Xoshiro128pp

class vulkpy.random.Xoshiro128pp

Bases: PRNG

xoshiro128++: Pseudo Random Number Generator

Notes

This class implements xoshiro128++ [1]. Initial internal states are sequentially generated during construction on CPU and are spaced 2^64 steps. Generating (pseudo-)random numbers are executed parallelly on GPU.

References

[1]

S. Vigna “xoshiro / xoroshiro generators and the PRNG shootout”, https://prng.di.unimi.it/

Methods Summary

normal(*[, shape, buffer, mean, stddev])	Generate Normal Distributing numbers
randint(*[, shape, buffer])	Generate [0, 2^32) unsigned integer numbers
random(*[, shape, buffer])	Generate [0, 1) floating numbers
randrange(*[, shape, buffer, low, high])	Generate [low, high) random numbers
wait()

Methods Documentation

normal(*, shape: Iterable[int] | None = None, buffer: Array | None = None, mean: float = 0.0, stddev: float = 1.0) → Array

Generate Normal Distributing numbers

Parameters:

• shape (iterable of ints, optional) – If specified, new vulkpy.Array with shape will be returned.

• buffer (vulkpy.Array) – If specified, generated numbers will be stored.

Returns:

Array which will get random numbers.

Return type:

vulkpy.Array

Raises:

ValueError – If neither shape or buffer are specified

Notes

This method first generates [0, 1) uniform random numbers, then transforms them to normal distribution with Box-Muller method. Box-Muller might have problem in terms of random number quality, however, it is quite GPU friendly.

randint(*, shape: Iterable[int] | None = None, buffer: U32Array | None = None) → U32Array

Generate [0, 2^32) unsigned integer numbers

Parameters:

• shape (iterable of ints, optional) – If specified, new vulkpy.U32Array with shape will be returned.

• buffer (vulkpy.U32Array) – If specified, generated numbers will be stored.

Returns:

Array which will get random numbers.

Return type:

vulkpy.U32Array

Raises:

ValueError – If neither shape or buffer are specified.

random(*, shape: Iterable[int] | None = None, buffer: Array | None = None) → Array

Generate [0, 1) floating numbers

Parameters:

• shape (iterable of ints, optional) – If specified, new vulkpy.Array with shape will be returned.

• buffer (vulkpy.Array) – If specified, generated numbers will be stored.

Returns:

Array which will get random numbers.

Return type:

vulkpy.Array

Raises:

ValueError – If neither shape or buffer are specified.

randrange(*, shape: Iterable[int] | None = None, buffer: U32Array | None = None, low: int = 0, high: int = 4294967296) → U32Array

Generate [low, high) random numbers

Parameters:

• shape (iterable of ints, optional) – If specified, new vulkpy.U32Array with shape will be returned.

• buffer (vulkpy.Array) – If specified, generated numbers will be stored.

• low (int, optional) – Inclusive lowest value. The default is 0.

• high (int, optional) – Exclusive highest value. The default is 2^32.

Returns:

Array which will get random numbers.

Return type:

vulkpy.U32Array

Raises:

• ValueError – If neither shape or buffer are specified.

• ValueError – If not 0 <= low < high <= 2^32.

wait()

__init__(gpu: GPU, size: int = 64, *, seed: int | None = None)

Initialize Xoshiro128pp

Parameters:

• gpu (vulkpy.GPU) – GPU where PRNG allocates

• size (int) – Number of internal states. These states generate random number parallelly.

• seed (int, optional) – Random seed. If None (default), use hardware random instead.