pulp::signal::Xorshift32 Class Reference

Xorshift32 (Marsaglia, "Xorshift RNGs", J. More...

#include <rng.hpp>

Public Member Functions
constexpr	Xorshift32 ()=default
constexpr	Xorshift32 (std::uint32_t s)
void	seed (std::uint32_t s)
void	set_seed (std::uint32_t s)
	Compatibility spelling used by the Round 2 DSP blocks.
void	reset ()
	Rewind to the seeded stream position.
std::uint32_t	next_u32 ()
std::uint32_t	next_uint ()
	Compatibility spelling used by the Round 2 DSP blocks.
float	next_unipolar ()
	Uniform in [0, 1).
float	next_bipolar ()
	Uniform in [-1, 1).
template<typename T>
T	next_bipolar ()
float	next_unit ()
	Uniform in [0, 1), with the caller-selected result precision.
template<typename T>
T	next_unit ()
float	gaussian ()
	Standard normal via the polar-free Box-Muller transform (Box & Muller, Ann.
std::uint32_t	state () const
std::uint32_t	seed () const

Static Public Attributes
static constexpr std::uint32_t	kDefaultSeed = 0x9E3779B9u
	Any nonzero constant works; this is the 32-bit golden-ratio odd constant, which is also what seed(0) normalizes to.

Detailed Description

Xorshift32 (Marsaglia, "Xorshift RNGs", J.

Stat. Soft. 8(14), 2003) with the canonical 13/17/5 triple.

Zero is the generator's fixed point — a zero-initialized instance would emit zeros forever — so the state is defaulted to a nonzero constant and a zero state is repaired lazily on use. That keeps "zero-init = fresh" true for aggregates that embed this without a constructor call.

Constructor & Destructor Documentation

Xorshift32() [1/2]

pulp::signal::Xorshift32::Xorshift32 ( )

constexprdefault

Xorshift32() [2/2]

pulp::signal::Xorshift32::Xorshift32 ( std::uint32_t s )

inlineexplicitconstexpr

Member Function Documentation

gaussian()

float pulp::signal::Xorshift32::gaussian ( )

inline

Standard normal via the polar-free Box-Muller transform (Box & Muller, Ann.

Math. Statist. 29(2), 1958). The transform produces two independent deviates per pair of uniforms; the second is cached so the amortized cost is one log/sqrt/sincos pair every other call.

u1 is drawn on (0, 1] rather than [0, 1) — log(0) is the one input that breaks the transform.

next_bipolar() [1/2]

template<typename T>

T pulp::signal::Xorshift32::next_bipolar ( )

inline

next_bipolar() [2/2]

float pulp::signal::Xorshift32::next_bipolar ( )

inline

Uniform in [-1, 1).

next_u32()

std::uint32_t pulp::signal::Xorshift32::next_u32 ( )

inline

next_uint()

std::uint32_t pulp::signal::Xorshift32::next_uint ( )

inline

Compatibility spelling used by the Round 2 DSP blocks.

next_unipolar()

float pulp::signal::Xorshift32::next_unipolar ( )

inline

Uniform in [0, 1).

Built from the top 24 bits so every result is exactly representable in float and the spacing is uniform.

next_unit() [1/2]

template<typename T>

T pulp::signal::Xorshift32::next_unit ( )

inline

next_unit() [2/2]

float pulp::signal::Xorshift32::next_unit ( )

inline

Uniform in [0, 1), with the caller-selected result precision.

reset()

void pulp::signal::Xorshift32::reset ( )

inline

Rewind to the seeded stream position.

reset() restores the seed the caller set, not the default one — a reset that silently discarded the seed would make every seeded object share one stream, which is the opposite of what seeding is for.

seed() [1/2]

std::uint32_t pulp::signal::Xorshift32::seed ( ) const

inline

seed() [2/2]

void pulp::signal::Xorshift32::seed ( std::uint32_t s )

inline

set_seed()

void pulp::signal::Xorshift32::set_seed ( std::uint32_t s )

inline

Compatibility spelling used by the Round 2 DSP blocks.

state()

std::uint32_t pulp::signal::Xorshift32::state ( ) const

inline

Member Data Documentation

kDefaultSeed

std::uint32_t pulp::signal::Xorshift32::kDefaultSeed = 0x9E3779B9u

staticconstexpr

Any nonzero constant works; this is the 32-bit golden-ratio odd constant, which is also what seed(0) normalizes to.

---

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

• /home/runner/work/pulp/pulp/core/signal/include/pulp/signal/rng.hpp