From: @jjfeing
Reviewed-by: @zhoufeng54,@kisnwang
Signed-off-by: @kisnwang

From: @kisnwang
Reviewed-by: @zhoufeng54,@jjfeing
Signed-off-by: @jjfeing

2. Dump the end graph in gpu session and cu session

From: @youui
Reviewed-by: 
Signed-off-by:

Fixed cpplint

From: @islam_amin
Reviewed-by: @john_tzanakakis,@yelihua
Signed-off-by: @yelihua

The basic idea is: exploits data dependency to control the execution order
of side-effect operations, and keep the semantics of ANF unchanged.

The ControlDepend primitive is removed and there are two primitives added:

1. UpdateState:
```
  a = Assign(para, value)
```
became:
```
  a = Assign(para, value, u)
  u = UpdateState(u, a)
```

2. Load:
```
  x = Add(para, value)
```
became:
```
  p = Load(para, u)
  x = Add(p, value)
  u = UpdateState(u, p)
```

From: @simson_wu
Reviewed-by: @chujinjin
Signed-off-by: @chujinjin

From: @liangzelang
Reviewed-by: 
Signed-off-by:

From: @HulkTang
Reviewed-by: 
Signed-off-by:

From: @chujinjin
Reviewed-by: 
Signed-off-by:

From: @hwjiaorui
Reviewed-by: @jjfeing,@kisnwang
Signed-off-by: @kisnwang

asyn dynamic shape

format

control flow error

format