This library is based on the XorShift128+ algorithm from Sebastiano Vigna. See:
- https://vigna.di.unimi.it/ftp/papers/xorshiftplus.pdf
- https://vigna.di.unimi.it/xorshift/xorshift128plus.c
- https://en.wikipedia.org/wiki/Xorshift#xorshift+
It uses a 128-bit state and returns 64-bit uniformly distributed numbers.
The rand_workspace struct (typedef'd) is defined as follows:
typedef struct{ uint64_t s[2]; } rand_workspace;
s is the seed of the RNG. The following functions are available:
-
rand_workspace *RandAlloc();Allocates a rand_workspace using malloc. -
void SetRandomSeed(rand_workspace *w);Randomly set the seed based on current CPU clocks counter. It should give different enough values even if called two times consecutively. -
double RSUnif(rand_workspace *w, double a, double b);Generate one random number uniformely distributed between a and b. a is meant to be < b but the function works also in the opposite case. -
double RSGauss(rand_workspace *w, double m, double s);Generate one random number following a Gaussian distribution with mean m and standard deviation s. This function uses the Box-Mueller algorithm. Here s is not required to be positive (the absolute value is used). -
double RSExpo(rand_workspace *w, double lambda);Generate one random number following an exponential distribution with parameter lambda. -
double RSCauchy(rand_workspace *w, double mean, double gamma);Generate one random number following a Cauchy distribution with parameters mean and gamma. gamma can be also negative.
A simple example is provided in the tests folder.
Before using the generator, allocates the workspace with RandAlloc or just instantiate one rand_workspace variable. This generator is thread-safe: allocate (or instantiate) one workspace for each thread. Apply SetRandomSeed to each one. At this point it is possible to generate numbers. If workspaces were allocated, remember to free memory before the end of your program (just use free()).