openrand::BaseRNG< RNG > Class Template Reference

Base class for random number generators. More...

#include <base_state.h>

Public Types
using	result_type = uint32_t

Public Member Functions
OPENRAND_DEVICE result_type	operator() ()
	Generates a 32 bit unsigned integer from a uniform distribution. More...
template<typename T = float>
OPENRAND_DEVICE T	rand ()
	Generates a random number from a uniform distribution between 0 and 1. More...
template<typename T = float>
OPENRAND_DEVICE T	uniform (const T low, const T high)
	Generates a number from a uniform distribution between a and b. More...
template<typename T = float>
OPENRAND_DEVICE void	fill_random (T *array, const int N)
template<typename T = float>
OPENRAND_DEVICE T	randn ()
	Generates a random number from a normal distribution with mean 0 and std 1. More...
template<typename T = float>
OPENRAND_DEVICE vec2< T >	randn2 ()
	More efficient version of randn, returns two values at once. More...
template<typename T = float>
OPENRAND_DEVICE T	randn (const T mean, const T std_dev)
	Generates a random number from a normal distribution with mean and std. More...
template<bool biased = true, typename T = int>
OPENRAND_DEVICE T	range (const T N)
	Generates a random integer of certain range. More...
template<typename T = float>
OPENRAND_DEVICE T	gamma (T alpha, T b)
	Generates a random number from a gamma distribution with shape alpha and scale b. More...
template<typename T = RNG>
std::enable_if_t< has_counter< T >::value, RNG >	forward_state (int n) const
	Returns a new generator with the internal state forwarded by a given number. More...

Static Public Member Functions
static constexpr result_type	min ()
static constexpr result_type	max ()

Protected Member Functions
template<typename Ftype , typename Utype >
OPENRAND_DEVICE Ftype	u01 (const Utype in) const

Detailed Description

template<typename RNG>
class openrand::BaseRNG< RNG >

Base class for random number generators.

This class utilizes the CRTP pattern to inject some common functionalities into random number generators.

Template Parameters

RNG	Random number generator class. It must contain two public methods, a constructor and a templated draw() function that returns the next 32 or 64 bit random unsigned int from its stream.

Member Function Documentation

forward_state()

template<typename RNG >

template<typename T = RNG>

std::enable_if_t<has_counter<T>::value, RNG> openrand::BaseRNG< RNG >::forward_state ( int  n ) const

inline

Returns a new generator with the internal state forwarded by a given number.

The new generator's first output will be the n+1th output of the current generator and so on. This is O(1) operation.

Parameters

n	Number of steps to move the state forward

Returns

RNG A new RNG object with the state moved forward by n steps

gamma()

template<typename RNG >

template<typename T = float>

OPENRAND_DEVICE T openrand::BaseRNG< RNG >::gamma ( T  alpha, T  b  )

inline

Generates a random number from a gamma distribution with shape alpha and scale b.

Adapted from the following implementation: https://www.hongliangjie.com/2012/12/19/how-to-generate-gamma-random-variables/

Template Parameters

T	floating point type to be returned

Parameters

alpha	shape parameter of the gamma distribution
b	scale parameter of the gamma distribution

Returns

T random number from a gamma distribution with shape alpha and scale b

operator()()

template<typename RNG >

OPENRAND_DEVICE result_type openrand::BaseRNG< RNG >::operator() ( )

inline

Generates a 32 bit unsigned integer from a uniform distribution.

This function is needed to conform to C++ engine interface

Returns

uint32_t random number from a uniform distribution

rand()

template<typename RNG >

template<typename T = float>

OPENRAND_DEVICE T openrand::BaseRNG< RNG >::rand ( )

inline

Generates a random number from a uniform distribution between 0 and 1.

Note

Some generators may expose a more efficient version of this function that returns multiple values at once.

Template Parameters

T	Data type to be returned. Can be 32 or 64 bit integer, float or double.

Returns

T random number from a uniform distribution between 0 and 1

randn() [1/2]

template<typename RNG >

template<typename T = float>

OPENRAND_DEVICE T openrand::BaseRNG< RNG >::randn ( )

inline

Generates a random number from a normal distribution with mean 0 and std 1.

This function implements box-muller method. This method avoids branching, and therefore more efficient on GPU. see: https://developer.nvidia.com/gpugems/gpugems3/part-vi-gpu-computing/chapter-37-efficient-random-number-generation-and-application

Template Parameters

T	floating point type to be returned

Returns

T random number from a normal distribution with mean 0 and std 1

randn() [2/2]

template<typename RNG >

template<typename T = float>

OPENRAND_DEVICE T openrand::BaseRNG< RNG >::randn ( const T  mean, const T  std_dev  )

inline

Generates a random number from a normal distribution with mean and std.

Template Parameters

T	floating point type to be returned

Parameters

mean	mean of the normal distribution
std_dev	standard deviation of the normal distribution

Returns

T random number from a normal distribution with mean and std

randn2()

template<typename RNG >

template<typename T = float>

OPENRAND_DEVICE vec2<T> openrand::BaseRNG< RNG >::randn2 ( )

inline

More efficient version of randn, returns two values at once.

Template Parameters

T	floating point type to be returned

Returns

T random number from a normal distribution with mean 0 and std 1

range()

template<typename RNG >

template<bool biased = true, typename T = int>

OPENRAND_DEVICE T openrand::BaseRNG< RNG >::range ( const T  N )

inline

Generates a random integer of certain range.

This uses the method described in [1] to generate a random integer of range [0..N)

Attention

if using non-biased version, please make sure that N is not too large [2]

Template Parameters

biased	if true, the faster, but slightly biased variant is used.
T	integer type (<=32 bit) to be returned

Parameters

N	A random integral of range [0..N) will be returned

Returns

T random number from a uniform distribution between 0 and N

Note

[1] https://lemire.me/blog/2016/06/30/fast-random-shuffling/

[2] If N=2^b (b<32), pr(taking the branch) = p = 1/2^(32-b). For N=2^24, this value is 1/2^8 = .4%, quite negligible. But GPU complicates this simple math. Assuming a warp size of 32, the probability of a thread taking the branch becomes 1 - (1-p)^32. For N=2^24, that value is 11.8%. For N=2^20, it's 0.8%.

uniform()

template<typename RNG >

template<typename T = float>

OPENRAND_DEVICE T openrand::BaseRNG< RNG >::uniform ( const T  low, const T  high  )

inline

Generates a number from a uniform distribution between a and b.

@Note For integer types, consider using range for greater control.

Template Parameters

T	Data type to be returned. Can be float or double. double.

Parameters

low	lower bound of the uniform distribution
high	upper bound of the uniform distribution

Returns

T random number from a uniform distribution between a and b

---

The documentation for this class was generated from the following file:

• /home/runner/work/OpenRAND/OpenRAND/include/openrand/base_state.h