Struct XorwowState 

pub struct XorwowState {
    pub s: [u32; 5],
    pub d: u32,
}

Compact per-row XORWOW state. Layout matches curand_kernel.h’s curandStateXORWOW_t for the five state lanes plus the addition counter; we omit the boxmuller cache (PG sampler doesn’t use it).

Fields

s: [u32; 5]

d: u32

Implementations

impl XorwowState

pub fn new(seed: u64, row: u64) -> Self

Stateless seeding from (seed, row). Each of the six state words is the high or low half of a SplitMix64 hash of splitmix64(seed ⊕ row·ROW_ZETA ⊕ word·WORD_GAMMA). The first non-zero state word is enforced so we never enter the all-zero XORWOW absorbing fixed point.

pub fn next_u32(&mut self) -> u32

Single XORWOW advance. Returns the next 32-bit output and mutates the state. Matches Marsaglia’s 2003 XORWOW formulation, which is also what curand_kernel.h::xorwow computes.

pub fn next_unit(&mut self) -> f64

Uniform double in (0, 1] — same (u32 + 1) / 2^32 convention the kernel uses (matches curand_uniform_double upper-open interval convention; we use the upper-closed variant so a zero u32 never produces exactly zero, which would crash log(u) in the Exp draw).

pub fn next_exp(&mut self) -> f64

Standard exponential via inverse CDF: -ln(U). U is on (0, 1] so -ln(U) is in [0, +inf), never +inf from a zero argument.

pub fn next_norm(&mut self) -> f64

Standard normal via Marsaglia polar method. Discards the second variate the polar pair produces (cleaner than caching it across calls — we’d need a per-row scratch slot, which the device kernel can’t afford to spill).

Trait Implementations

impl Clone for XorwowState

fn clone(&self) -> XorwowState

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Copy for XorwowState

impl Debug for XorwowState

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl PgRng for XorwowState

XorwowState is the randomness source for the bit-exact GPU oracle. Wiring it through PgRng lets the shared Devroye core run against the same RNG byte stream the device kernel consumes.

fn next_unit(&mut self) -> f64

Uniform variate. Callers must keep this on (0, 1] (strictly positive) so the inverse-CDF exponential never sees log(0).

fn next_exp(&mut self) -> f64

Standard exponential Exp(1) variate.

fn next_norm(&mut self) -> f64

Standard normal N(0, 1) variate.