for循环比整体操作慢几百倍
import time
import torch
import os
os.environ["CUDA_VISIBLE_DEVICES"] = '0'
t1 = time.time()
x = torch.rand(200,200)
for i in range(x.size(0)):
for j in range(x.size(1)):
x[i:j]*=2.0
cnt = time.time() - t1
print(cnt)
t1 = time.time()
x[:,:]*=2.0
cnt = time.time() - t1
print(cnt)