numba加速测试

import numpy as np
import time
from numba import jit

from multiprocessing.dummy import Pool

def numpy_original(a,b):
    y,x = np.indices((a,b),dtype=np.int32)
    res = np.multiply(x,y)
    return res

@jit(nopython=True,nogil=False)
def numba_gil(a,b):
    res = np.zeros((a,b),dtype=np.int32)
    for i in range(a):
        for j in range(b):
            res[i][j]=i*j
    return res

@jit(nopython=True,nogil=True)
def numba_nogil(a,b):
    res = np.zeros((a,b),dtype=np.int32)
    for i in range(a):
        for j in range(b):
            res[i][j]=i*j
    return res

def costtime(func,a,b):
    times=100
    param =[(a,b)]*times
    t0=time.time()
    with Pool(8) as p:
        p.map(lambda x:func(x[0],x[1]),param)
    t1 = time.time()
    return t1-t0

if __name__=="__main__":
    a = 10
    b = 10
    #numpy_dot(a,b)
    numba_gil(a,b) # warmup
    numba_nogil(a,b) # warmup
    a = 5000
    b = 5000
    cost1 = costtime(numpy_original,a,b)
    print('numpy 原生方法：',cost1)
    cost2 = costtime(numba_gil,a,b)
    print('numba加速关闭nogil:',cost2)
    cost3 = costtime(numba_nogil,a,b)
    print('numba加速开启nogil:',cost3)




"""
numpy 原生方法： 13.166807651519775
numba加速关闭nogil: 7.045409679412842
numba加速开启nogil: 5.6226959228515625
"""

结论：

numba对于数据量的的计算优化效果明显，nogil开启对多线程大约有20%以上的优化。