一些Python模拟角谷猜想的性能测试

公共部分

def solve(n):
    step = 0
    while n != 1:
        if n % 2 == 0:
            n //= 2
        else:
            n = 3 * n + 1
        step += 1
    return step
N = 1000000

v1

没啥好说的顺序求解
主要代码：

1 2	x = [i for i in range(1, N + 1)] y = list(map(solve, x))

v2

测试电脑为6C12T 故采用max_workers=12
主要代码：

from concurrent.futures import ThreadPoolExecutor
x = [i for i in range(1, N + 1)]
with ThreadPoolExecutor(max_workers=12) as e:
    y = list(e.map(solve, x))

v3

主要代码：

1
2
3

import numpy as np
x = np.arange(1, N + 1, 1)
y = solve(x)

np.vectorize版：

import numpy as np
x = np.arange(1, N + 1, 1)
solve_vector = np.vectorize(solve)
y = solve_vector(x)

测试结果

+njit指给solve()函数加上了@njit标签
+vectorize指给solve()函数加上了@vectorize()标签
numpy不带(vec)指使用numba的#vectorize()标签而不使用np.vectorize 反之亦然

版本	说明	最大内存占用	时间1	时间2	时间3	平均时间
v1	Serial	106.1MB	10272.937ms	10022.724ms	10065.625ms	10120.429ms
v1.1	Serial+njit	119.6MB	712.524ms	699.496ms	691.721ms	701.247ms
v1.1p	Serial+njit	105.8MB	232.681ms	239.268ms	234.367ms	235.439ms
v2	ThreadPoolExecutor(12)	1750.6MB	43345.733ms	44833.593ms	40681.212ms	42953.513ms
v2.1	ThreadPoolExecutor(12)+njit	2013.0MB	17897.805ms	16293.948ms	16591.327ms	16927.693ms
v3	numpy(vec)	108.8MB	9986.405ms	10205.962ms	9894.202ms	10028.856ms
v3.1	numpy(vec)+njit	115.1MB	736.445ms	796.545ms	749.636ms	760.875ms
v3.2	numpy+vectorize	93.2MB	538.277ms	532.579ms	542.114ms	537.657ms
v3.3	numpy+vectorize+njit	99.3MB	932.287ms	907.180ms	927.224ms	922.230ms
v3.3p	numpy+vectorize+njit	92.7MB	231.020ms	227.666ms	228.072ms	228.919ms