I did some quick benchmarks using some older version of Brian's benchmark code (we have a new version coming with more tests). The penryn machine was heavily loaded, so I don't trust the values very much, especially the pi ones, though I did all the benchmarks under identical conditions. The K10 ones should be more accurate:
penryn: Squaring: GMP 4.3.0 MPIR 1.2.0 MPIR 1.3.0 ========== ======== ========= ========== 128 x 128 : 50993028 51293922 51666716 512 x 512 : 9275740 7974568 8009900 8192 x 8192 : 98217 99896 100503 131072 x 131072 : 1537 1702 1724 2097152 x 2097152 : 68.2 79.5 80.6 Multiplication: ========== 128 x 128 : 50993028 51666716 51081525 512 x 512 : 7334238 7970325 8001954 8192 x 8192 : 70880 74547 74547 131072 x 131072 : 1159 1286 1294 2097152 x 2097152 : 47.3 55.9 57.3 Unbalanced: ========== 15000 x 10000 : 34617 36381 35832 20000 x 10000 : 26373 28333 27913 30000 x 10000 : 15672 16786 16804 16777216 x 512 : 223 232 233 16777216 x 262144 : 8.96 10.7 10.8 Division : ========= 8192 / 32 : 800782 672164 928075 8192 / 64 : 795968 683094 680328 8192 / 128 : 526394 365184 367509 8192 / 4096 : 119806 115299 114019 8192 / 8064 : 1646375 1618972 6714754 131072 / 65536 : 1373 1372 1393 8388608 / 4194304 : 4.03 4.40 4.47 16777216 / 262144 : 2.69 2.79 2.81 GCD : ==== 128 x 128 : 2229116 2131425 2191565 512 x 512 : 230145 156351 214626 8192 x 8192 : 5852 5453 5648 131072 x 131072 : 88.8 89.0 91.8 1048576 x 1048576 : 4.17 4.34 4.59 XGCD : ===== 128 x 128 : 991393 694605 909401 512 x 512 : 179531 110886 148946 8192 x 8192 : 3700 2438 3428 131072 x 131072 : 52.1 51.4 54.5 1048576 x 1048576 : 2.79 2.66 2.90 RSA : ==== 512 : 16190 15020 15019 1024 : 3021 3128 3112 2048 : 477 475 478 Pi : === 10000 : 328 332 335 100000 : 16.7 18.0 18.8 1000000 : 1.01 1.13 1.16 Overall : 1489 1460 1565 ====================================== K10: Squaring: GMP 4.3.0 MPIR 1.2.0 MPIR 1.3.0 ========== ======== ========= ========== 128 x 128 : 52219920 55564735 55553695 512 x 512 : 12321692 12420943 12148952 8192 x 8192 : 159844 161622 162057 131072 x 131072 : 2483 2689 2714 2097152 x 2097152 : 88.8 99.8 101 Multiplication: ========== 128 x 128 : 52219920 55605250 55505909 512 x 512 : 10656260 12418369 12251151 8192 x 8192 : 108145 112543 111985 131072 x 131072 : 1712 2022 2012 2097152 x 2097152 : 58.2 68.0 68.3 Unbalanced: ========== 15000 x 10000 : 53859 56976 54998 20000 x 10000 : 41146 44883 43507 30000 x 10000 : 23808 25990 25729 16777216 x 512 : 418 363 358 16777216 x 262144 : 10.9 11.4 11.4 Division : ========= 8192 / 32 : 1417991 1245813 1537953 8192 / 64 : 1432755 1247058 1248305 8192 / 128 : 871200 674187 678989 8192 / 4096 : 170418 178373 194700 8192 / 8064 : 2284081 2109811 7527466 131072 / 65536 : 2056 2138 2503 8388608 / 4194304 : 5.31 6.16 3.42 16777216 / 262144 : 3.93 4.33 4.41 GCD : ==== 128 x 128 : 2060071 1889462 1878091 512 x 512 : 235945 181500 195688 8192 x 8192 : 7351 6591 6551 131072 x 131072 : 129 128 129 1048576 x 1048576 : 5.72 6.25 6.51 XGCD : ===== 128 x 128 : 918836 396791 661980 512 x 512 : 176947 76129 135265 8192 x 8192 : 5030 3021 4395 131072 x 131072 : 79.6 77.0 81.6 1048576 x 1048576 : 3.69 3.90 4.26 RSA : ==== 512 : 19198 21685 21623 1024 : 4048 4750 4768 2048 : 752 823 822 Pi : === 10000 : 635 656 664 100000 : 27.7 31.6 32.7 1000000 : 1.43 1.71 1.75 Overall : 1420 1449 1539 I guess the first order of business will be to work out why the large speed regression in the 8388608 / 4194304 division on K10 (but not core2). Everything else is looking good though. We need to retune for the release, so these figures are preliminary. Bill. --~--~---------~--~----~------------~-------~--~----~ You received this message because you are subscribed to the Google Groups "mpir-devel" group. To post to this group, send email to mpir-devel@googlegroups.com To unsubscribe from this group, send email to mpir-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/mpir-devel?hl=en -~----------~----~----~----~------~----~------~--~---
