What is the fastest method to create uniform random numbers in c++

Question

I need to generate uniform random numbers in a for loop. for loop goes for 1000000 numbers. There is another for loop which goes to 2000 inside it. So I generate 2*10^9 uniform random numbers. I use below method:

#include <random>
    double zeta;
     unsigned seed = std::chrono::system_clock::now().time_since_epoch().count();
        auto uniform_rand = bind(uniform_real_distribution<double>(0,1), mt19937(seed)); 
    for(int j=0; j<1000000; j++)
               for(int i=0; i<2000; i++) 
       zeta=-eta/2.0+uniform_rand()*eta; // in the range (-eta/2, +eta/2)
theta[i]+=zeta;
    end
end

`std::bind` tends to be outclassed by lambdas, both in readability and performance. — chris, Oct 17 '17 at 16:55
That's an awful lot of random numbers. If you need that many, why not break this out into threads to make use of multiple cores? — tadman, Oct 17 '17 at 16:57
He means multi thread. Split up the job and generate 1/4 of each run on each of 4 threads — user4581301, Oct 17 '17 at 16:59
Why do you need so many random numbers? And why do you need to generate them all up-front rather when they are needed? — Jesper Juhl, Oct 17 '17 at 16:59
@OliverRange Than generate them as needed? Why are they generated upfront? — Passer By, Oct 17 '17 at 17:01
I am using them to change the angle of each walker in time. However I have not written that in the question.@PasserBy — Oliver Range, Oct 17 '17 at 17:04
It depends what you're doing with all of these things. So far it looks like nothing, so it's not a relevant concern. I'm just suggesting ways of making this code faster if that's your ultimate goal, but this is presumably just a single part of your problem. — tadman, Oct 17 '17 at 17:07
Yes, that is really ultimate goal or me. Could you please provide an answer with your suggestion? And do you think a better method to create them could efficiently change run time?@tadman — Oliver Range, Oct 17 '17 at 17:09
As far as my point goes, all I can suggest is [becoming familiar with lambdas](https://stackoverflow.com/questions/7627098/what-is-a-lambda-expression-in-c11). — chris, Oct 17 '17 at 17:12

navyblue · Accepted Answer · 2017-10-17T19:50:16.880

1

It's almost the same as yours. I just don't see any need for a binder or a lambda.

EDIT: I also changed the generator from mt19937 to minstd_rand which has made the code 88 times faster. With mt19937 the code has aproximately the same performance as the one in the question.

#include <random>

int main() {
    /*std::mt19937 gen(std::random_device{}());*/
    // Changing mt19937 to minstd_rand makes the code run 88 times faster!
    std::minstd_rand gen(std::random_device{}());
    std::uniform_real_distribution<double> dist(0, 1);
    for(unsigned int i = 0; i < 1000000; ++i) {
        for(unsigned int j = 0; j < 2000; ++j) {
            double anotherRandomNumber = dist(gen);
            // Do whatever you want with generated random number.
        }
    }
}

edited Oct 17 '17 at 19:50

answered Oct 17 '17 at 17:29

navyblue

776
4
8

Is it really faster than what I have written? – Oliver Range Oct 17 '17 at 17:32
Does it create different random numbers in different runs? – Oliver Range Oct 17 '17 at 17:33
@OliverRange yes, it seeds the random generator each time with `std::random_device`. But sometimes it won't work, see https://stackoverflow.com/questions/18880654/why-do-i-get-the-same-sequence-for-every-run-with-stdrandom-device-with-mingw – Mark Ransom Oct 17 '17 at 17:41
@OliverRange `std::random_device` will use a genuinely non deterministic random device if your platform has one. That means pretty much all desktop operating systems. The one problem is that `GCC MinGW` compiler on windows has a bug so it doesn't work on that specific compiler (yet). – Galik Oct 17 '17 at 17:48
@OliverRange Yes, it is just one Windows compiler that has the bug. – Galik Oct 17 '17 at 17:51
no match for call to ‘(std::uniform_real_distribution) (double&, double&, double&, double&, double&)’ It doesn't run @Galik – Oliver Range Oct 17 '17 at 18:13
@OliverRange The code in this answer runs for me. Did you type it in exactly as shown? – Galik Oct 17 '17 at 18:21
The error message suggests that you're doing something really strange. It seems that you're calling `dist` with x[i], y[i], x[*itp] etc. Make sure you call it exactly like I did in may answer. That is you can only call `dist` giving it a generator as the argument; in this case `gen`. – navyblue Oct 21 '17 at 17:52
Is it better to put std::minstd_rand gen(std::random_device{}()); std::uniform_real_distribution dist(0, 1); outside of the main fuction or not? – Oliver Range Oct 23 '17 at 09:10
It's too general a question to answer it. However, if you mean this particular piece of code that I've written as the answer then you shouldn't move `gen` or `dist` out of the `main` function because that would make this simple program overcompilcated or you would declare them as global variables which wouldn't make the code any better. – navyblue Oct 23 '17 at 14:17

score 0 · Answer 2 · answered Oct 17 '17 at 17:55

0

Use this algorithm from here:

uint64_t s[2] = { 0x41, 0x29837592 };

static inline uint64_t rotl(const uint64_t x, int k) {
    return (x << k) | (x >> (64 - k));
}

uint64_t next(void) {
    const uint64_t s0 = s[0];
    uint64_t s1 = s[1];
    const uint64_t result = s0 + s1;

    s1 ^= s0;
    s[0] = rotl(s0, 55) ^ s1 ^ (s1 << 14); // a, b
    s[1] = rotl(s1, 36); // c

    return result;
}

double uniform() {
    return next()*(1.0/18446744073709551616.0);
}

This is 4 times faster than your example on my machine.

Note: you need to seed s, perhaps with std::random_device.

answered Oct 17 '17 at 17:55

geza

28,403
6
61
135

Could you please provide an answer which use for loop as my case? – Oliver Range Oct 17 '17 at 18:00
@OliverRange: paste this code into yours, and replace `uniform_rand()` with `uniform()`. – geza Oct 17 '17 at 18:01
I don't know where to post it. after main and before for loop? – Oliver Range Oct 17 '17 at 18:03
@OliverRange: before main. But, to be honest, first you need to learn the basics of C++ before doing something like this... – geza Oct 17 '17 at 18:07
Segmentation fault – Oliver Range Oct 17 '17 at 18:18
@OliverRange: then you have some problem with some other part of your code. This code cannot cause segmentation fault. – geza Oct 17 '17 at 18:21

What is the fastest method to create uniform random numbers in c++

2 Answers2