Simple MRG speed tests using Vigna's harness. This is here mostly to share
between computers and test on different architectures. Calling make runs all
the generators for 10^9 numbers 10 times, averages the times and prints the
results. Check parameters before runing, numbers can be a bit high.
Example usages:
make
make mrg59k3a.res
./compile.sh mrg59k3a 1000000 5
Results for different processors and architectures (average of 10 runs for 10^9 numbers, Raspberry Pi uses 10*average for 10^8 numbers):
| CPU | OS | Architecture | Compiler | Misc. | lcg16 | mrg16_1 | mrg16_2 | mrg16_3 | mrg8 | mrg59k3a | mrg59p2a | mrg59p2b | mrg59p3c | mrg31k3p_original | mrg31k3p_fast | mrg32k3a_double | mrg32k3a_vigna |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| i5-7200U | Linux 5.3.15 | x86_64 (64 bits) | gcc 9.2.0 | - | 1.650 | 12.827 | 6.980 | 9.900 | 6.123 | 10.495 | 11.694 | 14.286 | 16.553 | 12.956 | 7.057 | 28.692 | 5.419 |
| i7-2600 | Linux 4.9.0 | x86_64 (64 bits) | 6.3.0 | - | 1.469 | 14.476 | 7.462 | 9.122 | 6.263 | 12.235 | 12.048 | 14.803 | 17.459 | 11.797 | 7.560 | 25.574 | 6.067 |
| SDM845 | Linux 4.9.165 | aarch64 (64 bits) | clang 8.0.1 | Termux on Pixel 3 | 1.513 | 13.459 | 7.150 | 12.983 | 8.872 | 25.210 | 13.720 | 21.636 | 26.385 | 8.873 | 9.380 | 19.813 | 9.994 |
| BCM2835 | Linux 4.14.98 | armv7l (32-bits) | gcc 6.3.0 | Raspberry Pi 3 B | 15.934 | 50.292 | 28.481 | 34.342 | 31.013 | 1521.133 | 345.154 | 373.797 | 425.014 | 148.653 | 202.937 | 181.729 | 513.926 |