add ips for dygraph mobilenet and resnet models

[luotao1]

增加IPS输出：

• mobilenet-V1：在CPU上运行，batch_size=4，输出如下：

[Epoch 0, batch 10], avg_loss 0.00000, acc_top1 1.00000, acc_top5 1.00000, batch_cost: 1.33183 s, net_t: 0.44647 s, backward_t: 0.78070 s, reader_t: 0.04040 s, ips: 3.00338 images/sec
[Epoch 0, batch 20], avg_loss 0.00000, acc_top1 1.00000, acc_top5 1.00000, batch_cost: 1.23914 s, net_t: 0.40186 s, backward_t: 0.78185 s, reader_t: 0.00170 s, ips: 3.22804 images/sec
[Epoch 0, batch 30], avg_loss 0.00000, acc_top1 1.00000, acc_top5 1.00000, batch_cost: 1.21990 s, net_t: 0.38759 s, backward_t: 0.77668 s, reader_t: 0.00150 s, ips: 3.27896 images/sec
[Epoch 0, batch 40], avg_loss 0.00000, acc_top1 1.00000, acc_top5 1.00000, batch_cost: 1.25072 s, net_t: 0.40059 s, backward_t: 0.79487 s, reader_t: 0.00151 s, ips: 3.19816 images/sec

• resnet： 在CPU上运行，batch_size=2，输出如下：

[Epoch 0, batch 0] loss 4.64205, acc1 0.00000, acc5 0.00000, batch_cost: 4.26375 s, reader_cost: 0.46242 s, ips: 0.46907 images/sec
[Epoch 0, batch 10] loss 44.68754, acc1 0.00000, acc5 0.00000, batch_cost: 2.80404 s, reader_cost: 0.00015 s, ips: 0.71326 images/sec
[Epoch 0, batch 20] loss 29.44468, acc1 0.02381, acc5 0.02381, batch_cost: 2.85184 s, reader_cost: 0.00013 s, ips: 0.70130 images/sec