OpenBench Testing Framework

Finished
master
8.0+0.08
Elo: 2.96 +- 5.84 (95%) [N=4096]
Games: 4104 W: 1070 L: 1035 D: 1999
Ptnml(0-2): 36, 467, 1001, 522, 26
previous (suspiciously high?) progtest to master
3A8E36707A31D685
8.0+0.08
Elo: 46.32 +- 5.95 (95%) [N=4096]
Games: 4112 W: 1303 L: 758 D: 2051
Ptnml(0-2): 20, 305, 896, 780, 55
v12 to previous (suspiciously high?) progtest
master
8.0+0.08
Elo: 43.10 +- 6.22 (95%) [N=4096]
Games: 4116 W: 1351 L: 843 D: 1922
Ptnml(0-2): 27, 323, 910, 711, 87
v12.0 to master progtest
multilayer-ranger *
40.0+0.40
Elo: 4.47 +- 3.26 [0.00, 5.00]
Games: 10794 W: 2735 L: 2596 D: 5463
Ptnml(0-2): 17, 1194, 2837, 1331, 18
ml ranger
history-with-capacity
2.0+0.02
Elo: 5.57 +- 9.15 [0.00, 5.00]
Games: 1310 W: 336 L: 315 D: 659
Ptnml(0-2): 5, 123, 379, 142, 6
master
40.0+0.40
Elo: 11.64 +- 11.59 (95%) [N=1024]
Games: 866 W: 223 L: 194 D: 449
Ptnml(0-2): 2, 87, 225, 118, 1
Fix neon non-dotprod SIMD fallback for dpbusd vmull_s8 takes signed integers, so we need to preserve i8s through the neon code and u8s through non-neon. I've introduced UI8s for this case.
master
40.0+0.40
Elo: 58.80 +- 11.40 (95%) [N=1024]
Games: 1026 W: 360 L: 188 D: 478
Ptnml(0-2): 2, 64, 217, 220, 10
master
8.0+0.08
Elo: -3.31 +- 8.90 [-5.00, 0.00]
Games: 1576 W: 374 L: 389 D: 813
Ptnml(0-2): 12, 181, 407, 186, 2
master
40.0+0.40
Elo: 37.78 +- 15.02 (95%) [N=1024]
Games: 554 W: 174 L: 114 D: 266
Ptnml(0-2): 1, 43, 130, 101, 2
multilayer-ranger
N=25000
Elo: 8.60 +- 5.54 [0.00, 5.00]
Games: 6912 W: 2230 L: 2059 D: 2623
Ptnml(0-2): 174, 757, 1468, 838, 219
ml ranger
467F2C23BB07255C
8.0+0.08
Elo: -1.79 +- 6.71 [0.00, 5.00]
Games: 3106 W: 748 L: 764 D: 1594
Ptnml(0-2): 28, 382, 731, 402, 10
vectorsed-inference-int
40.0+0.40
Elo: 16.47 +- 7.08 [0.00, 5.00]
Games: 2322 W: 609 L: 499 D: 1214
Ptnml(0-2): 5, 217, 609, 323, 7
avx2 simd
master
40.0+0.40
Elo: 6.42 +- 11.05 (95%) [N=1024]
Games: 1028 W: 255 L: 236 D: 537
Ptnml(0-2): 1, 121, 255, 132, 5
master
40.0+0.40
Elo: 56.82 +- 10.93 (95%) [N=1024]
Games: 1024 W: 339 L: 173 D: 512
Ptnml(0-2): 0, 63, 227, 215, 7
continuous-best-move-tm
40.0+0.40
Elo: -4.62 +- 6.52 [-5.00, 0.00]
Games: 2556 W: 556 L: 590 D: 1410
Ptnml(0-2): 8, 292, 704, 274, 0
continuous best move tm
reintroduce-qsearch-lmp
8.0+0.08
Elo: -1.70 +- 3.60 [0.00, 5.00]
Games: 9610 W: 2243 L: 2290 D: 5077
Ptnml(0-2): 30, 1183, 2437, 1114, 41
Reintroduce qsearch lmp
qs-bad-tacticals-in-check *
8.0+0.08
Elo: 2.28 +- 3.69 [0.00, 5.00]
Games: 9298 W: 2227 L: 2166 D: 4905
Ptnml(0-2): 39, 1079, 2355, 1134, 42
search bad tacticals when in check in qsearch
master *
4.0+0.04
Elo: 2.92 +- 9.09 (95%) [N=40000]
Games: 1070 W: 274 L: 265 D: 531
Ptnml(0-2): 4, 81, 356, 90, 4
Revert "Add late move pruning to quiescence" This reverts commit bf4c5d4362044f474c54faadfaf63f6cf834e5e7. It appears that this introduced time losses.
simplify-tb-logic-in-search
N=5000
Elo: -0.00 +- 0.00 [0.00, 5.00]
Games: 3076 W: 1030 L: 1030 D: 1016
Ptnml(0-2): 0, 0, 1538, 0, 0
simplify tb logic in search cardinality logic is already in can_probe
qsearch-lmp *
8.0+0.08
Elo: 4.79 +- 3.45 [0.00, 5.00]
Games: 10656 W: 2590 L: 2443 D: 5623
Ptnml(0-2): 37, 1227, 2664, 1352, 48
qsearch lmp
scalar-inference-int
N=25000
Elo: 25.09 +- 10.06 [0.00, 5.00]
Games: 2136 W: 738 L: 584 D: 814
Ptnml(0-2): 47, 207, 434, 305, 75
scalar inference impl
scalar-inference-int
N=25000
Elo: 18.46 +- 8.57 [0.00, 5.00]
Games: 2770 W: 901 L: 754 D: 1115
Ptnml(0-2): 53, 288, 591, 365, 88
scalar inference impl
scalar-inference-int
N=25000
Elo: 16.53 +- 8.16 [0.00, 5.00]
Games: 3408 W: 1163 L: 1001 D: 1244
Ptnml(0-2): 101, 322, 723, 430, 128
scalar inference impl
qsearch-lmp *
8.0+0.08
Elo: -0.92 +- 3.15 [0.00, 5.00]
Games: 12878 W: 3017 L: 3051 D: 6810
Ptnml(0-2): 59, 1565, 3220, 1541, 54
qsearch lmp
fix-qsearch-evasions
8.0+0.08
Elo: 2.38 +- 2.42 [0.00, 5.00]
Games: 21432 W: 5144 L: 4997 D: 11291
Ptnml(0-2): 67, 2542, 5358, 2675, 74
alternate attempt