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
|
Add NNUE evaluation
This patch ports the efficiently updatable neural network (NNUE) evaluation to Stockfish.
Both the NNUE and the classical evaluations are available, and can be used to
assign a value to a position that is later used in alpha-beta (PVS) search to find the
best move. The classical evaluation computes this value as a function of various chess
concepts, handcrafted by experts, tested and tuned using fishtest. The NNUE evaluation
computes this value with a neural network based on basic inputs. The network is optimized
and trained on the evalutions of millions of positions at moderate search depth.
The NNUE evaluation was first introduced in shogi, and ported to Stockfish afterward.
It can be evaluated efficiently on CPUs, and exploits the fact that only parts
of the neural network need to be updated after a typical chess move.
[The nodchip repository](https://github.com/nodchip/Stockfish) provides additional
tools to train and develop the NNUE networks.
This patch is the result of contributions of various authors, from various communities,
including: nodchip, ynasu87, yaneurao (initial port and NNUE authors), domschl, FireFather,
rqs, xXH4CKST3RXx, tttak, zz4032, joergoster, mstembera, nguyenpham, erbsenzaehler,
dorzechowski, and vondele.
This new evaluation needed various changes to fishtest and the corresponding infrastructure,
for which tomtor, ppigazzini, noobpwnftw, daylen, and vondele are gratefully acknowledged.
The first networks have been provided by gekkehenker and sergiovieri, with the latter
net (nn-97f742aaefcd.nnue) being the current default.
The evaluation function can be selected at run time with the `Use NNUE` (true/false) UCI option,
provided the `EvalFile` option points the the network file (depending on the GUI, with full path).
The performance of the NNUE evaluation relative to the classical evaluation depends somewhat on
the hardware, and is expected to improve quickly, but is currently on > 80 Elo on fishtest:
60000 @ 10+0.1 th 1
https://tests.stockfishchess.org/tests/view/5f28fe6ea5abc164f05e4c4c
ELO: 92.77 +-2.1 (95%) LOS: 100.0%
Total: 60000 W: 24193 L: 8543 D: 27264
Ptnml(0-2): 609, 3850, 9708, 10948, 4885
40000 @ 20+0.2 th 8
https://tests.stockfishchess.org/tests/view/5f290229a5abc164f05e4c58
ELO: 89.47 +-2.0 (95%) LOS: 100.0%
Total: 40000 W: 12756 L: 2677 D: 24567
Ptnml(0-2): 74, 1583, 8550, 7776, 2017
At the same time, the impact on the classical evaluation remains minimal, causing no significant
regression:
sprt @ 10+0.1 th 1
https://tests.stockfishchess.org/tests/view/5f2906a2a5abc164f05e4c5b
LLR: 2.94 (-2.94,2.94) {-6.00,-4.00}
Total: 34936 W: 6502 L: 6825 D: 21609
Ptnml(0-2): 571, 4082, 8434, 3861, 520
sprt @ 60+0.6 th 1
https://tests.stockfishchess.org/tests/view/5f2906cfa5abc164f05e4c5d
LLR: 2.93 (-2.94,2.94) {-6.00,-4.00}
Total: 10088 W: 1232 L: 1265 D: 7591
Ptnml(0-2): 49, 914, 3170, 843, 68
The needed networks can be found at https://tests.stockfishchess.org/nns
It is recommended to use the default one as indicated by the `EvalFile` UCI option.
Guidelines for testing new nets can be found at
https://github.com/glinscott/fishtest/wiki/Creating-my-first-test#nnue-net-tests
Integration has been discussed in various issues:
https://github.com/official-stockfish/Stockfish/issues/2823
https://github.com/official-stockfish/Stockfish/issues/2728
The integration branch will be closed after the merge:
https://github.com/official-stockfish/Stockfish/pull/2825
https://github.com/official-stockfish/Stockfish/tree/nnue-player-wip
closes https://github.com/official-stockfish/Stockfish/pull/2912
This will be an exciting time for computer chess, looking forward to seeing the evolution of
this approach.
Bench: 4746616
2020-08-05 09:11:15 -06:00
|
|
|
Copyright (C) 2004-2020 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 MISC_H_INCLUDED
|
2008-08-31 23:59:13 -06:00
|
|
|
#define MISC_H_INCLUDED
|
|
|
|
|
2014-12-08 00:23:09 -07:00
|
|
|
#include <cassert>
|
2015-02-24 04:24:53 -07:00
|
|
|
#include <chrono>
|
2014-09-14 02:06:36 -06:00
|
|
|
#include <ostream>
|
2008-08-31 23:59:13 -06:00
|
|
|
#include <string>
|
2012-03-31 05:15:57 -06:00
|
|
|
#include <vector>
|
2011-11-05 04:19:21 -06:00
|
|
|
|
2009-05-10 05:21:45 -06:00
|
|
|
#include "types.h"
|
2008-08-31 23:59:13 -06:00
|
|
|
|
2014-11-01 23:10:31 -06:00
|
|
|
const std::string engine_info(bool to_uci = false);
|
2019-09-24 11:00:27 -06:00
|
|
|
const std::string compiler_info();
|
2015-02-07 11:13:41 -07:00
|
|
|
void prefetch(void* addr);
|
2016-06-12 17:12:24 -06:00
|
|
|
void start_logger(const std::string& fname);
|
Add NNUE evaluation
This patch ports the efficiently updatable neural network (NNUE) evaluation to Stockfish.
Both the NNUE and the classical evaluations are available, and can be used to
assign a value to a position that is later used in alpha-beta (PVS) search to find the
best move. The classical evaluation computes this value as a function of various chess
concepts, handcrafted by experts, tested and tuned using fishtest. The NNUE evaluation
computes this value with a neural network based on basic inputs. The network is optimized
and trained on the evalutions of millions of positions at moderate search depth.
The NNUE evaluation was first introduced in shogi, and ported to Stockfish afterward.
It can be evaluated efficiently on CPUs, and exploits the fact that only parts
of the neural network need to be updated after a typical chess move.
[The nodchip repository](https://github.com/nodchip/Stockfish) provides additional
tools to train and develop the NNUE networks.
This patch is the result of contributions of various authors, from various communities,
including: nodchip, ynasu87, yaneurao (initial port and NNUE authors), domschl, FireFather,
rqs, xXH4CKST3RXx, tttak, zz4032, joergoster, mstembera, nguyenpham, erbsenzaehler,
dorzechowski, and vondele.
This new evaluation needed various changes to fishtest and the corresponding infrastructure,
for which tomtor, ppigazzini, noobpwnftw, daylen, and vondele are gratefully acknowledged.
The first networks have been provided by gekkehenker and sergiovieri, with the latter
net (nn-97f742aaefcd.nnue) being the current default.
The evaluation function can be selected at run time with the `Use NNUE` (true/false) UCI option,
provided the `EvalFile` option points the the network file (depending on the GUI, with full path).
The performance of the NNUE evaluation relative to the classical evaluation depends somewhat on
the hardware, and is expected to improve quickly, but is currently on > 80 Elo on fishtest:
60000 @ 10+0.1 th 1
https://tests.stockfishchess.org/tests/view/5f28fe6ea5abc164f05e4c4c
ELO: 92.77 +-2.1 (95%) LOS: 100.0%
Total: 60000 W: 24193 L: 8543 D: 27264
Ptnml(0-2): 609, 3850, 9708, 10948, 4885
40000 @ 20+0.2 th 8
https://tests.stockfishchess.org/tests/view/5f290229a5abc164f05e4c58
ELO: 89.47 +-2.0 (95%) LOS: 100.0%
Total: 40000 W: 12756 L: 2677 D: 24567
Ptnml(0-2): 74, 1583, 8550, 7776, 2017
At the same time, the impact on the classical evaluation remains minimal, causing no significant
regression:
sprt @ 10+0.1 th 1
https://tests.stockfishchess.org/tests/view/5f2906a2a5abc164f05e4c5b
LLR: 2.94 (-2.94,2.94) {-6.00,-4.00}
Total: 34936 W: 6502 L: 6825 D: 21609
Ptnml(0-2): 571, 4082, 8434, 3861, 520
sprt @ 60+0.6 th 1
https://tests.stockfishchess.org/tests/view/5f2906cfa5abc164f05e4c5d
LLR: 2.93 (-2.94,2.94) {-6.00,-4.00}
Total: 10088 W: 1232 L: 1265 D: 7591
Ptnml(0-2): 49, 914, 3170, 843, 68
The needed networks can be found at https://tests.stockfishchess.org/nns
It is recommended to use the default one as indicated by the `EvalFile` UCI option.
Guidelines for testing new nets can be found at
https://github.com/glinscott/fishtest/wiki/Creating-my-first-test#nnue-net-tests
Integration has been discussed in various issues:
https://github.com/official-stockfish/Stockfish/issues/2823
https://github.com/official-stockfish/Stockfish/issues/2728
The integration branch will be closed after the merge:
https://github.com/official-stockfish/Stockfish/pull/2825
https://github.com/official-stockfish/Stockfish/tree/nnue-player-wip
closes https://github.com/official-stockfish/Stockfish/pull/2912
This will be an exciting time for computer chess, looking forward to seeing the evolution of
this approach.
Bench: 4746616
2020-08-05 09:11:15 -06:00
|
|
|
void* std_aligned_alloc(size_t alignment, size_t size);
|
|
|
|
void std_aligned_free(void* ptr);
|
2020-02-05 07:18:24 -07:00
|
|
|
void* aligned_ttmem_alloc(size_t size, void*& mem);
|
2020-05-14 03:00:35 -06:00
|
|
|
void aligned_ttmem_free(void* mem); // nop if mem == nullptr
|
2014-11-01 23:10:31 -06:00
|
|
|
|
|
|
|
void dbg_hit_on(bool b);
|
2015-02-07 03:15:38 -07:00
|
|
|
void dbg_hit_on(bool c, bool b);
|
2014-11-01 23:10:31 -06:00
|
|
|
void dbg_mean_of(int v);
|
|
|
|
void dbg_print();
|
2008-10-11 11:25:16 -06:00
|
|
|
|
2015-02-24 04:24:53 -07:00
|
|
|
typedef std::chrono::milliseconds::rep TimePoint; // A value in milliseconds
|
2018-03-27 08:22:53 -06:00
|
|
|
static_assert(sizeof(TimePoint) == sizeof(int64_t), "TimePoint should be 64 bits");
|
2015-02-24 04:24:53 -07:00
|
|
|
inline TimePoint now() {
|
|
|
|
return std::chrono::duration_cast<std::chrono::milliseconds>
|
|
|
|
(std::chrono::steady_clock::now().time_since_epoch()).count();
|
2012-09-04 01:38:51 -06:00
|
|
|
}
|
2012-03-03 01:35:56 -07:00
|
|
|
|
2012-03-31 05:15:57 -06:00
|
|
|
template<class Entry, int Size>
|
|
|
|
struct HashTable {
|
2014-12-07 16:53:33 -07:00
|
|
|
Entry* operator[](Key key) { return &table[(uint32_t)key & (Size - 1)]; }
|
2012-03-31 05:15:57 -06:00
|
|
|
|
|
|
|
private:
|
2019-03-31 03:47:36 -06:00
|
|
|
std::vector<Entry> table = std::vector<Entry>(Size); // Allocate on the heap
|
2012-03-31 05:15:57 -06:00
|
|
|
};
|
|
|
|
|
2012-08-29 03:25:11 -06:00
|
|
|
|
2014-02-09 09:31:45 -07:00
|
|
|
enum SyncCout { IO_LOCK, IO_UNLOCK };
|
2012-08-29 03:25:11 -06:00
|
|
|
std::ostream& operator<<(std::ostream&, SyncCout);
|
|
|
|
|
2014-02-09 09:31:45 -07:00
|
|
|
#define sync_cout std::cout << IO_LOCK
|
|
|
|
#define sync_endl std::endl << IO_UNLOCK
|
2012-08-29 03:25:11 -06:00
|
|
|
|
Simpler PRNG and faster magics search
This patch replaces RKISS by a simpler and faster PRNG, xorshift64* proposed
by S. Vigna (2014). It is extremely simple, has a large enough period for
Stockfish's needs (2^64), requires no warming-up (allowing such code to be
removed), and offers slightly better randomness than MT19937.
Paper: http://xorshift.di.unimi.it/
Reference source code (public domain):
http://xorshift.di.unimi.it/xorshift64star.c
The patch also simplifies how init_magics() searches for magics:
- Old logic: seed the PRNG always with the same seed,
then use optimized bit rotations to tailor the RNG sequence per rank.
- New logic: seed the PRNG with an optimized seed per rank.
This has two advantages:
1. Less code and less computation to perform during magics search (not ROTL).
2. More choices for random sequence tuning. The old logic only let us choose
from 4096 bit rotation pairs. With the new one, we can look for the best seeds
among 2^64 values. Indeed, the set of seeds[][] provided in the patch reduces
the effort needed to find the magics:
64-bit SF:
Old logic -> 5,783,789 rand64() calls needed to find the magics
New logic -> 4,420,086 calls
32-bit SF:
Old logic -> 2,175,518 calls
New logic -> 1,895,955 calls
In the 64-bit case, init_magics() take 25 ms less to complete (Intel Core i5).
Finally, when playing with strength handicap, non-determinism is achieved
by setting the seed of the static RNG only once. Afterwards, there is no need
to skip output values.
The bench only changes because the Zobrist keys are now different (since they
are random numbers straight out of the PRNG).
The RNG seed has been carefully chosen so that the
resulting Zobrist keys are particularly well-behaved:
1. All triplets of XORed keys are unique, implying that it
would take at least 7 keys to find a 64-bit collision
(test suggested by ceebo)
2. All pairs of XORed keys are unique modulo 2^32
3. The cardinality of { (key1 ^ key2) >> 48 } is as close
as possible to the maximum (65536)
Point 2 aims at ensuring a good distribution among the bits
that determine an TT entry's cluster, likewise point 3
among the bits that form the TT entry's key16 inside a
cluster.
Details:
Bitset card(key1^key2)
------ ---------------
RKISS
key16 64894 = 99.020% of theoretical maximum
low18 180117 = 99.293%
low32 305362 = 99.997%
Xorshift64*, old seed
key16 64918 = 99.057%
low18 179994 = 99.225%
low32 305350 = 99.993%
Xorshift64*, new seed
key16 65027 = 99.223%
low18 181118 = 99.845%
low32 305371 = 100.000%
Bench: 9324905
Resolves #148
2014-12-07 17:10:57 -07:00
|
|
|
|
|
|
|
/// xorshift64star Pseudo-Random Number Generator
|
|
|
|
/// This class is based on original code written and dedicated
|
|
|
|
/// to the public domain by Sebastiano Vigna (2014).
|
|
|
|
/// It has the following characteristics:
|
2014-12-08 00:23:09 -07:00
|
|
|
///
|
Simpler PRNG and faster magics search
This patch replaces RKISS by a simpler and faster PRNG, xorshift64* proposed
by S. Vigna (2014). It is extremely simple, has a large enough period for
Stockfish's needs (2^64), requires no warming-up (allowing such code to be
removed), and offers slightly better randomness than MT19937.
Paper: http://xorshift.di.unimi.it/
Reference source code (public domain):
http://xorshift.di.unimi.it/xorshift64star.c
The patch also simplifies how init_magics() searches for magics:
- Old logic: seed the PRNG always with the same seed,
then use optimized bit rotations to tailor the RNG sequence per rank.
- New logic: seed the PRNG with an optimized seed per rank.
This has two advantages:
1. Less code and less computation to perform during magics search (not ROTL).
2. More choices for random sequence tuning. The old logic only let us choose
from 4096 bit rotation pairs. With the new one, we can look for the best seeds
among 2^64 values. Indeed, the set of seeds[][] provided in the patch reduces
the effort needed to find the magics:
64-bit SF:
Old logic -> 5,783,789 rand64() calls needed to find the magics
New logic -> 4,420,086 calls
32-bit SF:
Old logic -> 2,175,518 calls
New logic -> 1,895,955 calls
In the 64-bit case, init_magics() take 25 ms less to complete (Intel Core i5).
Finally, when playing with strength handicap, non-determinism is achieved
by setting the seed of the static RNG only once. Afterwards, there is no need
to skip output values.
The bench only changes because the Zobrist keys are now different (since they
are random numbers straight out of the PRNG).
The RNG seed has been carefully chosen so that the
resulting Zobrist keys are particularly well-behaved:
1. All triplets of XORed keys are unique, implying that it
would take at least 7 keys to find a 64-bit collision
(test suggested by ceebo)
2. All pairs of XORed keys are unique modulo 2^32
3. The cardinality of { (key1 ^ key2) >> 48 } is as close
as possible to the maximum (65536)
Point 2 aims at ensuring a good distribution among the bits
that determine an TT entry's cluster, likewise point 3
among the bits that form the TT entry's key16 inside a
cluster.
Details:
Bitset card(key1^key2)
------ ---------------
RKISS
key16 64894 = 99.020% of theoretical maximum
low18 180117 = 99.293%
low32 305362 = 99.997%
Xorshift64*, old seed
key16 64918 = 99.057%
low18 179994 = 99.225%
low32 305350 = 99.993%
Xorshift64*, new seed
key16 65027 = 99.223%
low18 181118 = 99.845%
low32 305371 = 100.000%
Bench: 9324905
Resolves #148
2014-12-07 17:10:57 -07:00
|
|
|
/// - Outputs 64-bit numbers
|
|
|
|
/// - Passes Dieharder and SmallCrush test batteries
|
|
|
|
/// - Does not require warm-up, no zeroland to escape
|
|
|
|
/// - Internal state is a single 64-bit integer
|
|
|
|
/// - Period is 2^64 - 1
|
|
|
|
/// - Speed: 1.60 ns/call (Core i7 @3.40GHz)
|
2014-12-08 00:23:09 -07:00
|
|
|
///
|
Simpler PRNG and faster magics search
This patch replaces RKISS by a simpler and faster PRNG, xorshift64* proposed
by S. Vigna (2014). It is extremely simple, has a large enough period for
Stockfish's needs (2^64), requires no warming-up (allowing such code to be
removed), and offers slightly better randomness than MT19937.
Paper: http://xorshift.di.unimi.it/
Reference source code (public domain):
http://xorshift.di.unimi.it/xorshift64star.c
The patch also simplifies how init_magics() searches for magics:
- Old logic: seed the PRNG always with the same seed,
then use optimized bit rotations to tailor the RNG sequence per rank.
- New logic: seed the PRNG with an optimized seed per rank.
This has two advantages:
1. Less code and less computation to perform during magics search (not ROTL).
2. More choices for random sequence tuning. The old logic only let us choose
from 4096 bit rotation pairs. With the new one, we can look for the best seeds
among 2^64 values. Indeed, the set of seeds[][] provided in the patch reduces
the effort needed to find the magics:
64-bit SF:
Old logic -> 5,783,789 rand64() calls needed to find the magics
New logic -> 4,420,086 calls
32-bit SF:
Old logic -> 2,175,518 calls
New logic -> 1,895,955 calls
In the 64-bit case, init_magics() take 25 ms less to complete (Intel Core i5).
Finally, when playing with strength handicap, non-determinism is achieved
by setting the seed of the static RNG only once. Afterwards, there is no need
to skip output values.
The bench only changes because the Zobrist keys are now different (since they
are random numbers straight out of the PRNG).
The RNG seed has been carefully chosen so that the
resulting Zobrist keys are particularly well-behaved:
1. All triplets of XORed keys are unique, implying that it
would take at least 7 keys to find a 64-bit collision
(test suggested by ceebo)
2. All pairs of XORed keys are unique modulo 2^32
3. The cardinality of { (key1 ^ key2) >> 48 } is as close
as possible to the maximum (65536)
Point 2 aims at ensuring a good distribution among the bits
that determine an TT entry's cluster, likewise point 3
among the bits that form the TT entry's key16 inside a
cluster.
Details:
Bitset card(key1^key2)
------ ---------------
RKISS
key16 64894 = 99.020% of theoretical maximum
low18 180117 = 99.293%
low32 305362 = 99.997%
Xorshift64*, old seed
key16 64918 = 99.057%
low18 179994 = 99.225%
low32 305350 = 99.993%
Xorshift64*, new seed
key16 65027 = 99.223%
low18 181118 = 99.845%
low32 305371 = 100.000%
Bench: 9324905
Resolves #148
2014-12-07 17:10:57 -07:00
|
|
|
/// For further analysis see
|
|
|
|
/// <http://vigna.di.unimi.it/ftp/papers/xorshift.pdf>
|
|
|
|
|
|
|
|
class PRNG {
|
|
|
|
|
2014-12-08 00:23:09 -07:00
|
|
|
uint64_t s;
|
Simpler PRNG and faster magics search
This patch replaces RKISS by a simpler and faster PRNG, xorshift64* proposed
by S. Vigna (2014). It is extremely simple, has a large enough period for
Stockfish's needs (2^64), requires no warming-up (allowing such code to be
removed), and offers slightly better randomness than MT19937.
Paper: http://xorshift.di.unimi.it/
Reference source code (public domain):
http://xorshift.di.unimi.it/xorshift64star.c
The patch also simplifies how init_magics() searches for magics:
- Old logic: seed the PRNG always with the same seed,
then use optimized bit rotations to tailor the RNG sequence per rank.
- New logic: seed the PRNG with an optimized seed per rank.
This has two advantages:
1. Less code and less computation to perform during magics search (not ROTL).
2. More choices for random sequence tuning. The old logic only let us choose
from 4096 bit rotation pairs. With the new one, we can look for the best seeds
among 2^64 values. Indeed, the set of seeds[][] provided in the patch reduces
the effort needed to find the magics:
64-bit SF:
Old logic -> 5,783,789 rand64() calls needed to find the magics
New logic -> 4,420,086 calls
32-bit SF:
Old logic -> 2,175,518 calls
New logic -> 1,895,955 calls
In the 64-bit case, init_magics() take 25 ms less to complete (Intel Core i5).
Finally, when playing with strength handicap, non-determinism is achieved
by setting the seed of the static RNG only once. Afterwards, there is no need
to skip output values.
The bench only changes because the Zobrist keys are now different (since they
are random numbers straight out of the PRNG).
The RNG seed has been carefully chosen so that the
resulting Zobrist keys are particularly well-behaved:
1. All triplets of XORed keys are unique, implying that it
would take at least 7 keys to find a 64-bit collision
(test suggested by ceebo)
2. All pairs of XORed keys are unique modulo 2^32
3. The cardinality of { (key1 ^ key2) >> 48 } is as close
as possible to the maximum (65536)
Point 2 aims at ensuring a good distribution among the bits
that determine an TT entry's cluster, likewise point 3
among the bits that form the TT entry's key16 inside a
cluster.
Details:
Bitset card(key1^key2)
------ ---------------
RKISS
key16 64894 = 99.020% of theoretical maximum
low18 180117 = 99.293%
low32 305362 = 99.997%
Xorshift64*, old seed
key16 64918 = 99.057%
low18 179994 = 99.225%
low32 305350 = 99.993%
Xorshift64*, new seed
key16 65027 = 99.223%
low18 181118 = 99.845%
low32 305371 = 100.000%
Bench: 9324905
Resolves #148
2014-12-07 17:10:57 -07:00
|
|
|
|
|
|
|
uint64_t rand64() {
|
2014-12-08 00:23:09 -07:00
|
|
|
|
|
|
|
s ^= s >> 12, s ^= s << 25, s ^= s >> 27;
|
|
|
|
return s * 2685821657736338717LL;
|
Simpler PRNG and faster magics search
This patch replaces RKISS by a simpler and faster PRNG, xorshift64* proposed
by S. Vigna (2014). It is extremely simple, has a large enough period for
Stockfish's needs (2^64), requires no warming-up (allowing such code to be
removed), and offers slightly better randomness than MT19937.
Paper: http://xorshift.di.unimi.it/
Reference source code (public domain):
http://xorshift.di.unimi.it/xorshift64star.c
The patch also simplifies how init_magics() searches for magics:
- Old logic: seed the PRNG always with the same seed,
then use optimized bit rotations to tailor the RNG sequence per rank.
- New logic: seed the PRNG with an optimized seed per rank.
This has two advantages:
1. Less code and less computation to perform during magics search (not ROTL).
2. More choices for random sequence tuning. The old logic only let us choose
from 4096 bit rotation pairs. With the new one, we can look for the best seeds
among 2^64 values. Indeed, the set of seeds[][] provided in the patch reduces
the effort needed to find the magics:
64-bit SF:
Old logic -> 5,783,789 rand64() calls needed to find the magics
New logic -> 4,420,086 calls
32-bit SF:
Old logic -> 2,175,518 calls
New logic -> 1,895,955 calls
In the 64-bit case, init_magics() take 25 ms less to complete (Intel Core i5).
Finally, when playing with strength handicap, non-determinism is achieved
by setting the seed of the static RNG only once. Afterwards, there is no need
to skip output values.
The bench only changes because the Zobrist keys are now different (since they
are random numbers straight out of the PRNG).
The RNG seed has been carefully chosen so that the
resulting Zobrist keys are particularly well-behaved:
1. All triplets of XORed keys are unique, implying that it
would take at least 7 keys to find a 64-bit collision
(test suggested by ceebo)
2. All pairs of XORed keys are unique modulo 2^32
3. The cardinality of { (key1 ^ key2) >> 48 } is as close
as possible to the maximum (65536)
Point 2 aims at ensuring a good distribution among the bits
that determine an TT entry's cluster, likewise point 3
among the bits that form the TT entry's key16 inside a
cluster.
Details:
Bitset card(key1^key2)
------ ---------------
RKISS
key16 64894 = 99.020% of theoretical maximum
low18 180117 = 99.293%
low32 305362 = 99.997%
Xorshift64*, old seed
key16 64918 = 99.057%
low18 179994 = 99.225%
low32 305350 = 99.993%
Xorshift64*, new seed
key16 65027 = 99.223%
low18 181118 = 99.845%
low32 305371 = 100.000%
Bench: 9324905
Resolves #148
2014-12-07 17:10:57 -07:00
|
|
|
}
|
|
|
|
|
|
|
|
public:
|
2014-12-08 00:23:09 -07:00
|
|
|
PRNG(uint64_t seed) : s(seed) { assert(seed); }
|
Simpler PRNG and faster magics search
This patch replaces RKISS by a simpler and faster PRNG, xorshift64* proposed
by S. Vigna (2014). It is extremely simple, has a large enough period for
Stockfish's needs (2^64), requires no warming-up (allowing such code to be
removed), and offers slightly better randomness than MT19937.
Paper: http://xorshift.di.unimi.it/
Reference source code (public domain):
http://xorshift.di.unimi.it/xorshift64star.c
The patch also simplifies how init_magics() searches for magics:
- Old logic: seed the PRNG always with the same seed,
then use optimized bit rotations to tailor the RNG sequence per rank.
- New logic: seed the PRNG with an optimized seed per rank.
This has two advantages:
1. Less code and less computation to perform during magics search (not ROTL).
2. More choices for random sequence tuning. The old logic only let us choose
from 4096 bit rotation pairs. With the new one, we can look for the best seeds
among 2^64 values. Indeed, the set of seeds[][] provided in the patch reduces
the effort needed to find the magics:
64-bit SF:
Old logic -> 5,783,789 rand64() calls needed to find the magics
New logic -> 4,420,086 calls
32-bit SF:
Old logic -> 2,175,518 calls
New logic -> 1,895,955 calls
In the 64-bit case, init_magics() take 25 ms less to complete (Intel Core i5).
Finally, when playing with strength handicap, non-determinism is achieved
by setting the seed of the static RNG only once. Afterwards, there is no need
to skip output values.
The bench only changes because the Zobrist keys are now different (since they
are random numbers straight out of the PRNG).
The RNG seed has been carefully chosen so that the
resulting Zobrist keys are particularly well-behaved:
1. All triplets of XORed keys are unique, implying that it
would take at least 7 keys to find a 64-bit collision
(test suggested by ceebo)
2. All pairs of XORed keys are unique modulo 2^32
3. The cardinality of { (key1 ^ key2) >> 48 } is as close
as possible to the maximum (65536)
Point 2 aims at ensuring a good distribution among the bits
that determine an TT entry's cluster, likewise point 3
among the bits that form the TT entry's key16 inside a
cluster.
Details:
Bitset card(key1^key2)
------ ---------------
RKISS
key16 64894 = 99.020% of theoretical maximum
low18 180117 = 99.293%
low32 305362 = 99.997%
Xorshift64*, old seed
key16 64918 = 99.057%
low18 179994 = 99.225%
low32 305350 = 99.993%
Xorshift64*, new seed
key16 65027 = 99.223%
low18 181118 = 99.845%
low32 305371 = 100.000%
Bench: 9324905
Resolves #148
2014-12-07 17:10:57 -07:00
|
|
|
|
|
|
|
template<typename T> T rand() { return T(rand64()); }
|
|
|
|
|
|
|
|
/// Special generator used to fast init magic numbers.
|
|
|
|
/// Output values only have 1/8th of their bits set on average.
|
2014-12-08 00:23:09 -07:00
|
|
|
template<typename T> T sparse_rand()
|
|
|
|
{ return T(rand64() & rand64() & rand64()); }
|
Simpler PRNG and faster magics search
This patch replaces RKISS by a simpler and faster PRNG, xorshift64* proposed
by S. Vigna (2014). It is extremely simple, has a large enough period for
Stockfish's needs (2^64), requires no warming-up (allowing such code to be
removed), and offers slightly better randomness than MT19937.
Paper: http://xorshift.di.unimi.it/
Reference source code (public domain):
http://xorshift.di.unimi.it/xorshift64star.c
The patch also simplifies how init_magics() searches for magics:
- Old logic: seed the PRNG always with the same seed,
then use optimized bit rotations to tailor the RNG sequence per rank.
- New logic: seed the PRNG with an optimized seed per rank.
This has two advantages:
1. Less code and less computation to perform during magics search (not ROTL).
2. More choices for random sequence tuning. The old logic only let us choose
from 4096 bit rotation pairs. With the new one, we can look for the best seeds
among 2^64 values. Indeed, the set of seeds[][] provided in the patch reduces
the effort needed to find the magics:
64-bit SF:
Old logic -> 5,783,789 rand64() calls needed to find the magics
New logic -> 4,420,086 calls
32-bit SF:
Old logic -> 2,175,518 calls
New logic -> 1,895,955 calls
In the 64-bit case, init_magics() take 25 ms less to complete (Intel Core i5).
Finally, when playing with strength handicap, non-determinism is achieved
by setting the seed of the static RNG only once. Afterwards, there is no need
to skip output values.
The bench only changes because the Zobrist keys are now different (since they
are random numbers straight out of the PRNG).
The RNG seed has been carefully chosen so that the
resulting Zobrist keys are particularly well-behaved:
1. All triplets of XORed keys are unique, implying that it
would take at least 7 keys to find a 64-bit collision
(test suggested by ceebo)
2. All pairs of XORed keys are unique modulo 2^32
3. The cardinality of { (key1 ^ key2) >> 48 } is as close
as possible to the maximum (65536)
Point 2 aims at ensuring a good distribution among the bits
that determine an TT entry's cluster, likewise point 3
among the bits that form the TT entry's key16 inside a
cluster.
Details:
Bitset card(key1^key2)
------ ---------------
RKISS
key16 64894 = 99.020% of theoretical maximum
low18 180117 = 99.293%
low32 305362 = 99.997%
Xorshift64*, old seed
key16 64918 = 99.057%
low18 179994 = 99.225%
low32 305350 = 99.993%
Xorshift64*, new seed
key16 65027 = 99.223%
low18 181118 = 99.845%
low32 305371 = 100.000%
Bench: 9324905
Resolves #148
2014-12-07 17:10:57 -07:00
|
|
|
};
|
|
|
|
|
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
|
|
|
inline uint64_t mul_hi64(uint64_t a, uint64_t b) {
|
|
|
|
#if defined(__GNUC__) && defined(IS_64BIT)
|
|
|
|
__extension__ typedef unsigned __int128 uint128;
|
|
|
|
return ((uint128)a * (uint128)b) >> 64;
|
|
|
|
#else
|
|
|
|
uint64_t aL = (uint32_t)a, aH = a >> 32;
|
|
|
|
uint64_t bL = (uint32_t)b, bH = b >> 32;
|
|
|
|
uint64_t c1 = (aL * bL) >> 32;
|
|
|
|
uint64_t c2 = aH * bL + c1;
|
|
|
|
uint64_t c3 = aL * bH + (uint32_t)c2;
|
|
|
|
return aH * bH + (c2 >> 32) + (c3 >> 32);
|
|
|
|
#endif
|
|
|
|
}
|
2016-11-21 23:41:46 -07:00
|
|
|
|
|
|
|
/// Under Windows it is not possible for a process to run on more than one
|
|
|
|
/// logical processor group. This usually means to be limited to use max 64
|
|
|
|
/// cores. To overcome this, some special platform specific API should be
|
|
|
|
/// called to set group affinity for each thread. Original code from Texel by
|
|
|
|
/// Peter Österlund.
|
|
|
|
|
|
|
|
namespace WinProcGroup {
|
|
|
|
void bindThisThread(size_t idx);
|
|
|
|
}
|
|
|
|
|
Embed default net, and simplify using non-default nets
covers the most important cases from the user perspective:
It embeds the default net in the binary, so a download of that binary will result
in a working engine with the default net. The engine will be functional in the default mode
without any additional user action.
It allows non-default nets to be used, which will be looked for in up to
three directories (working directory, location of the binary, and optionally a specific default directory).
This mechanism is also kept for those developers that use MSVC,
the one compiler that doesn't have an easy mechanism for embedding data.
It is possible to disable embedding, and instead specify a specific directory, e.g. linux distros might want to use
CXXFLAGS="-DNNUE_EMBEDDING_OFF -DDEFAULT_NNUE_DIRECTORY=/usr/share/games/stockfish/" make -j ARCH=x86-64 profile-build
passed STC non-regression:
https://tests.stockfishchess.org/tests/view/5f4a581c150f0aef5f8ae03a
LLR: 2.95 (-2.94,2.94) {-1.25,-0.25}
Total: 66928 W: 7202 L: 7147 D: 52579
Ptnml(0-2): 291, 5309, 22211, 5360, 293
closes https://github.com/official-stockfish/Stockfish/pull/3070
fixes https://github.com/official-stockfish/Stockfish/issues/3030
No functional change.
2020-08-23 05:43:38 -06:00
|
|
|
namespace CommandLine {
|
|
|
|
void init(int argc, char* argv[]);
|
|
|
|
|
|
|
|
extern std::string binaryDirectory; // path of the executable directory
|
|
|
|
extern std::string workingDirectory; // path of the working directory
|
|
|
|
}
|
|
|
|
|
2013-07-23 07:31:57 -06:00
|
|
|
#endif // #ifndef MISC_H_INCLUDED
|