Created by: yihuaxu

According to the performance status of DAM model, implemented the pass of convolution 3D and bias for mkldnn to accelerate the data processing.

Platform: Intel(R) Xeon(R) CPU E5-2699 v3 @ 2.30GHz Model Path: third_party/inference_demo/small_dam/model Batch Size: 300 Command: ./paddle/fluid/inference/tests/api/test_analyzer_small_dam --infer_model=third_party/inference_demo/small_dam/model/ --infer_data=third_party/inference_demo/small_dam/data.txt --gtest_filter=Analyzer_dam.profile_mkldnn --paddle_num_threads=1 --repeat=1 --batch_size=300 --max_turn_num=1 --test_all_data --profile Data Source: third_party/inference_demo/small_dam/data.txt.

Dependency PR: 14707

The following is the comparison with the different scenarios.