[Graph][Optimization]split+concat fusion to improve performance

[shanzhou2186]

split+concat fusion optimization
Goal
Optimize performance through split+concat fusion

Problem Description
In some of recommendation model, there is potential performance gain through split and concat fusion.

The step to reproduce the performance issue will be updated later.

Requirement Details

• Fusion 2 operators split and concat into 1 operator. Both of the forward and backward need to be covered. And make sure it could be applied in the real model at least.
• Datatype: FP32 is needed.
• Follow grappler mechanism https://www.tensorflow.org/guide/graph_optimization
• Attached code can be used to reproduce it.
  
  training-attention-mlp.py.txt

Test

• Unit test code and benchmark is needed.
• Using 1 model from model zoo to validate the performance gain. The performance data and analysis result could be described and reproduced.

Code Style and commit

• C++ and python: Keep aligned with DeepRec code.

Maintain

• All of the issue and bugs related with this op need to be covered in the future.

Definition of Done

• Run successfully in DeepRec and could get better performance.
• Integrated into DeepRec successfully and commit the code follow DeepRec commit standard.