WAC Falls into a manhole. Someone pulled the cover off

[Dann Corbit]

Geschrieben von: / Posted by: Dann Corbit at 14 April 2000 22:29:18:
Gullydeckel 2 solved 240/300 on the WAC test at 5 seconds/position.
This, on a weak 300 MHz PII.
The interesting thing is the high NPS figures for some positions. While generally getting around 200K NPS it sometimes dips under 100K and sometimes goes way up. Consider this position:
+----+----+----+----+----+----+----+----+
| | *k | | | | | | |
+----+----+----+----+----+----+----+----+
| | | | | | r | p | |
+----+----+----+----+----+----+----+----+
| | p | | | | | | |
+----+----+----+----+----+----+----+----+
| | k | | | | | | |
+----+----+----+----+----+----+----+----+
| | | | | | | *r | |
+----+----+----+----+----+----+----+----+
| | | | | | | | |
+----+----+----+----+----+----+----+----+
| | | | | | | | |
+----+----+----+----+----+----+----+----+
| | | | | | | | |
+----+----+----+----+----+----+----+----+
pos: "WAC.041" Solution: Ka5 Kc5
1+ 0.40 0.00 1 f7f8
1& 2.68 0.00 2 f7f8
1& 2.69 0.00 14 b5c6
1& 2.71 0.00 17 b5c5
1& 3.02 0.00 26 b6b7
1. 3.02 0.00 26 b6b7
2& 3.02 0.00 41 b6b7 b8a7
2+ 3.42 0.00 87 f7f8
2& 6.17 0.00 109 f7f8 b8b7 g7g8q g4g8 f8g8
2. 6.17 0.00 109 f7f8 b8b7 g7g8q g4g8 f8g8
3& 6.17 0.00 174 f7f8 b8b7 g7g8q g4g8 f8g8
3. 6.17 0.00 535 f7f8 b8b7 g7g8q g4g8 f8g8
4- 5.77 0.02 3092 f7f8 b8b7 g7g8q g4g8 f8g8
4& 4.83 0.02 3503 f7f8 b8b7 g7g8q g4g5 b5c4 g5g8 f8g8 b7b6

4. 4.83 0.03 4856 f7f8 b8b7 g7g8q g4g5 b5c4 g5g8 f8g8 b7b6

5+ 5.23 0.04 6596 f7e7
5& 6.12 0.05 7705 f7e7 g4g5 b5a6 b8c8 e7e8 c8d7 g7g8q g5g8
e8g8
5. 6.12 0.07 14153 f7e7 g4g5 b5a6 b8c8 e7e8 c8d7 g7g8q g5g8
e8g8
6& 6.13 0.18 32524 b5a5 b8c8 f7f8 c8d7 g7g8q g4g8 f8g8
6. 6.13 0.19 36629 b5a5 b8c8 f7f8 c8d7 g7g8q g4g8 f8g8
7& 6.11 0.23 42678 b5a5 b8c8 f7f8 c8d7 g7g8q g4g8 f8g8 d7e6

7. 6.11 0.52 100159 b5a5 b8c8 f7f8 c8d7 g7g8q g4g8 f8g8 d7e6

8& 6.13 0.66 124865 b5a5 b8c8 f7f8 c8b7 g7g8q g4g8 f8g8 b7c6
g8f8
8. 6.13 1.26 239242 b5a5 b8c8 f7f8 c8b7 g7g8q g4g8 f8g8 b7c6
g8f8
9& 6.17 1.61 307788 b5a5 b8c8 f7f8 c8b7 g7g8q g4g8 f8g8 b7c6
g8g6 c6b7 a5b4
9. 6.17 3.26 623367 b5a5 b8c8 f7f8 c8b7 g7g8q g4g8 f8g8 b7c6
g8g6 c6b7 a5b4
10& 6.43 3.60 675562 b5a5 b8a8 f7f8 a8b7 g7g8q g4g8 f8g8 b7c6
g8g6 c6d5 b6b7 d5e4
10i 6.17 5.01 926592 b5a5 b8a8 f7f8 a8b7 g7g8q g4g8 f8g8 b7c6
g8g6 c6d5 b6b7 d5e4
game_phase: 3
evals:926592 (full:39142 lazy:95.78%), time: 5.01 [305896.21 nps]
Nodes in search: 605948 , quies: 926592 (60.46%)
moves in search - generated: 6180847 , looked at: 1792655 (29.00%)
pawn hash: hits: 39134 misses: 8 (99.980%)
Correct.
As you can see, G2 managed over 300K NPS.
It's quite an interesting patch of code! I have never seen the same mechanisms used before.

My FTP site

[Pete Galati]

Geschrieben von: / Posted by: Pete Galati at 14 April 2000 23:31:29:
Als Antwort auf: / As an answer to: WAC Falls into a manhole. Someone pulled the cover off geschrieben von: / posted by: Dann Corbit at 14 April 2000 22:29:18:

Gullydeckel 2 solved 240/300 on the WAC test at 5 seconds/position.
This, on a weak 300 MHz PII.
The interesting thing is the high NPS figures for some positions. While generally getting around 200K NPS it sometimes dips under 100K and sometimes goes way up. Consider this position:
+----+----+----+----+----+----+----+----+
| | *k | | | | | | |
+----+----+----+----+----+----+----+----+
| | | | | | r | p | |
+----+----+----+----+----+----+----+----+
| | p | | | | | | |
+----+----+----+----+----+----+----+----+
| | k | | | | | | |
+----+----+----+----+----+----+----+----+
| | | | | | | *r | |
+----+----+----+----+----+----+----+----+
| | | | | | | | |
+----+----+----+----+----+----+----+----+
| | | | | | | | |
+----+----+----+----+----+----+----+----+
| | | | | | | | |
+----+----+----+----+----+----+----+----+
pos: "WAC.041" Solution: Ka5 Kc5
1+ 0.40 0.00 1 f7f8
1& 2.68 0.00 2 f7f8
1& 2.69 0.00 14 b5c6
1& 2.71 0.00 17 b5c5
1& 3.02 0.00 26 b6b7
1. 3.02 0.00 26 b6b7
2& 3.02 0.00 41 b6b7 b8a7
2+ 3.42 0.00 87 f7f8
2& 6.17 0.00 109 f7f8 b8b7 g7g8q g4g8 f8g8
2. 6.17 0.00 109 f7f8 b8b7 g7g8q g4g8 f8g8
3& 6.17 0.00 174 f7f8 b8b7 g7g8q g4g8 f8g8
3. 6.17 0.00 535 f7f8 b8b7 g7g8q g4g8 f8g8
4- 5.77 0.02 3092 f7f8 b8b7 g7g8q g4g8 f8g8
4& 4.83 0.02 3503 f7f8 b8b7 g7g8q g4g5 b5c4 g5g8 f8g8 b7b6

4. 4.83 0.03 4856 f7f8 b8b7 g7g8q g4g5 b5c4 g5g8 f8g8 b7b6

5+ 5.23 0.04 6596 f7e7
5& 6.12 0.05 7705 f7e7 g4g5 b5a6 b8c8 e7e8 c8d7 g7g8q g5g8
e8g8
5. 6.12 0.07 14153 f7e7 g4g5 b5a6 b8c8 e7e8 c8d7 g7g8q g5g8
e8g8
6& 6.13 0.18 32524 b5a5 b8c8 f7f8 c8d7 g7g8q g4g8 f8g8
6. 6.13 0.19 36629 b5a5 b8c8 f7f8 c8d7 g7g8q g4g8 f8g8
7& 6.11 0.23 42678 b5a5 b8c8 f7f8 c8d7 g7g8q g4g8 f8g8 d7e6

7. 6.11 0.52 100159 b5a5 b8c8 f7f8 c8d7 g7g8q g4g8 f8g8 d7e6

8& 6.13 0.66 124865 b5a5 b8c8 f7f8 c8b7 g7g8q g4g8 f8g8 b7c6
g8f8
8. 6.13 1.26 239242 b5a5 b8c8 f7f8 c8b7 g7g8q g4g8 f8g8 b7c6
g8f8
9& 6.17 1.61 307788 b5a5 b8c8 f7f8 c8b7 g7g8q g4g8 f8g8 b7c6
g8g6 c6b7 a5b4
9. 6.17 3.26 623367 b5a5 b8c8 f7f8 c8b7 g7g8q g4g8 f8g8 b7c6
g8g6 c6b7 a5b4
10& 6.43 3.60 675562 b5a5 b8a8 f7f8 a8b7 g7g8q g4g8 f8g8 b7c6
g8g6 c6d5 b6b7 d5e4
10i 6.17 5.01 926592 b5a5 b8a8 f7f8 a8b7 g7g8q g4g8 f8g8 b7c6
g8g6 c6d5 b6b7 d5e4
game_phase: 3
evals:926592 (full:39142 lazy:95.78%), time: 5.01 [305896.21 nps]
Nodes in search: 605948 , quies: 926592 (60.46%)
moves in search - generated: 6180847 , looked at: 1792655 (29.00%)
pawn hash: hits: 39134 misses: 8 (99.980%)
Correct.
As you can see, G2 managed over 300K NPS.
It's quite an interesting patch of code! I have never seen the same mechanisms used before.

Hmmm, I can't tell what I'm looking at yet, here try this formatting thing:


Does the program work in Windows? Is that p problem with the nps being so different with different position? It doesn't seem like it should be.
Pete

[Dann Corbit]

Geschrieben von: / Posted by: Dann Corbit at 14 April 2000 23:40:52:
Als Antwort auf: / As an answer to: Re: WAC Falls into a manhole. Someone pulled the cover off geschrieben von: / posted by: Pete Galati at 14 April 2000 23:31:29:
[snip]

Does the program work in Windows?
Is that p problem with the nps being so different with different position?
It doesn't seem like it should be.

I am doing a Win32 port. The result set I posted was from a Win32 binary.
No, it's normal. The variation is larger than other engines I have seen. The interesting thing is that 300K NPS was achieved without using 64 bit bitboard data types. It's definitely (by a large margin) the highest NPS for a non-64 bit bitboard approach that I have ever seen.
[snip]


My FTP site