xiphos 64 bit random number

[pedrox]

I have seen the 64 bit random number of xiphos and I wanted to try it, but I have found that it generates different pv for the same position even with 1 thread (the same for my engine).

xiphos 1 thread.

first run

FEN: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1

Xiphos-0.4.24-w64-sse:
1 00:00 20 20k +0,82 1.d4
2 00:00 52 26k +0,92 1.e3 h5
3 00:00 88 44k +1,11 1.e3 h6 2.Nf3
4 00:00 139 70k +1,13 1.e3 g6 2.Nf3 h5
5 00:00 794 265k +0,37 1.d4 e6 2.Nf3 d5 3.Nc3
6 00:00 1k 357k +0,14 1.d4 e6 2.Nf3 Nf6 3.Nc3 d5
7 00:00 2k 360k +0,29 1.d4 e6 2.Nf3 Nf6 3.Nc3 d5 4.e3
8 00:00 3k 524k +0,10 1.d4 e6 2.Nf3 Nc6 3.e3 d5 4.Nc3 Nf6
9 00:00 8k 732k +0,06 1.d4 Nf6 2.h3 d5 3.Nf3 Nc6 4.Nc3 e6 5.e3
10 00:00 23k 899k +0,31 1.e4 d5 2.exd5 c6 3.dxc6 Nxc6 4.b3 e5 5.Nc3 Nf6 6.Nf3
11 00:00 28k 943k +0,31 1.e4 d5 2.exd5 c6 3.dxc6 Nxc6 4.b3 e5 5.Nc3 Nf6 6.Nf3
12 00:00 49k 1.012k +0,33 1.e4 d5 2.exd5 c6 3.c4 cxd5 4.d4 e6 5.Nc3 Nc6 6.Nf3 Nf6
13 00:00 201k 1.196k +0,15 1.d4 Nf6 2.c4 e6 3.Nc3 d5 4.Bg5 Bb4 5.e3 0-0 6.Nf3 Bxc3+ 7.bxc3
14 00:00 271k 1.206k +0,12 1.d4 Nf6 2.c4 e6 3.Nc3 d5 4.Bg5 Bb4 5.e3 0-0 6.Nf3 Bxc3+ 7.bxc3 Nc6
15 00:00 377k 1.223k +0,09 1.d4 Nf6 2.c4 e6 3.Nc3 d5 4.Nf3 Be7 5.Bf4 0-0 6.e3 Nc6 7.Bd3 Nb4 8.0-0
16 00:00 632k 1.198k +0,24 1.e4 e5 2.Nf3 Nc6 3.Nc3 Nf6 4.Bc4 Be7 5.d3 0-0 6.0-0 d6 7.Bb3 Be6 8.Nd5 Bxd5
17 00:00 1.146k 1.222k +0,27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.0-0 Bd6 5.c3 a6 6.Bd3 0-0 7.Bc2 Re8 8.d3 b5 9.Bd2 Bb7 10.h3
18 00:01 1.296k 1.228k +0,27 1.e4 e5 2.Nf3 Nc6 3.Bb5 Nf6 4.0-0 Bd6 5.c3 a6 6.Bd3 0-0 7.Bc2 Re8 8.d3 b5 9.Bd2 Bb7
19 00:01 1.887k 1.231k +0,29 1.e4 e5 2.Nf3 Nc6 3.Bb5 a6 4.Bxc6 dxc6 5.0-0 Bg4 6.d3 Nf6 7.h3 Bxf3 8.Qxf3 Qe7 9.Be3 0-0-0 10.Nd2 Kb8
20 00:03 4.070k 1.231k +0,19 1.e4 e6 2.d4 d5 3.exd5 exd5 4.Nf3 Nf6 5.Bd3 Be7 6.0-0 0-0 7.Re1 h6 8.Nc3 Nc6 9.Bf4 Nb4 10.h3 Re8 11.Ne5
21 00:06 7.668k 1.229k +0,22 1.d4 Nf6 2.c4 e6 3.Nc3 d5 4.Nf3 Bb4 5.e3 0-0 6.Bd3 c5 7.a3 Bxc3+ 8.bxc3 Qa5 9.Bd2 Qc7 10.cxd5 exd5 11.dxc5
22 00:08 10.790k 1.224k +0,23 1.d4 Nf6 2.c4 e6 3.Nc3 d5 4.Nf3 h6 5.Bf4 Bb4 6.e3 0-0 7.Bd3 dxc4 8.Bxc4 Nd5 9.Bxd5 exd5 10.0-0 Bxc3 11.bxc3 Nc6 12.Bg3
Best move: d4, Value: +0,23, Depth: 22, Time: 00:08,870, 10.790.277 Nodes, 1.224 kN/sec.

second run

FEN: rnbqkbnr/pppppppp/8/8/8/8/PPPPPPPP/RNBQKBNR w KQkq - 0 1

Xiphos-0.4.24-w64-sse:
1 00:00 20 20k +0,82 1.d4
2 00:00 52 26k +0,92 1.e3 h5
3 00:00 88 44k +1,11 1.e3 h6 2.Nf3
4 00:00 139 70k +1,13 1.e3 g6 2.Nf3 h5
5 00:00 794 265k +0,37 1.d4 e6 2.Nf3 d5 3.Nc3
6 00:00 1k 268k +0,14 1.d4 e6 2.Nf3 Nf6 3.Nc3 d5
7 00:00 2k 360k +0,29 1.d4 e6 2.Nf3 Nf6 3.Nc3 d5 4.e3
8 00:00 3k 524k +0,10 1.d4 e6 2.Nf3 Nc6 3.e3 d5 4.Nc3 Nf6
9 00:00 8k 732k +0,06 1.d4 Nf6 2.h3 d5 3.Nf3 Nc6 4.Nc3 e6 5.e3
10 00:00 23k 899k +0,31 1.e4 d5 2.exd5 c6 3.dxc6 Nxc6 4.b3 e5 5.Nc3 Nf6 6.Nf3
11 00:00 28k 943k +0,31 1.e4 d5 2.exd5 c6 3.dxc6 Nxc6 4.b3 e5 5.Nc3 Nf6 6.Nf3
12 00:00 49k 971k +0,33 1.e4 d5 2.exd5 c6 3.c4 cxd5 4.d4 e6 5.Nc3 Nc6 6.Nf3 Nf6
13 00:00 176k 1.167k +0,20 1.d4 Nf6 2.c4 e6 3.Nf3 d5 4.e3 Nc6 5.a3 Be7 6.Nc3 0-0 7.Be2
14 00:00 283k 1.180k +0,20 1.d4 Nf6 2.c4 e6 3.e3 Nc6 4.Nf3 d5 5.a3 Be7 6.Nc3 0-0 7.Be2 dxc4
15 00:00 533k 1.227k +0,12 1.d4 d5 2.Nf3 e6 3.c4 Nf6 4.Nc3 Be7 5.Bg5 0-0 6.e3 h6 7.Bxf6 Bxf6 8.Bd3
16 00:00 667k 1.191k +0,12 1.d4 d5 2.Nf3 e6 3.c4 Nf6 4.Nc3 Be7 5.Bg5 0-0 6.e3 h6 7.Bxf6 Bxf6 8.Bd3 dxc4
17 00:00 1.179k 1.224k +0,16 1.d4 d5 2.Nf3 e6 3.e3 c5 4.c4 Nf6 5.Nc3 Nc6 6.Be2 cxd4 7.exd4 Be7 8.a3 0-0 9.0-0 dxc4
18 00:01 1.716k 1.242k +0,19 1.d4 d5 2.Nf3 e6 3.c4 Nf6 4.e3 Bd6 5.cxd5 exd5 6.Nc3 Bd7 7.Bd3 h6 8.h3 0-0 9.0-0 Nc6 10.a3
19 00:01 2.488k 1.245k +0,19 1.d4 d5 2.Nf3 e6 3.c4 Nf6 4.Nc3 Be7 5.cxd5 exd5 6.Bf4 0-0 7.e3 Ne4 8.Bd3 Bb4 9.Rc1 Bxc3+ 10.bxc3 Bf5
20 00:02 3.696k 1.246k +0,20 1.d4 d5 2.Nf3 e6 3.c4 Nf6 4.Nc3 Be7 5.cxd5 exd5 6.Bf4 Nc6 7.e3 0-0 8.Be2 Ne4 9.0-0 Nxc3 10.bxc3 Bf5
21 00:05 6.722k 1.189k +0,23 1.d4 d5 2.Nf3 Nf6 3.c4 e6 4.Nc3 h6 5.e3 c5 6.cxd5 exd5 7.Be2 Be7 8.dxc5 Bxc5 9.a3 0-0 10.0-0 Bf5 11.Bd2 Nc6
22 00:12 13.809k 1.150k +0,21 1.d4 d5 2.Nf3 Nf6 3.c4 e6 4.Nc3 h6 5.e3 c5 6.cxd5 exd5 7.Be2 Be7 8.dxc5 0-0 9.0-0 Bxc5 10.a3 Re8 11.h3 Nc6
22 00:12 14.790k 1.147k +0,21 1.d4 d5 2.Nf3 Nf6 3.c4 e6 4.Nc3 h6 5.e3 c5 6.cxd5 exd5 7.Be2 Be7 8.dxc5 0-0 9.0-0 Bxc5 10.a3 Re8 11.h3 Nc6
Best move: d4, Value: +0,21, Depth: 22, Time: 00:12,948, 14.790.190 Nodes, 1.147 kN/sec.

pv is different from ply 13

64 bit random number used:

uint64_t rand64()
{
int i;
uint64_t r;

r = 0;
for (i = 0; i < 8; i ++)
r |= (uint64_t)(rand() & 0xff) << (i << 3);
return r;
}

This is a bug?
This is a problem? I do not think it affect the strength
maybe used to have more variety with pv for a lazy smp? (I do not know how xiphos does smp and I really do not know how smp works)

[D Sceviour]

I have seen the 64 bit random number of xiphos and I wanted to try it, but I have found that it generates different pv for the same position even with 1 thread (the same for my engine).

This is a bug?

No, that gives it some style. Random hash collisions and the random moves it generates might give it some advantage over the NN engines that are trying to predict its moves!

This is a problem? I do not think it affect the strength.

It does not effect overall strength. However, Schooner uses the standard polyglot keys. This saves space, makes debugging a lot easier, and keeps the hash system ready for polyglot book immediately.

(I do not know how xiphos does smp and I really do not know how smp works)

Xiphos uses shared hash table threading with staggered aspiration depths (taken from Ethereal). Hash keys probably have no effect on this.

[xr_a_y]

Just recompile with a fixed seed.

[Sven]

Code: Select all

uint64_t rand64()
{
  int i;
  uint64_t r;

  r = 0;
  for (i = 0; i < 8; i ++)
    r |= (uint64_t)(rand() & 0xff) << (i << 3);
  return r;
}

This is a bug?

The implementation of rand64() looks correct, the 64-bit return value "r" is composed from 8 pseudo-random numbers of 8 bit each, shifted correctly to the expected bit positions. But the implementation of reset_hash_key() in hash.c uses srand(time(0)) so zobrist keys are not the same for each program run and thus the results are not reproducible. Typically the set of zobrist keys should be fixed, either by using fixed constants or through a fixed random seed.