Skylake incompatible with some compilers ??

[Werner]

Hi,
when I make games with Nirvana Chess 2.2 x64 on my Skylake PC I get about 50% of the games with such funny moves like here. The same .exe makes no Problems on a an older I7. Any idea whats the reason?
best wishes
Werner

[Event "Nirvanana"]
[Site "I56600K"]
[Date "2015.12.27"]
[Round "4"]
[White "Nirvana Chess 2.2n x64 4CPU"]
[Black "Fritz 15 x64 4CPU"]
[Result "0-1"]
[BlackElo "2200"]
[ECO "D12"]
[Opening "Damengambit (Slawische Verteidigung)"]
[Time "16:21:57"]
[Variation "4.e3 Bf5 5.Nc3"]
[WhiteElo "2200"]
[TimeControl "40/480:40/480:40/480"]
[Termination "time forfeit"]
[PlyCount "128"]
[WhiteType "human"]
[BlackType "human"]

1. d4 Nf6 2. Bg5 d5 3. e3 g6 4. Bxf6 exf6 5. c4 dxc4 6. Bxc4 Bd6 7. Nc3 O-O
8. h4 h5 9. Nf3 {+0.30/21 8} Nd7 {-0.08/20 26} 10. Qc2 {+0.28/23 39} Kg7
{0.00/21 17} 11. O-O {+0.25/24 6} Nb6 {0.00/20 14} 12. Be2 {+0.22/25 16} c6
{+0.05/20 11} 13. Ne4 {+0.26/25 6} Bf5 {+0.04/21 16} 14. Bd3 {+0.29/23 7}
Bxe4 {+0.04/22 18} 15. Bxe4 {+0.19/24 7} f5 {+0.05/22 13} 16. Bd3 {+0.26/22
5} a5 {+0.04/22 9} 17. Rac1 {+0.14/22 9} Re8 {+0.04/22 16} 18. a3 {+0.18/22
9} a4 {+0.04/21 13} 19. Rfe1 {+0.10/22 13} Nd5 {+0.03/20 8} 20. Bc4
{+0.16/23 8} Re7 {+0.03/20 53} 21. g3 {+0.10/21 11} Nf6 {+0.04/19 22} 22.
Ng5 {+0.05/22 15} Qd7 {+0.04/18 8} 23. Qd1 {0.00/22 6} Ng4 {+0.16/19 14}
24. Qf3 {0.00/21 7} Rae8 {+0.16/18 6} 25. Kg2 {0.00/24 9} Qc7 {+0.16/18 10}
26. Re2 {0.00/24 4} Qb8 {+0.16/17 6} 27. Rc3 {+0.08/24 4} Qc8 {+0.14/18 14}
28. Rec2 {+0.08/23 5} Bc7 {+0.11/18 11} 29. Rc1 {+0.08/22 4} Qb8 {+0.10/18
17} 30. Rb1 {+0.08/22 6} Qa8 {+0.10/18 23} 31. Ra1 {+0.08/21 9} Ba5
{+0.11/18 11} 32. Rd3 {+0.04/23 8} Bd8 {+0.11/19 8} 33. Rc1 {0.00/21 8} Rd7
{+0.10/19 16} 34. Nh3 {0.00/22 4} Bc7 {+0.10/19 13} 35. Rdc3 {+0.08/22 4}
Qc8 {+0.10/19 17} 36. Ng5 {+0.08/20 5} Ree7 {+0.07/17 24} 37. R3c2
{+0.09/21 16} Rd8 {+0.09/18 8} 38. Kh1 {+0.08/20 3} Qd7 {+0.11/17 7} 39.
Qg2 {+0.08/20 3} Rde8 {+0.14/18 5} 40. Rd2 {+0.08/21 3} Qd6 {+0.14/17 2}
41. Nf3 {0.00/23 21} Nxe3 {+0.24/18 16} 42. fxe3 {-0.02/25 10} Rxe3
{+0.19/18 5} 43. Bd3 {-0.03/26 16} Ba5 {+0.24/17 8} 44. Re2 {-0.13/25 9} f4
{+0.39/20 18} 45. gxf4 {-0.39/23 8} Qxf4 {+0.41/20 7} 46. Rf1 {-0.40/25 7}
Bd8 {+0.41/20 9} 47. Rxe3 {-0.47/26 9} Rxe3 {+0.41/21 3} 48. Bc2 {-0.40/27
10} Bxh4 {+0.53/20 9} 49. Ne5 {-0.24/29 10} Bf2 {+0.58/20 11} 50. Nxg6
{-0.26/29 11} fxg6 {+0.58/20 2} 51. Qxg6+ {-0.33/31 11} Kf8 {+0.58/20 1}
52. Qf5+ {-0.33/31 17} Qxf5 {+0.59/18 2} 53. Bxf5 {-0.32/29 7} Rf3
{+0.64/23 12} 54. Bc8 {-0.42/29 9} Rf7 {+0.64/25 6} 55. d5 {-0.26/31 15}
cxd5 {+0.63/26 12} 56. Kg2 {-0.14/31 11} Be3 {+0.63/26 8} 57. Bxb7
{-0.12/31 14} d4 {+0.62/26 14} 58. Rxf7+ {+7.17/15 20} Kxf7 {+0.62/26 2}
59. Bd5+ {+M1/100 3} Kf6 {+0.53/27 12} 60. Kg3 {+95.05/19 36} Bc1 {+1.13/30
12} 61. Ba8 {+M1/100 5} Bxb2 {+2.14/27 4} 62. Bc6 {+M1/100 3} Bxa3
{+2.21/32 10} 63. Bg2 {+M1/16 19} Bb4 {+79.99/18 5} 64. Kh3 {+M1/25 7} a3
{+M11/31 14 Weiss überschreitet die Zeit} 0-1

[Ozymandias]

Mate evals are usual with Nirvana, which is why I don't use it for analysis. I hadn't seen those evals also creep out, during play.

[cetormenter]

Mate evals are usual with Nirvana, which is why I don't use it for analysis. I hadn't seen those evals also creep out, during play.

Really? Do you have an example position where you saw this behavior? I have never seen this when analyzing positions nor in any of the games that I have observed in Graham's tournaments. There was something that would cause incorrect values to be displayed during fail highs/lows but I think I fixed this before 2.1c. It should be documented in the change logs.

As for the main topic, I remember you had sent me an email about a month ago concerning this. I haven't been able to figure out any information about why it would not work on skylake processors. The only special instructions that I use in Nirvana are bsf/bsr and popcount.

[Sven]

Mate evals are usual with Nirvana, which is why I don't use it for analysis. I hadn't seen those evals also creep out, during play.

Really? Do you have an example position where you saw this behavior? I have never seen this when analyzing positions nor in any of the games that I have observed in Graham's tournaments. There was something that would cause incorrect values to be displayed during fail highs/lows but I think I fixed this before 2.1c. It should be documented in the change logs.

As for the main topic, I remember you had sent me an email about a month ago concerning this. I haven't been able to figure out any information about why it would not work on skylake processors. The only special instructions that I use in Nirvana are bsf/bsr and popcount.

Apart from the displayed values which are obviously wrong, the only faulty move is 63.Bg2? -+, where 63.Bxa4 would result in a TB draw. But that might have been a time problem?

[Werner]

As for the main topic, I remember you had sent me an email about a month ago concerning this. I haven't been able to figure out any information about why it would not work on skylake processors. The only special instructions that I use in Nirvana are bsf/bsr and popcount.

Yes, I have sent you an email longer time ago. I now made accidently another match on the skylake
...and here too: I am not able to see the same moves when I use analyze on these positions. I too tried the non popcorn version - not better. So perhaps there is something with the hash ?

best wishes

[Modern Times]

I too tried the non popcorn version - not better.

best wishes

Wish I had a popcorn version

[Werner]

I too tried the non popcorn version - not better.

best wishes

Wish I had a popcorn version

good spellcheck version

[Ozymandias]

Mate evals are usual with Nirvana, which is why I don't use it for analysis. I hadn't seen those evals also creep out, during play.

Really? Do you have an example position where you saw this behavior?

I would need to check my games from the last three CWTs, I don't recall exactly when I used Nirvana, I'll get back to you.