我很惊讶居然没人提到numba。

该版本在2.47 ms±36.5µs内达到1M标记。

几年前，维基百科页面上出现了一个阿特金筛子的伪代码。这已经不存在了，参考阿特金筛似乎是一个不同的算法。一个2007/03/01版本的维基百科页面(Primer number as 2007-03-01)显示了我用作参考的伪代码。

import numpy as np
from numba import njit

@njit
def nb_primes(n):
    # Generates prime numbers 2 <= p <= n
    # Atkin's sieve -- see https://en.wikipedia.org/w/index.php?title=Prime_number&oldid=111775466
    sqrt_n = int(np.sqrt(n)) + 1

    # initialize the sieve
    s = np.full(n + 1, -1, dtype=np.int8)
    s[2] = 1
    s[3] = 1

    # put in candidate primes:
    # integers which have an odd number of
    # representations by certain quadratic forms
    for x in range(1, sqrt_n):
        x2 = x * x
        for y in range(1, sqrt_n):
            y2 = y * y
            k = 4 * x2 + y2
            if k <= n and (k % 12 == 1 or k % 12 == 5): s[k] *= -1
            k = 3 * x2 + y2
            if k <= n and (k % 12 == 7): s[k] *= -1
            k = 3 * x2 - y2
            if k <= n and x > y and k % 12 == 11: s[k] *= -1

    # eliminate composites by sieving
    for k in range(5, sqrt_n):
        if s[k]:
            k2 = k*k
            # k is prime, omit multiples of its square; this is sufficient because
            # composites which managed to get on the list cannot be square-free
            for i in range(1, n // k2 + 1):
                j = i * k2 # j ∈ {k², 2k², 3k², ..., n}
                s[j] = -1
    return np.nonzero(s>0)[0]

# initial run for "compilation" 
nb_primes(10)

时机

In[10]:
%timeit nb_primes(1_000_000)

Out[10]:
2.47 ms ± 36.5 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)

In[11]:
%timeit nb_primes(10_000_000)

Out[11]:
33.4 ms ± 373 µs per loop (mean ± std. dev. of 7 runs, 10 loops each)

In[12]:
%timeit nb_primes(100_000_000)

Out[12]:
828 ms ± 5.64 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

下面是我在Python中通常用来生成质数的代码:

$ python -mtimeit -s'import sieve' 'sieve.sieve(1000000)' 
10 loops, best of 3: 445 msec per loop
$ cat sieve.py
from math import sqrt

def sieve(size):
 prime=[True]*size
 rng=xrange
 limit=int(sqrt(size))

 for i in rng(3,limit+1,+2):
  if prime[i]:
   prime[i*i::+i]=[False]*len(prime[i*i::+i])

 return [2]+[i for i in rng(3,size,+2) if prime[i]]

if __name__=='__main__':
 print sieve(100)

它不能与这里发布的更快的解决方案竞争，但至少它是纯python。

谢谢你提出这个问题。我今天真的学到了很多东西。

这里是最快的函数之一的两个更新版本(纯Python 3.6)，

from itertools import compress

def rwh_primes1v1(n):
    """ Returns  a list of primes < n for n > 2 """
    sieve = bytearray([True]) * (n//2)
    for i in range(3,int(n**0.5)+1,2):
        if sieve[i//2]:
            sieve[i*i//2::i] = bytearray((n-i*i-1)//(2*i)+1)
    return [2,*compress(range(3,n,2), sieve[1:])]

def rwh_primes1v2(n):
    """ Returns a list of primes < n for n > 2 """
    sieve = bytearray([True]) * (n//2+1)
    for i in range(1,int(n**0.5)//2+1):
        if sieve[i]:
            sieve[2*i*(i+1)::2*i+1] = bytearray((n//2-2*i*(i+1))//(2*i+1)+1)
    return [2,*compress(range(3,n,2), sieve[1:])]

如果你可以控制N，列出所有质数的最快方法就是预先计算它们。认真对待。预计算是一种被忽视的优化方法。

从2021年的答案开始，我还没有发现二进制数组方法对10亿以下的质数有利。

但我可以用几个技巧将质数从2加速到接近x2:

使用numexpr库将numpy表达式转换为分配较少的紧循环 取代np。有更快的选择 以某种方式操作筛选的前9个元素，因此不需要改变数组的形状

总之，在我的机器上，质数< 10亿的时间从25秒变成了14.5秒

import numexpr as ne
import numpy as np

def primesfrom2to_numexpr(n):
    # https://stackoverflow.com/questions/2068372/fastest-way-to-list-all-primes-below-n-in-python/3035188#3035188
    """ Input n>=24, Returns a array of primes, 2 <= p < n + a few over"""
    sieve = np.zeros((n // 3 + (n % 6 == 2))//4+1, dtype=np.int32)
    ne.evaluate('sieve + 0x01010101', out=sieve)
    sieve = sieve.view('int8')
    #sieve = np.ones(n // 3 + (n % 6 == 2), dtype=np.bool_)
    sieve[0] = 0
    for i in np.arange(int(n ** 0.5) // 3 + 1):
        if sieve[i]:
            k = 3 * i + 1 | 1
            sieve[((k * k) // 3)::2 * k] = 0
            sieve[(k * k + 4 * k - 2 * k * (i & 1)) // 3::2 * k] = 0
    sieve[[0,8]] = 1
    result = np.flatnonzero(sieve)
    ne.evaluate('result * 3 + 1 + result%2', out=result)
    result[:9] = [2,3,5,7,11,13,17,19,23]
    return result

对于最快的代码，numpy解决方案是最好的。不过，出于纯粹的学术原因，我发布了我的纯python版本，它比上面发布的食谱版本快不到50%。由于我将整个列表放在内存中，所以需要足够的空间来容纳所有内容，但它的可伸缩性似乎相当好。

def daniel_sieve_2(maxNumber):
    """
    Given a number, returns all numbers less than or equal to
    that number which are prime.
    """
    allNumbers = range(3, maxNumber+1, 2)
    for mIndex, number in enumerate(xrange(3, maxNumber+1, 2)):
        if allNumbers[mIndex] == 0:
            continue
        # now set all multiples to 0
        for index in xrange(mIndex+number, (maxNumber-3)/2+1, number):
            allNumbers[index] = 0
    return [2] + filter(lambda n: n!=0, allNumbers)

结果是:

>>>mine = timeit.Timer("daniel_sieve_2(1000000)",
...                    "from sieves import daniel_sieve_2")
>>>prev = timeit.Timer("get_primes_erat(1000000)",
...                    "from sieves import get_primes_erat")
>>>print "Mine: {0:0.4f} ms".format(min(mine.repeat(3, 1))*1000)
Mine: 428.9446 ms
>>>print "Previous Best {0:0.4f} ms".format(min(prev.repeat(3, 1))*1000)
Previous Best 621.3581 ms