intel处理器在winubuntu和mklopenblas下的numpy性能测试

intel处理器在win/ubuntu和mkl/openblas下的numpy性能测试

测试代码参考自

共测试三个项目，numpy矩阵乘法，numpy矩阵resize和numpy跑图像滤波

测试代码如下，欢迎补充其他cpu成绩或指出测试不当的点

更多是为了满足自己好奇心hhh

import numpy as np

from time import time

d = 100  #矩阵尺寸

times = 5

a = np.random.rand(d,d,d)

b = np.random.rand(d,d,d)

start = time()

for i in range (times):

    c = np.dot(a,b)

end = time()

print('running time is :%f'%(end/times-start/times))

print(np.__config__.show())

# 版权声明：本文为CSDN博主「weixin_39827506」的原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接及本声明。

# 原文链接：https://blog.csdn.net/weixin_39827506/article/details/111435520

import numpy as np

from time import time

times = 10

a = np.random.rand(100000000)

start = time()

for t in range(times):

    for i in range(100):

        b = np.resize(a,(1000,200,500))

end = time()

print('running time is :%f'%(end/times-start/times))

print(np.__config__.show())

# 版权声明：本文为CSDN博主「weixin_39827506」的原创文章，遵循CC 4.0 BY-SA版权协议，转载请附上原文出处链接及本声明。

# 原文链接：https://blog.csdn.net/weixin_39827506/article/details/111435520

import numpy as np

import cv2

from time import time

def rgb2gray(rgb):

    return np.dot(rgb[..., :3], [0.2989, 0.5870, 0.1140])

def myImageFilter(img_in, h):

    img_w = img_in.shape[0]

    img_h = img_in.shape[1]

    padding_width = int((h.shape[0] - 1) / 2)  

    padding_height = int((h.shape[1] - 1) / 2)  #求padding步长

    new_image = np.zeros([img_w + 2 * padding_width, img_h + 2 * padding_height])

    new_image[padding_width : (img_w + padding_width), padding_height : (img_h + padding_height)] = img_in   #创建一个新的temp图像，边框置0，中心为原图

    res = np.zeros([img_w, img_h])      #结果图像

    h = np.fliplr(np.flipud(h))

    for i in range(padding_width, img_w + padding_width):           #使用最普通的均值滤波

        for j in range(padding_height, img_h + padding_height):

            res[i - padding_width, j - padding_height] = np.sum(np.multiply(

                new_image[i - padding_width:i + padding_width + 1, j - padding_height:j + padding_height + 1], h))/(len(h)**2)

    return res

   

if __name__ == '__main__': # 主函数s

    image_path = "test.jpg"

    img = cv2.imread(image_path)

    gray = rgb2gray(img)

    h = np.ones((3, 3)) #卷积核

    strat = time()

    new_image = myImageFilter(gray, h)

    end = time()

    print("filter cost time:",(end - strat))

    print(np.__config__.show())

附一个自己的b站视频 ，会有比较详细的测试过程图和视频AMD,Intel,Windows,openblas,mkl究竟谁会负优化numpy呢？ （实测多款处理器多个环境但不好给结论）_哔哩哔哩_bilibili