Created by: tensor-tang

enable MKLDNN test on AVX fix C++ part of #10682 (closed)

Steps to run

Take resnet of image_classification as an example：

1. make test ARGS="-R test_image_classification -V" to generate the CPU model.

2. ./build/paddle/fluid/inference/tests/book/test_inference_image_classification_resnet --dirname=./build/python/paddle/fluid/tests/book/image_classification_resnet.inference.model --batch_size=2 --repeat=10 --use_mkldnn=True