fastPRNG

fastPRNG is a single header-only FAST 32/64 bit PRNG (pseudo-random generator), highly optimized to obtain faster code from compilers, it's based on xoshiro / xoroshiro (Blackman/Vigna), xorshift and other Marsaglia algorithms.

64bit algorithms

• Blackman/Vigna
  • xoshiro256+ / xoshiro256++ / xoshiro256**
  • xoroshiro128+ / xoroshiro128++ / xoroshiro128**
• Marsaglia
  • xorshift
  • znew / wnew / MWC / CNG / FIB / XSH / KISS

32bit algorithms

• Blackman/Vigna
  • xoshiro128+ / xoshiro128++ / xoshiro128**
  • xoroshiro64+ / xoroshiro64++
• Marsaglia
  • xorshift
  • znew / wnew / MWC / CNG / FIB / XSH / KISS
  • LFIB4 / SWB

 

fastPRNG distribution tests - live WebGL

All functions are tested, below the distribution test screenshots in a cube with [-1.0, 1.0] side (OpenGL/WebGL)

30M dots/spheres	30M dots/spheres clipping planes	3 thin boards from 30M cube dots/spheres

==>  *view the Live WebGL distribution test section.

 

Return values and floating point

All base functions return integers:

• 32bit ==> uint32_t in [0, UINT32_MAX] interval
• 64bit ==> uint64_t in [0, UINT64_MAX] interval.

*If you need (e.g.) values between [INT32_MIN, INT32_MAX], just cast result to int32_t, same for 64bit (cast to int64_t): look at the examples below

Floating point helpers

• There are a single/double precision floating point template functions, to generate fast numbers in:
    (they have same base name, but with the following suffix)
  • [-1.0, 1.0] interval ==> suffix _VNI<T> (Vector Normalized Interval)
  • [ 0.0, 1.0] interval ==> suffix _UNI<T> (Unity Normalized Interval)
  • [ min, max] interval ==> suffix _Range<T>(min, max)

*look at the examples below

 

How to use - Examples

To use it just include fastPRNG.h in your code:

#include "fastPRNG.h"

It contains following classes and member functions, inside the namespace fastPRNG:

64bit classes and members

• fastXS64 ==> contains xor-shift algorithms
  • xoshiro256p / xoshiro256pp / xoshiro256xx
  • xoroshiro128p / xoroshiro128pp / xoroshiro128xx
  • xorShift
• fastXS64s ==> same as before, but class with static members, to use directly w/o declaration
  • fastXS64s::xoshiro256p / fastXS64s::xoshiro256pp / fastXS64s::xoshiro256xx
  • fastXS64s::xoroshiro128p / fastXS64s::xoroshiro128pp / fastXS64s::xoroshiro128xx
  • fastXS64s::xorShift
• fastRand64 ==> other Marsaglia algorithms
  • znew / wnew / MWC / CNG / FIB / XSH / KISS

32bit classes and members

• fastXS32 ==> contains xor-shift algorithms
  • xoshiro128p / xoshiro128pp / xoshiro128xx
  • xoroshiro64p / xoroshiro64pp
  • xorShift
• fastXS32s ==> same as before, but class with static members, to use directly w/o declaration
  • fastXS32s::xoshiro128p / fastXS32s::xoshiro128pp / fastXS32s::xoshiro128xx
  • fastXS32s::xoroshiro64p / fastXS32s::xoroshiro64pp
  • fastXS32s::xorShift
• fastRand32 ==> other Marsaglia algorithms
  • znew / wnew / MWC / CNG / FIB / XSH / KISS
  • LFIB4 / SWB

Examples

• Example: use xoshiro256+ 64bit algorithm:

    using fastPRNG;

    fastXS64 fastR; // default "chrono" seed
//  fastXS64 fastR(0x123456789ABCDEF0); // personal seed also to (re)generate a specific random numbers sequence

    for(int i=0; i<10000; i++) {
        cout <<         fastR.xoshiro256p()  << endl;    // returns number in [0, UINT64_MAX] interval
        cout << int64_t(fastR.xoshiro256p()) << endl;    // returns number in [INT64_MIN, INT32_MAX] interval
        cout << fastR.xoshiro256p_VNI<float>()) << endl; // returns number in [-1.0, 1.0] interval in single precision
        cout << fastR.xoshiro256p_UNI<float>()) << endl; // returns number in [ 0.0, 1.0] interval in single precision
        cout << fastR.xoshiro256p_Range<double>(-3.0, 7.0)) << endl; // returns number in [-3.0, 7.0] interval in double precision
    }
//  N.B. all members/functions share same seed and subsequent xor & shift operations on it.
//       it is usually not a problem, but if need different seeds (or separate PRNG) have to declare 
//       more/different fastXS64 objects

//  or you can also use static members from fastXS64s class w/o declaration: the seed is always "chrono"
    for(int i=0; i<10000; i++) {
        cout <<         fastXS64s::xoshiro256p()  << endl;   // returns number in [0, UINT64_MAX] interval
        cout << int64_t(fastXS64s::xoshiro256p()) << endl;   // returns number in [INT64_MIN, INT32_MAX] interval
        cout << fastXS64s::xoshiro256p_VNI<float>()) << endl;// returns number in [-1.0, 1.0] interval in single precision from 64bit PRNG
        cout << fastXS64s::xoshiro256p_UNI<float>()) << endl;// returns number in [ 0.0, 1.0] interval in single precision from 64bit PRNG
        cout << fastXS64s::xoshiro256p_Range<double>(-5.0, 5.0)) << endl; // returns number in [-5.0, 5.0] interval in double precision from 64bit PRNG
    }
//  N.B. all members/functions share same seed, and subsequent xor & shift operations on it.
//       it is usually not a problem, but if need different seeds (or separate PRNG) have to use 
//       fastXS64 (non static) class, and have to declare different fastXS64 objects.

// static declaration of a non static class (e.g. if you need to initialize it to specific seed)
    for(int i=0; i<10000; i++) {
        static fastXS64 fastR(0x123456789ABCDEF0); // personal seed also to (re)generate a specific random numbers sequence
        cout <<         fastR.xoshiro256p()  << endl;    // returns number in [0, UINT64_MAX] interval
        cout << int64_t(fastR.xoshiro256p()) << endl;    // returns number in [INT64_MIN, INT32_MAX] interval
        cout << fastR.xoshiro256p_VNI<float>()) << endl; // returns number in [-1.0, 1.0] interval in single precision
        cout << fastR.xoshiro256p_UNI<float>()) << endl; // returns number in [ 0.0, 1.0] interval in single precision
        cout << fastR.xoshiro256p_Range<double>(-3.0, 7.0)) << endl; // returns number in [-3.0, 7.0] interval in double precision
    }

• Example: use KISS 32bit algorithm:

    fastPRNG::fastRand32 fastRandom; // for 32bit
//  fstRnd::fastRand32 fastRandom(0x12345678); or with seed initialization: to (re)generate a specific random numbers sequence 
    for(int i=0; i<10000; i++) {
        cout <<         fastRandom.KISS()  << endl;   // returns number in [0, UINT32_MAX] interval
        cout << int32_t(fastRandom.KISS()) << endl;   // returns number in [INT32_MIN, INT32_MAX] interval
        cout << fastRandom.KISS_VNI<float>()) << endl;// returns number in [-1.0, 1.0] interval in single precision from 32bit PRNG
        cout << fastRandom.KISS_UNI<float>()) << endl;// returns number in [ 0.0, 1.0] interval in single precision  from 32bit PRNG
        cout << fastRandom.KISS_Range<double>(-3.0, 7.0)) << endl; // returns number in [-3.0, 7.0] interval in from 32bit PRNG
    }

• Any class object can to re-initialized with a new seed calling seed()

    fastPRNG::fastXS32 fastR(0x12345678); // seed to specific value

    for(int i=0; i<10000; i++) 
        cout << fastR.xoshiro256p()  << endl;   // returns number in [0, UINT64_MAX] interval

    fastR.seed(0x12345678); // same seed to obtain same sequence
    for(int i=0; i<10000; i++) 
        cout << fastR.xoshiro256p()  << endl;   // returns same number sequence in [0, UINT64_MAX] interval

    fastR.seed(); // new seed to 'chrono" to obtain different sequence
    for(int i=0; i<10000; i++) 
        cout << fastR.xoshiro256p()  << endl;   // returns number in [0, UINT64_MAX] interval

*classes fastXS32s and fastXS64s don't have seed() function: they have/are only static members.

For more details look at the source file: it's well documented.

 

Where it is used

Classes and functions are currently used

• glChAoS.P / wglChAoS.P Realtime 3D GPUs Strange Attractors and Hypercomplex Fractals scout

Hypercomplex fractals with stochastic IIM (Inverse Iteration Method) algorithm	DLA 3D (Diffusion Limited Aggregation) algorithm

• DLAf-optimized fast DLA (Diffusion Limited Aggregation) 2D/3D

 

Distribution Test

All functions are tested, below the distribution test in a cube with [-1.0, 1.0] side.

Live WebGL 2 / WebAssemly ==> fastPRNG distribution test
*Only FireFox and Chromium based web-browsers (Chrome / Opera / new Edge / etc.) are supported

30M dots/spheres cube	30M dots/spheres cube with clipping planes	3 thin boards from 30M dots cube
thin board from a 10M dots cube	2.5M dots/spheres with clipping planes	3 thin boards from 5M dots cube

*It's builded on the rendering particles engine of glChAoS.P / wglChAoS.P

*N.B. it's a distribution test, NOT a speed/benchmark test, since the rendering time/calculus is preeminent

Using distribution live WebGL test

The JavaScript / WebGL version is slower of Desktop one, so the test starts with 2.5M of particles, and a pre-allocated maxbuffer of 15M (for slow / low memory GPUs)

Particles panel
A) Start / Stop particles emitter
B) Continue / FullStop: continue endless (circular buffer) / stop when buffer is full
C) Endless / Restart: rewrite circular buffer / restart deleting all circular buffer
D) Set circular buffer size (drag with mouse)
          from .01M (10'000) to maxbuffer (default 15M pre-allocated) particles

*You can resize pre-allocated buffer changing the URL value of maxbuffer (in your browser address box) (e.g. maxbuffer=30 pre allocate a 30M particles memory buffer)

Desktop

If you want use the desktop version (available for Windows / Linux / MacOS), please download glChAoS.P / wglChAoS.P and build it with -DGLCHAOSP_TEST_RANDOM_DISTRIBUTION or enable the #define GLCHAOSP_TEST_RANDOM_DISTRIBUTION in attractorsBase.h file