Re: [boost] Looking for some "real world" extended precision integer arithmetic tests

Subject: Re: [boost] Looking for some "real world" extended precision integer arithmetic tests
From: John Maddock (boost.regex_at_[hidden])
Date: 2012-03-21 13:17:18

Andrii,

Many thanks for the use case - in point of fact it's proved pretty hard to 
get a general purpose solution anywhere near your special-purpose code - 
simply because your code is so simple ;-)  In the general purpose case, even 
something as simple as an extra "if" statement can make a big difference in 
your benchmarks, largely because they mainly test construction of 
temporaries!

Anyhow, I think I'm as close as it's going to get for now, here it is on 
Linux (sorry about the mangled names, but your integer type is first):

Testing... N5boost7polygon6detail12extended_intILm32EEE
Voronoi Benchmark Test (time in seconds):
| Number of points | Number of tests | Time per one test |
|               10 |           10000 |          0.000035 |
|              100 |            1000 |          0.000620 |
|             1000 |             100 |          0.006900 |
|            10000 |              10 |          0.087000 |
|           100000 |               1 |          0.880000 |
Testing... 
N5boost14multiprecision9mp_numberINS0_8backends15cpp_int_backendILj256ELb1EvEELb1EEE
Voronoi Benchmark Test (time in seconds):
| Number of points | Number of tests | Time per one test |
|               10 |           10000 |          0.000041 |
|              100 |            1000 |          0.000720 |
|             1000 |             100 |          0.008200 |
|            10000 |              10 |          0.097000 |
|           100000 |               1 |          1.010000 |
Testing... 
N5boost14multiprecision9mp_numberINS0_8backends15cpp_int_backendILj1024ELb1EvEELb1EEE
Voronoi Benchmark Test (time in seconds):
| Number of points | Number of tests | Time per one test |
|               10 |           10000 |          0.000041 |
|              100 |            1000 |          0.000730 |
|             1000 |             100 |          0.008100 |
|            10000 |              10 |          0.097000 |
|           100000 |               1 |          1.000000 |
Testing... 
N5boost14multiprecision9mp_numberINS0_8backends15cpp_int_backendILj0ELb1ESaImEEELb1EEE
Voronoi Benchmark Test (time in seconds):
| Number of points | Number of tests | Time per one test |
|               10 |           10000 |          0.000052 |
|              100 |            1000 |          0.000930 |
|             1000 |             100 |          0.010300 |
|            10000 |              10 |          0.121000 |
|           100000 |               1 |          1.240000 |
Testing... N5boost14multiprecision9mp_numberINS0_8backends7gmp_intELb1EEE
Voronoi Benchmark Test (time in seconds):
| Number of points | Number of tests | Time per one test |
|               10 |           10000 |          0.000075 |
|              100 |            1000 |          0.001400 |
|             1000 |             100 |          0.015500 |
|            10000 |              10 |          0.174000 |
|           100000 |               1 |          1.780000 |
Testing... 
N5boost14multiprecision9mp_numberINS0_8backends11tommath_intELb1EEE
Voronoi Benchmark Test (time in seconds):
| Number of points | Number of tests | Time per one test |
|               10 |           10000 |          0.001447 |
|              100 |            1000 |          0.026940 |
|             1000 |             100 |          0.269500 |
|            10000 |              10 |          2.590000 |
|           100000 |               1 |         24.870000 |

On Windows, I can't get quite as close, and MPIR on Win32 is noticeably much 
much worse than GMP on Linux x64:

Testing... class boost::polygon::detail::extended_int<32>
Voronoi Benchmark Test (time in seconds):
| Number of points | Number of tests | Time per one test |
|               10 |           10000 |          0.000072 |
|              100 |            1000 |          0.001117 |
|             1000 |             100 |          0.012220 |
|            10000 |              10 |          0.134600 |
|           100000 |               1 |          1.377000 |
Testing... class boost::multiprecision::mp_number<struct 
boost::multiprecision::backends::cpp_int_backend<1024,1,void>,0>
Voronoi Benchmark Test (time in seconds):
| Number of points | Number of tests | Time per one test |
|               10 |           10000 |          0.000094 |
|              100 |            1000 |          0.001601 |
|             1000 |             100 |          0.017600 |
|            10000 |              10 |          0.187100 |
|           100000 |               1 |          1.893000 |
Testing... class boost::multiprecision::mp_number<struct 
boost::multiprecision::backends::cpp_int_backend<1024,1,void>,1>
Voronoi Benchmark Test (time in seconds):
| Number of points | Number of tests | Time per one test |
|               10 |           10000 |          0.000104 |
|              100 |            1000 |          0.001693 |
|             1000 |             100 |          0.018740 |
|            10000 |              10 |          0.196900 |
|           100000 |               1 |          1.988000 |
Testing... class boost::multiprecision::mp_number<struct 
boost::multiprecision::backends::cpp_int_backend<0,1,class 
std::allocator<unsigned int> >,1>
Voronoi Benchmark Test (time in seconds):
| Number of points | Number of tests | Time per one test |
|               10 |           10000 |          0.000126 |
|              100 |            1000 |          0.002210 |
|             1000 |             100 |          0.024230 |
|            10000 |              10 |          0.254400 |
|           100000 |               1 |          2.563000 |
Testing... class boost::multiprecision::mp_number<struct 
boost::multiprecision::backends::gmp_int,1>
Voronoi Benchmark Test (time in seconds):
| Number of points | Number of tests | Time per one test |
|               10 |           10000 |          0.000392 |
|              100 |            1000 |          0.007060 |
|             1000 |             100 |          0.078010 |
|            10000 |              10 |          0.803300 |
|           100000 |               1 |          8.080000 |

Cheers, John.