一个非常快速的Atkin Sieve的实现是Dan Bernstein的primegen。这个筛子比埃拉托色尼的筛子更有效率。他的页面有一些基准测试信息。

#include <iostream>

using namespace std;

int set [1000000];

int main (){

    for (int i=0; i<1000000; i++){
        set [i] = 0;
    }
    int set_size= 1000;
    set [set_size];
    set [0] = 2;
    set [1] = 3;
    int Ps = 0;
    int last = 2;

    cout << 2 << " " << 3 << " ";

    for (int n=1; n<10000; n++){
        int t = 0;
        Ps = (n%2)+1+(3*n);
        for (int i=0; i==i; i++){
            if (set [i] == 0) break;
            if (Ps%set[i]==0){
                t=1;
                break;
            }
        }
        if (t==0){
            cout << Ps << " ";
            set [last] = Ps;
            last++;
        }
    }
    //cout << last << endl;


    cout << endl;

    system ("pause");
    return 0;
}

另一个Python实现比死亡面具推销员的答案更直接，也更快:

import numpy as np

def prime_numbers(limit: int) -> list[int]:
    """Provide a list of all prime numbers <= the limit."""
    is_prime = np.full((limit + 1, ), True)
    is_prime[0:2] = False
    for n in range(2, limit + 1):
        if is_prime[n]:
            is_prime[n**2::n] = False
    return list(np.where(is_prime)[0])

你可以进一步优化，例如，排除2，或者硬编码更多质数，但我想保持简单。

*示例运行时比较(注意:我使用了其他实现的优化形式，见我的评论):

我会让你决定这是不是最快的。

using System;
namespace PrimeNumbers
{

public static class Program
{
    static int primesCount = 0;


    public static void Main()
    {
        DateTime startingTime = DateTime.Now;

        RangePrime(1,1000000);   

        DateTime endingTime = DateTime.Now;

        TimeSpan span = endingTime - startingTime;

        Console.WriteLine("span = {0}", span.TotalSeconds);

    }


    public static void RangePrime(int start, int end)
    {
        for (int i = start; i != end+1; i++)
        {
            bool isPrime = IsPrime(i);
            if(isPrime)
            {
                primesCount++;
                Console.WriteLine("number = {0}", i);
            }
        }
        Console.WriteLine("primes count = {0}",primesCount);
    }



    public static bool IsPrime(int ToCheck)
    {

        if (ToCheck == 2) return true;
        if (ToCheck < 2) return false;


        if (IsOdd(ToCheck))
        {
            for (int i = 3; i <= (ToCheck / 3); i += 2)
            {
                if (ToCheck % i == 0) return false;
            }
            return true;
        }
        else return false; // even numbers(excluding 2) are composite
    }

    public static bool IsOdd(int ToCheck)
    {
        return ((ToCheck % 2 != 0) ? true : false);
    }
}
}

在我使用2.40 GHz处理器的酷睿2 Duo笔记本电脑上，查找并打印1到1,000,000范围内的质数大约需要82秒。它找到了78,498个质数。

这取决于您的应用程序。这里有一些注意事项:

你需要的仅仅是一些数字是否是质数的信息，你需要所有的质数达到一定的限度，还是你需要(潜在的)所有的质数? 你要处理的数字有多大?

米勒-拉宾和模拟测试只比筛选超过一定规模的数字(我相信大约在几百万左右)的速度快。在这以下，使用试除法(如果你只有几个数字)或筛子会更快。

Rabin-Miller是一个标准的概率质数检验。(你运行K次，输入数字要么肯定是合数，要么可能是素数，误差概率为4-K。(经过几百次迭代，它几乎肯定会告诉你真相)

拉宾·米勒有一个非概率(确定性)的变体。

The Great Internet Mersenne Prime Search (GIMPS) which has found the world's record for largest proven prime (274,207,281 - 1 as of June 2017), uses several algorithms, but these are primes in special forms. However the GIMPS page above does include some general deterministic primality tests. They appear to indicate that which algorithm is "fastest" depends upon the size of the number to be tested. If your number fits in 64 bits then you probably shouldn't use a method intended to work on primes of several million digits.