Dev Builds » 20210730-1515

You are viewing an old NCM Stockfish dev build test. You may find the most recent dev build tests using Stockfish 15 as the baseline here.

Use this dev build

NCM plays each Stockfish dev build 20,000 times against Stockfish 14. 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 11:57:55 1227258 3984 1019 937 2028 +7.15 ± 5.36 2 440 1029 516 5 +13.79 ± 10.61
ncm-dbt-02 11:53:30 1261752 3924 986 906 2032 +7.08 ± 5.26 1 414 1054 490 3 +13.82 ± 10.46
ncm-dbt-03 11:56:39 1269043 3980 1005 904 2071 +8.82 ± 5.37 6 420 1036 523 5 +17.82 ± 10.57
ncm-dbt-04 00:30:20 1481711 168 40 40 88 0.0 ± 25.68 0 20 44 20 0 -0.0 ± 51.65
ncm-dbt-05 11:53:36 1258668 3960 1016 910 2034 +9.3 ± 5.32 2 416 1041 516 5 +18.09 ± 10.54
ncm-dbt-06 11:57:09 1254938 3984 1048 915 2021 +11.6 ± 5.31 3 406 1042 537 4 +23.06 ± 10.54
20000 5114 4612 10274 +8.72 ± 2.37 14 2116 5246 2602 22 +17.18 ± 4.69

Test Detail

ID Host Base NPS Games WLD Standard Elo Ptnml(0-2) Gamepair Elo CLI PGN
209949 ncm-dbt-02 1237178 260 68 60 132 +10.69 ± 20.25 0 26 70 34 0 +21.41 ± 40.75 ↓
209948 ncm-dbt-05 1205181 296 76 65 155 +12.92 ± 19.04 0 29 79 40 0 +25.87 ± 38.36 ↓
209947 ncm-dbt-03 1256869 310 78 80 152 -2.24 ± 19.67 1 37 81 35 1 -4.48 ± 37.94 ↓
209946 ncm-dbt-01 1190920 316 69 76 171 -7.7 ± 18.9 0 41 84 32 1 -17.61 ± 37.21 ↓
209945 ncm-dbt-06 1215990 318 96 81 141 +16.4 ± 18.92 0 30 86 41 2 +28.47 ± 36.71 ↓
209944 ncm-dbt-02 1238049 500 122 106 272 +11.12 ± 14.75 1 48 135 66 0 +23.66 ± 29.27 ↓
209943 ncm-dbt-05 1231370 500 143 115 242 +19.48 ± 14.65 0 45 132 73 0 +39.08 ± 29.64 ↓
209942 ncm-dbt-03 1227550 500 113 123 264 -6.95 ± 15.16 1 64 129 56 0 -12.51 ± 30.04 ↓
209941 ncm-dbt-01 1216619 500 113 126 261 -9.04 ± 15.21 0 68 128 53 1 -19.48 ± 30.16 ↓
209940 ncm-dbt-06 1248970 500 117 107 276 +6.95 ± 15.52 0 60 120 70 0 +13.9 ± 31.14 ↓
209939 ncm-dbt-02 1242264 500 122 124 254 -1.39 ± 13.9 0 53 146 51 0 -2.78 ± 27.84 ↓
209938 ncm-dbt-03 1240517 500 133 102 265 +21.57 ± 14.55 0 43 133 74 0 +43.3 ± 29.5 ↓
209937 ncm-dbt-05 1226095 500 132 114 254 +12.51 ± 14.99 1 49 131 69 0 +26.46 ± 29.78 ↓
209936 ncm-dbt-01 1203608 500 144 115 241 +20.17 ± 16.04 0 55 112 82 1 +39.08 ± 32.13 ↓
209935 ncm-dbt-06 1220747 500 125 111 264 +9.73 ± 14.77 0 51 135 63 1 +18.08 ± 29.27 ↓
209934 ncm-dbt-02 1227568 500 123 119 258 +2.78 ± 15.17 0 58 132 58 2 +2.78 ± 29.66 ↓
209933 ncm-dbt-03 1221347 500 127 112 261 +10.43 ± 15.07 2 48 133 67 0 +23.66 ± 29.53 ↓
209932 ncm-dbt-05 1246824 500 118 115 267 +2.09 ± 14.99 1 56 132 61 0 +5.56 ± 29.66 ↓
209931 ncm-dbt-01 1176062 500 140 128 232 +8.34 ± 15.51 1 55 126 67 1 +16.69 ± 30.41 ↓
209930 ncm-dbt-06 1222744 500 137 116 247 +14.6 ± 15.4 0 53 124 72 1 +27.85 ± 30.66 ↓
209929 ncm-dbt-02 1232970 500 130 118 252 +8.34 ± 14.65 0 52 134 64 0 +16.69 ± 29.4 ↓
209928 ncm-dbt-03 1246178 500 128 121 251 +4.86 ± 16.06 1 61 120 66 2 +8.34 ± 31.14 ↓
209927 ncm-dbt-01 1197584 500 127 127 246 -0.0 ± 14.68 0 58 134 58 0 -0.0 ± 29.41 ↓
209926 ncm-dbt-05 1225875 500 130 114 256 +11.12 ± 15.61 0 57 121 71 1 +20.87 ± 31.02 ↓
209925 ncm-dbt-06 1213525 500 127 136 237 -6.25 ± 14.34 0 60 139 51 0 -12.51 ± 28.76 ↓
209924 ncm-dbt-02 1244902 500 116 114 270 +1.39 ± 14.93 0 58 133 58 1 +1.39 ± 29.53 ↓
209923 ncm-dbt-01 1193296 500 129 118 253 +7.64 ± 14.59 0 52 135 63 0 +15.3 ± 29.27 ↓
209922 ncm-dbt-03 1223831 500 123 113 264 +6.95 ± 14.4 0 51 138 61 0 +13.9 ± 28.89 ↓
209921 ncm-dbt-05 1228463 500 121 115 264 +4.17 ± 14.8 0 56 132 62 0 +8.34 ± 29.66 ↓
209920 ncm-dbt-06 1244590 500 129 114 257 +10.43 ± 14.95 1 50 132 67 0 +22.27 ± 29.65 ↓
209919 ncm-dbt-02 1203353 500 131 112 257 +13.21 ± 14.67 0 49 133 68 0 +26.46 ± 29.52 ↓
209918 ncm-dbt-03 1257890 500 124 102 274 +15.3 ± 15.09 0 50 129 70 1 +29.25 ± 30.03 ↓
209917 ncm-dbt-01 1186847 500 133 114 253 +13.21 ± 15.53 0 55 122 72 1 +25.06 ± 30.9 ↓
209916 ncm-dbt-06 1208070 500 127 104 269 +15.99 ± 14.64 1 44 136 69 0 +33.46 ± 29.13 ↓
209915 ncm-dbt-05 1215134 500 118 116 266 +1.39 ± 14.55 0 54 142 52 2 -0.0 ± 28.37 ↓
209914 ncm-dbt-03 1242651 500 139 102 259 +25.76 ± 15.11 1 41 129 78 1 +51.8 ± 30.01 ↓
209913 ncm-dbt-05 1257998 500 135 112 253 +15.99 ± 15.14 0 49 131 68 2 +29.25 ± 29.78 ↓
209912 ncm-dbt-02 1237017 500 130 115 255 +10.43 ± 14.82 0 52 131 67 0 +20.87 ± 29.78 ↓
209911 ncm-dbt-06 1222187 500 137 103 260 +23.66 ± 14.71 1 40 133 76 0 +48.96 ± 29.49 ↓
209910 ncm-dbt-01 1182966 500 120 99 281 +14.6 ± 14.0 0 43 143 64 0 +29.25 ± 28.2 ↓
178054 ncm-dbt-02 1492474 164 44 38 82 +12.72 ± 26.86 0 18 40 24 0 +25.47 ± 54.27 ↓
178053 ncm-dbt-06 1497627 166 53 43 70 +20.95 ± 27.62 0 18 37 28 0 +42.06 ± 56.23 ↓
178052 ncm-dbt-04 1481711 168 40 40 88 0.0 ± 25.68 0 20 44 20 0 -0.0 ± 51.65 ↓
178051 ncm-dbt-05 1491076 164 43 44 77 -2.12 ± 26.64 0 21 41 20 0 -4.24 ± 53.6 ↓
178050 ncm-dbt-03 1504558 170 40 49 81 -18.41 ± 25.47 0 25 44 16 0 -36.93 ± 51.66 ↓
178049 ncm-dbt-01 1497426 168 44 34 90 +20.71 ± 26.04 1 13 45 25 0 +45.76 ± 50.93 ↓

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 2011–2024 Next Chess Move LLC
Feedback & Support Terms of Service Privacy Policy Desktop Version