Created
October 21, 2014 14:54
-
-
Save andreasnoack/9f3f63d7f1d00a07359b to your computer and use it in GitHub Desktop.
A new RNG for Julia
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
module XORShift | |
import Base.rand | |
type XORShiftStar1024 <: AbstractRNG | |
p::Int | |
state::Vector{Uint64} | |
end | |
XORShiftStar() = XORShiftStar1024(1, rand(Uint64, 16)) | |
const globalXORShiftStar1024 = XORShiftStar() | |
function rand(rng::XORShiftStar1024) | |
s = rng.state | |
p = rng.p | |
s0 = s[p] | |
p = p %16 + 1 | |
s1 = s[p] | |
s1 $= s1 << 31 | |
s1 $= s1 >> 11 | |
s0 $= s0 >> 30 | |
s0 $= s1 | |
s[p] = s0 | |
rng.p = p | |
return s0 * 1181783497276652981 | |
end | |
randXORShiftStar() = rand(globalXORShiftStar1024)*5.421010862427522e-20 # The number is 1/2.0^64 | |
end #module |
@stevengj Did you benchmark that? I would assume that the heterogeneous type results in type instability getfield
(although you could always use unsafe_load
if you wanted). Also I think the increments should be optimized out, since computing the address to load requires decrementing the argument to getfield
/unsafe_load!
.
I wonder whether this could be vectorized to compute two or four random numbers per iteration, but that would probably make getting a single number out of rand
more complicated. I also don't think it can be made to work without llvmcall
or maybe the SLP vectorizer.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
You can slightly modify it to avoid a couple of unnecessary
+ 1
operations (just makep
anInt
and start it at1
). But in this case you need& 15
since%
is more expensive forInt
.