RNG list to be implemented

[sunoru]

As discussed before, I'd like to implement various RNGs in this package. And temporarily they are listed in the following:

Four from Random123:

• [x] Threefry
• [x] Philox
• [x] AESNI
• [x] ARS

The xorshift families:

• [x] xoroshiro128+
• [x] xorshift128+
• [x] xorshift1024*

Six from SPRNG

• [ ] Combined Multiple Recursive Generator
• [ ] 48 Bit Linear Congruential Generator with Prime Addend
• [ ] 64 Bit Linear Congruential Generator with Prime Addend
• [ ] Modified Lagged Fibonacci Generator
• [ ] Multiplicative Lagged Fibonacci Generator
• [ ] Prime Modulus Linear Congruential Generator

[sunoru]

• [x] PCG families seem to be a nice choice, it's also worth considering.

[ChrisRackauckas]

Philox via Random123 (and others) can be used on a GPU: any way for this to plug in with ArrayFire.jl or CUDArt.jl?

[sunoru]

Hi Chris, thank you for the suggestion.

Yes, I have considered accelerating some RNGs with GPU, but I was trying to write code in pure Julia at first.

Actually I want to make this package have as few requires as possible, and I don't know much about these GPU packages, so currently they are not used. Do you have any idea about which to choose?

[ChrisRackauckas]

You can think about making it a conditional dependency which requires the user to using CUDArt for it to work (and not have it in requires). CUDArt.jl is definitely the way to go for this.

[sunoru]

It sounds a good idea. I will have a try.

[kafisatz]

Here are two additional (fast?) RNGs which may be worth considering: wyhash and lehmer's https://lemire.me/blog/2019/03/19/the-fastest-conventional-random-number-generator-that-can-pass-big-crush/

[sunoru]

@kafisatz thanks, I'll check it out.