Dev Builds » 20210730-1515

Use this dev build

NCM plays each Stockfish dev build 20,000 times against Stockfish 15. This yields an approximate Elo difference and establishes confidence in the strength of the dev builds.

Summary

Host Duration Avg Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo
ncm-dbt-01 06:54:58 583698 4000 518 1519 1963 -88.83 ± 5.09 29 1056 803 111 1 -184.65 ± 12.01
ncm-dbt-02 06:55:25 586956 4000 515 1529 1956 -90.04 ± 5.09 29 1069 789 113 0 -187.39 ± 12.12
ncm-dbt-03 06:54:43 585412 4000 489 1540 1971 -93.48 ± 5.0 26 1098 777 99 0 -196.69 ± 12.21
ncm-dbt-04 06:55:11 567188 4000 521 1492 1987 -86.06 ± 5.14 27 1040 814 115 4 -179.0 ± 11.93
ncm-dbt-05 06:56:42 582460 4000 568 1530 1902 -85.23 ± 5.12 29 1026 824 120 1 -175.88 ± 11.85
20000 2611 7610 9779 -88.72 ± 2.28 140 5289 4007 558 6 -184.65 ± 5.37

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN
436824 ncm-dbt-01 580489 500 70 190 240 -85.04 ± 13.75 3 125 111 11 0 -176.33 ± 32.16 ↓
436823 ncm-dbt-04 569669 500 55 190 255 -96.19 ± 14.85 4 142 90 13 1 -204.07 ± 36.13 ↓
436822 ncm-dbt-02 586308 500 60 179 261 -84.3 ± 14.76 5 125 104 16 0 -171.02 ± 33.51 ↓
436821 ncm-dbt-05 583866 500 55 190 255 -96.19 ± 15.0 7 135 94 14 0 -196.45 ± 35.33 ↓
436820 ncm-dbt-03 583698 500 56 184 260 -90.96 ± 13.4 1 136 103 10 0 -194.57 ± 33.53 ↓
436819 ncm-dbt-01 585590 500 60 183 257 -87.26 ± 14.69 4 131 99 16 0 -179.9 ± 34.4 ↓
436818 ncm-dbt-04 569390 500 63 184 253 -85.78 ± 14.65 3 132 98 17 0 -178.11 ± 34.58 ↓
436817 ncm-dbt-05 585042 500 77 190 233 -79.9 ± 14.21 3 122 110 15 0 -164.07 ± 32.48 ↓
436816 ncm-dbt-03 586181 500 68 193 239 -88.74 ± 14.28 5 127 106 12 0 -181.7 ± 33.07 ↓
436815 ncm-dbt-02 587622 500 60 199 241 -99.2 ± 14.6 4 145 87 14 0 -209.91 ± 36.75 ↓
436814 ncm-dbt-01 581693 500 78 179 243 -71.16 ± 15.32 4 117 105 24 0 -142.26 ± 33.37 ↓
436813 ncm-dbt-05 583824 500 75 202 223 -90.22 ± 14.6 6 128 103 13 0 -183.51 ± 33.63 ↓
436812 ncm-dbt-03 586054 500 47 199 254 -109.07 ± 12.53 3 148 97 2 0 -238.66 ± 34.07 ↓
436811 ncm-dbt-04 565469 500 65 182 253 -82.83 ± 14.58 4 125 105 16 0 -169.27 ± 33.34 ↓
436810 ncm-dbt-02 587877 500 58 187 255 -91.71 ± 13.42 2 134 105 9 0 -194.57 ± 33.11 ↓
436809 ncm-dbt-01 582778 500 60 190 250 -92.46 ± 13.9 1 140 98 10 1 -200.24 ± 34.5 ↓
436808 ncm-dbt-04 568236 500 71 193 236 -86.52 ± 13.32 3 124 115 8 0 -179.9 ± 31.32 ↓
436807 ncm-dbt-02 585843 500 67 204 229 -97.69 ± 13.5 2 142 97 9 0 -209.91 ± 34.63 ↓
436806 ncm-dbt-05 583866 500 66 185 249 -84.31 ± 14.76 0 139 92 18 1 -181.7 ± 35.67 ↓
436805 ncm-dbt-03 586308 500 61 200 239 -99.2 ± 13.83 4 140 97 9 0 -209.91 ± 34.63 ↓
436804 ncm-dbt-01 585632 500 66 193 241 -90.22 ± 14.16 3 134 100 13 0 -189.0 ± 34.18 ↓
436803 ncm-dbt-04 567799 500 64 170 266 -74.79 ± 15.04 3 121 106 19 1 -153.86 ± 33.21 ↓
436802 ncm-dbt-02 586943 500 71 179 250 -76.25 ± 14.95 3 123 103 21 0 -155.54 ± 33.71 ↓
436801 ncm-dbt-03 585548 500 59 205 236 -104.49 ± 14.04 3 151 85 11 0 -226.0 ± 37.21 ↓
436800 ncm-dbt-05 582861 500 69 195 236 -89.48 ± 13.53 1 135 103 11 0 -190.85 ± 33.57 ↓
436799 ncm-dbt-01 585590 500 61 196 243 -96.19 ± 15.0 7 135 94 14 0 -196.45 ± 35.33 ↓
436798 ncm-dbt-04 561292 500 72 193 235 -85.78 ± 13.77 2 129 107 12 0 -179.9 ± 32.89 ↓
436797 ncm-dbt-03 585801 500 64 185 251 -85.78 ± 15.08 5 129 98 18 0 -174.55 ± 34.58 ↓
436796 ncm-dbt-05 581818 500 74 196 230 -86.52 ± 13.47 2 128 110 10 0 -181.7 ± 32.28 ↓
436795 ncm-dbt-02 588004 500 64 202 234 -98.44 ± 15.32 8 137 90 15 0 -200.24 ± 36.12 ↓
436794 ncm-dbt-01 583572 500 60 192 248 -93.95 ± 13.61 2 138 100 10 0 -200.24 ± 34.09 ↓
436793 ncm-dbt-04 566060 500 74 192 234 -83.57 ± 15.43 4 131 95 19 1 -172.78 ± 35.1 ↓
436792 ncm-dbt-03 586012 500 71 187 242 -82.1 ± 14.7 3 128 101 18 0 -169.27 ± 34.05 ↓
436791 ncm-dbt-05 582318 500 81 187 232 -74.79 ± 15.96 5 123 95 27 0 -148.85 ± 34.96 ↓
436790 ncm-dbt-02 584201 500 57 183 260 -89.48 ± 14.59 3 136 95 16 0 -187.16 ± 35.14 ↓
436789 ncm-dbt-01 584243 500 63 196 241 -94.69 ± 14.54 5 136 96 13 0 -196.45 ± 34.93 ↓
436788 ncm-dbt-04 569589 500 57 188 255 -93.2 ± 14.51 4 136 98 11 1 -196.45 ± 34.53 ↓
436787 ncm-dbt-03 583698 500 63 187 250 -88.0 ± 14.85 2 139 90 19 0 -185.33 ± 36.06 ↓
436786 ncm-dbt-02 588856 500 78 196 226 -83.57 ± 13.87 2 127 108 13 0 -174.55 ± 32.75 ↓
436785 ncm-dbt-05 576088 500 71 185 244 -80.63 ± 14.09 5 116 117 12 0 -162.35 ± 31.23 ↓

Commit

Commit ID 26edf9534ad571a6d26bf9db47d21776cbf45d54
Author Tomasz Sobczyk
Date 2021-07-30 15:15:52 UTC
Avoid unnecessary stores in the affine transform This patch improves the codegen in the AffineTransform::forward function for architectures >=SSSE3. Current code works directly on memory and the compiler cannot see that the stores through outptr do not alias the loads through weights and input32. The solution implemented is to perform the affine transform with local variables as accumulators and only store the result to memory at the end. The number of accumulators required is OutputDimensions / OutputSimdWidth, which means that for the 1024->16 affine transform it requires 4 registers with SSSE3, 2 with AVX2, 1 with AVX512. It also cuts the number of stores required by NumRegs * 256 for each node evaluated. The local accumulators are expected to be assigned to registers, but even if this cannot be done in some case due to register pressure it will help the compiler to see that there is no aliasing between the loads and stores and may still result in better codegen. See https://godbolt.org/z/59aTKbbYc for codegen comparison. passed STC: LLR: 2.94 (-2.94,2.94) <-0.50,2.50> Total: 140328 W: 10635 L: 10358 D: 119335 Ptnml(0-2): 302, 8339, 52636, 8554, 333 closes https://github.com/official-stockfish/Stockfish/pull/3634 No functional change
Copyright 2011–2025 Next Chess Move LLC
Feedback & Support Terms of Service Privacy Policy Desktop Version