CUDA学习3 Max pooling （python c++ cuda）

1.Python

在CNN4 参数优化中有一个CNN模型，其中的限速步是max pooling。

如下所示，Python中运行一个50*100*24*24的max pooling需要3秒。

import numpy as np
import time

def simple_pool(input, ds=(2, 2)):
    n, m, h, w = input.shape
    d, s = ds
    zh = h / d + h % d
    zw = w / s + w % s
    z = np.zeros((n, m,zh,zw))
    for k in range(n):
        for o in range(m):
            for i in range(zh):
                for j in range(zw):
                    maxd = -10000
                    for u in range(min(d,h-d*i) ):
                        for v in range(min(d,w-d*j)):
                            if input[k,o,d*i+u,d*j+v]>maxd:
                                maxd=input[k,o,d*i+u,d*j+v]
                    z[k, o, i, j] = maxd

    return z

N,M,H,W=[50,100,24,24]
a=np.reshape(range(N*M*H*W),(N,M,H,W))*0.01
start_time= time.time()
out_data=simple_pool(a)
print "Cost:",time.time()-start_time,"s"
print out_data[0,0,0,:10]

"""
Cost: 3.08899998665 s
[ 0.25  0.27  0.29  0.31  0.33  0.35  0.37  0.39  0.41  0.43]
"""

2.C++

采用c++，仅需16~30ms。