupied entry and the
otherwise lowest possible depth, i.e., DEPTH_NONE (depth8==1).
To prevent a performance regression, we'll reorder the TT entry fields
by the access order of TranspositionTable::probe(). Memory in general
works fastest when accessed in sequential order. We'll also match the
store order in TTEntry::save() with the entry field order, and
re-order the 'if-or' expressions in TTEntry::save() from the cheapest
to the most expensive.
Finally, as we now have a proper TT entry occupancy test, we'll fix a
minor corner case with hashfull reporting. To reproduce:
- Use a big hash
- Either:
a. Start 31 very quick searches (this wraparounds generation to 0); or
b. Force generation of the first search to 0.
- go depth infinite
Before the fix, hashfull would incorrectly report nearly full hash
immediately after the search start, since
TranspositionTable::hashfull() used to consider only the entry
generation and not whether the entry was actually occupied.
STC:
LLR: 2.95 (-2.94,2.94) {-0.25,1.25}
Total: 36848 W: 4091 L: 3898 D: 28859
Ptnml(0-2): 158, 2996, 11972, 3091, 207
https://tests.stockfishchess.org/tests/view/5f3f98d5dc02a01a0c2881f7
LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 32280 W: 1828 L: 1653 D: 28799
Ptnml(0-2): 34, 1428, 13051, 1583, 44
https://tests.stockfishchess.org/tests/view/5f3fe77a87a5c3c63d8f5332
closes https://github.com/official-stockfish/Stockfish/pull/3048
Bench: 3760677
2020-08-21 03:12:39 -06:00
|
|
|
/// depth 8 bit
|
2019-01-09 07:05:28 -07:00
|
|
|
/// generation 5 bit
|
2019-01-01 06:13:08 -07:00
|
|
|
/// pv node 1 bit
|
2014-06-28 12:05:58 -06:00
|
|
|
/// bound type 2 bit
|
Allow TT entries with key16==0 to be fetched
Fix the issue where a TT entry with key16==0 would always be reported
as a miss. Instead, we'll use depth8 to detect whether the TT entry is
occupied. In order to do that, we'll change DEPTH_OFFSET to -7
(depth8==0) to distinguish between an unoccupied entry and the
otherwise lowest possible depth, i.e., DEPTH_NONE (depth8==1).
To prevent a performance regression, we'll reorder the TT entry fields
by the access order of TranspositionTable::probe(). Memory in general
works fastest when accessed in sequential order. We'll also match the
store order in TTEntry::save() with the entry field order, and
re-order the 'if-or' expressions in TTEntry::save() from the cheapest
to the most expensive.
Finally, as we now have a proper TT entry occupancy test, we'll fix a
minor corner case with hashfull reporting. To reproduce:
- Use a big hash
- Either:
a. Start 31 very quick searches (this wraparounds generation to 0); or
b. Force generation of the first search to 0.
- go depth infinite
Before the fix, hashfull would incorrectly report nearly full hash
immediately after the search start, since
TranspositionTable::hashfull() used to consider only the entry
generation and not whether the entry was actually occupied.
STC:
LLR: 2.95 (-2.94,2.94) {-0.25,1.25}
Total: 36848 W: 4091 L: 3898 D: 28859
Ptnml(0-2): 158, 2996, 11972, 3091, 207
https://tests.stockfishchess.org/tests/view/5f3f98d5dc02a01a0c2881f7
LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 32280 W: 1828 L: 1653 D: 28799
Ptnml(0-2): 34, 1428, 13051, 1583, 44
https://tests.stockfishchess.org/tests/view/5f3fe77a87a5c3c63d8f5332
closes https://github.com/official-stockfish/Stockfish/pull/3048
Bench: 3760677
2020-08-21 03:12:39 -06:00
|
|
|
/// move 16 bit
|
|
|
|
/// value 16 bit
|
|
|
|
/// eval value 16 bit
|
2008-08-31 23:59:13 -06:00
|
|
|
|
2013-06-29 02:23:43 -06:00
|
|
|
struct TTEntry {
|
2008-08-31 23:59:13 -06:00
|
|
|
|
2015-01-09 04:35:44 -07:00
|
|
|
Move move() const { return (Move )move16; }
|
|
|
|
Value value() const { return (Value)value16; }
|
|
|
|
Value eval() const { return (Value)eval16; }
|
2019-09-28 14:27:23 -06:00
|
|
|
Depth depth() const { return (Depth)depth8 + DEPTH_OFFSET; }
|
2020-03-14 10:04:50 -06:00
|
|
|
bool is_pv() const { return (bool)(genBound8 & 0x4); }
|
2015-01-09 04:35:44 -07:00
|
|
|
Bound bound() const { return (Bound)(genBound8 & 0x3); }
|
2019-01-01 06:13:08 -07:00
|
|
|
void save(Key k, Value v, bool pv, Bound b, Depth d, Move m, Value ev);
|
2014-12-13 00:16:35 -07:00
|
|
|
|
2014-05-24 15:53:01 -06:00
|
|
|
private:
|
|
|
|
friend class TranspositionTable;
|
|
|
|
|
2014-06-28 12:05:58 -06:00
|
|
|
uint16_t key16;
|
Allow TT entries with key16==0 to be fetched
Fix the issue where a TT entry with key16==0 would always be reported
as a miss. Instead, we'll use depth8 to detect whether the TT entry is
occupied. In order to do that, we'll change DEPTH_OFFSET to -7
(depth8==0) to distinguish between an unoccupied entry and the
otherwise lowest possible depth, i.e., DEPTH_NONE (depth8==1).
To prevent a performance regression, we'll reorder the TT entry fields
by the access order of TranspositionTable::probe(). Memory in general
works fastest when accessed in sequential order. We'll also match the
store order in TTEntry::save() with the entry field order, and
re-order the 'if-or' expressions in TTEntry::save() from the cheapest
to the most expensive.
Finally, as we now have a proper TT entry occupancy test, we'll fix a
minor corner case with hashfull reporting. To reproduce:
- Use a big hash
- Either:
a. Start 31 very quick searches (this wraparounds generation to 0); or
b. Force generation of the first search to 0.
- go depth infinite
Before the fix, hashfull would incorrectly report nearly full hash
immediately after the search start, since
TranspositionTable::hashfull() used to consider only the entry
generation and not whether the entry was actually occupied.
STC:
LLR: 2.95 (-2.94,2.94) {-0.25,1.25}
Total: 36848 W: 4091 L: 3898 D: 28859
Ptnml(0-2): 158, 2996, 11972, 3091, 207
https://tests.stockfishchess.org/tests/view/5f3f98d5dc02a01a0c2881f7
LTC:
LLR: 2.95 (-2.94,2.94) {0.25,1.25}
Total: 32280 W: 1828 L: 1653 D: 28799
Ptnml(0-2): 34, 1428, 13051, 1583, 44
https://tests.stockfishchess.org/tests/view/5f3fe77a87a5c3c63d8f5332
closes https://github.com/official-stockfish/Stockfish/pull/3048
Bench: 3760677
2020-08-21 03:12:39 -06:00
|
|
|
uint8_t depth8;
|
|
|
|
uint8_t genBound8;
|
2011-01-08 04:34:29 -07:00
|
|
|
uint16_t move16;
|
2014-06-28 12:05:58 -06:00
|
|
|
int16_t value16;
|
2015-01-09 04:35:44 -07:00
|
|
|
int16_t eval16;
|
2008-08-31 23:59:13 -06:00
|
|
|
};
|
|
|
|
|
2009-08-11 01:30:19 -06:00
|
|
|
|
2019-12-21 12:41:42 -07:00
|
|
|
/// A TranspositionTable is an array of Cluster, of size clusterCount. Each
|
|
|
|
/// cluster consists of ClusterSize number of TTEntry. Each non-empty TTEntry
|
|
|
|
/// contains information on exactly one position. The size of a Cluster should
|
2020-06-24 14:19:58 -06:00
|
|
|
/// divide the size of a cache line for best performance, as the cacheline is
|
|
|
|
/// prefetched when possible.
|
2008-08-31 23:59:13 -06:00
|
|
|
|
|
|
|
class TranspositionTable {
|
|
|
|
|
2018-03-18 16:38:58 -06:00
|
|
|
static constexpr int ClusterSize = 3;
|
2015-01-09 04:35:44 -07:00
|
|
|
|
2015-01-17 01:04:25 -07:00
|
|
|
struct Cluster {
|
2015-01-17 14:15:15 -07:00
|
|
|
TTEntry entry[ClusterSize];
|
2019-12-21 12:41:42 -07:00
|
|
|
char padding[2]; // Pad to 32 bytes
|
2015-01-09 04:35:44 -07:00
|
|
|
};
|
|
|
|
|
2019-12-21 12:41:42 -07:00
|
|
|
static_assert(sizeof(Cluster) == 32, "Unexpected Cluster size");
|
2015-01-21 05:18:19 -07:00
|
|
|
|
2021-02-11 14:29:28 -07:00
|
|
|
// Constants used to refresh the hash table periodically
|
|
|
|
static constexpr unsigned GENERATION_BITS = 3; // nb of bits reserved for other things
|
|
|
|
static constexpr int GENERATION_DELTA = (1 << GENERATION_BITS); // increment for generation field
|
|
|
|
static constexpr int GENERATION_CYCLE = 255 + (1 << GENERATION_BITS); // cycle length
|
|
|
|
static constexpr int GENERATION_MASK = (0xFF << GENERATION_BITS) & 0xFF; // mask to pull out generation number
|
|
|
|
|
2008-08-31 23:59:13 -06:00
|
|
|
public:
|
Add large page support for NNUE weights and simplify TT mem management
Use TT memory functions to allocate memory for the NNUE weights. This
should provide a small speed-up on systems where large pages are not
automatically used, including Windows and some Linux distributions.
Further, since we now have a wrapper for std::aligned_alloc(), we can
simplify the TT memory management a bit:
- We no longer need to store separate pointers to the hash table and
its underlying memory allocation.
- We also get to merge the Linux-specific and default implementations
of aligned_ttmem_alloc().
Finally, we'll enable the VirtualAlloc code path with large page
support also for Win32.
STC: https://tests.stockfishchess.org/tests/view/5f66595823a84a47b9036fba
LLR: 2.94 (-2.94,2.94) {-0.25,1.25}
Total: 14896 W: 1854 L: 1686 D: 11356
Ptnml(0-2): 65, 1224, 4742, 1312, 105
closes https://github.com/official-stockfish/Stockfish/pull/3081
No functional change.
2020-08-30 10:41:30 -06:00
|
|
|
~TranspositionTable() { aligned_large_pages_free(table); }
|
2021-02-11 14:29:28 -07:00
|
|
|
void new_search() { generation8 += GENERATION_DELTA; } // Lower bits are used for other things
|
2014-12-13 00:16:35 -07:00
|
|
|
TTEntry* probe(const Key key, bool& found) const;
|
2015-01-25 00:57:51 -07:00
|
|
|
int hashfull() const;
|
2014-07-01 04:13:20 -06:00
|
|
|
void resize(size_t mbSize);
|
2008-08-31 23:59:13 -06:00
|
|
|
void clear();
|
|
|
|
|
2015-01-09 04:35:44 -07:00
|
|
|
TTEntry* first_entry(const Key key) const {
|
Use 128 bit multiply for TT index
Remove super cluster stuff from TT and just use a 128 bit multiply.
STC https://tests.stockfishchess.org/tests/view/5ee719b3aae8aec816ab7548
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 12736 W: 2502 L: 2333 D: 7901
Ptnml(0-2): 191, 1452, 2944, 1559, 222
LTC https://tests.stockfishchess.org/tests/view/5ee732d1aae8aec816ab7556
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 27584 W: 3431 L: 3350 D: 20803
Ptnml(0-2): 173, 2500, 8400, 2511, 208
Scheme back to being derived from https://lemire.me/blog/2016/06/27/a-fast-alternative-to-the-modulo-reduction/
Also the default optimized version of the index calculation now uses fewer instructions.
https://godbolt.org/z/Tktxbv
Might benefit from mulx (requires -mbmi2)
closes https://github.com/official-stockfish/Stockfish/pull/2744
bench: 4320954
2020-06-15 00:35:07 -06:00
|
|
|
return &table[mul_hi64(key, clusterCount)].entry[0];
|
2015-01-09 04:35:44 -07:00
|
|
|
}
|
|
|
|
|
2020-09-10 05:54:47 -06:00
|
|
|
static bool enable_transposition_table;
|
|
|
|
|
2008-08-31 23:59:13 -06:00
|
|
|
private:
|
2018-07-03 16:58:16 -06:00
|
|
|
friend struct TTEntry;
|
|
|
|
|
Use 128 bit multiply for TT index
Remove super cluster stuff from TT and just use a 128 bit multiply.
STC https://tests.stockfishchess.org/tests/view/5ee719b3aae8aec816ab7548
LLR: 2.94 (-2.94,2.94) {-1.50,0.50}
Total: 12736 W: 2502 L: 2333 D: 7901
Ptnml(0-2): 191, 1452, 2944, 1559, 222
LTC https://tests.stockfishchess.org/tests/view/5ee732d1aae8aec816ab7556
LLR: 2.93 (-2.94,2.94) {-1.50,0.50}
Total: 27584 W: 3431 L: 3350 D: 20803
Ptnml(0-2): 173, 2500, 8400, 2511, 208
Scheme back to being derived from https://lemire.me/blog/2016/06/27/a-fast-alternative-to-the-modulo-reduction/
Also the default optimized version of the index calculation now uses fewer instructions.
https://godbolt.org/z/Tktxbv
Might benefit from mulx (requires -mbmi2)
closes https://github.com/official-stockfish/Stockfish/pull/2744
bench: 4320954
2020-06-15 00:35:07 -06:00
|
|
|
size_t clusterCount;
|
2015-01-17 01:04:25 -07:00
|
|
|
Cluster* table;
|
2014-12-13 01:27:39 -07:00
|
|
|
uint8_t generation8; // Size must be not bigger than TTEntry::genBound8
|
2008-08-31 23:59:13 -06:00
|
|
|
};
|
|
|
|
|
2009-08-09 17:20:54 -06:00
|
|
|
extern TranspositionTable TT;
|
|
|
|
|
2021-02-26 02:02:13 -07:00
|
|
|
} // namespace Stockfish
|
|
|
|
|
2013-07-23 07:31:57 -06:00
|
|
|
#endif // #ifndef TT_H_INCLUDED
|