* add group norm plugin

* fix compile problems

* move concat axis check to trt op teller

* add nbDims for scale and bias nv dims

* add group norm unit test

* fix unittest

* add trt version restriction for group norm op teller

* fix unittest

* fix entry

* fix distributed lookup table fuse case

* fix entry bug at first time

* move entry from paddle.fluid -> paddle.distributed

* fix ut with paddle.enable_static()
Co-authored-by: Nmalin10 <malin10@baidu.com>

* add error msg when dtypes of operator are not same

* add error msg when dtypes of operator are not same

* change error msg to warning msg when dtypes of operator are not same

* modify test case to fit for python2

* Support 'bool' and 'int' for attention mask.

* Update docs.

* Add unittest for Transformer.

* fix bugs.

* add default argument  for paddle.save/static.save

* edit documentation of

* Add comments for special processing for protocol=2 and protocol=3.

* Update python/paddle/fluid/io.py
Co-authored-by: Nlanxianghit <47554610+lanxianghit@users.noreply.github.com>
Co-authored-by: Nlanxianghit <47554610+lanxianghit@users.noreply.github.com>

**Problem**
In our old shape transformer logic, if user write:
```
s = tensor.shape
...
y = paddle.some_api(s)
```
Dy2stat will change it to
```
...
y = paddle.some_api(convert_var_shape(tensor))
```
However it will cause fatal bug if user changes the shape of `x` after assign. For example:
```
s = tensor.shape
...
tensor = paddle.some_change_shape_api(tensor)
...
y = paddle.some_api(s)
```
Then the Dy2stat will get wrong result because the code is translated into:
```
tensor = paddle.some_change_shape_api(tensor)
...
y = paddle.some_api(convert_var_shape(tensor)) # tensor shape has been changed, not origin `s` value
```

**Solution Logic**

It can not be solved in the old logic, so I refactoring tensor_shape_transformer logic. Now we will use `s` to store shape attribute and generate a var `s__STATIC_CONVERT_VAR_SHAPE_SUFFIX` to store static shape API `shape(tensor)`
```
s = tensor.shape
...
y = paddle.some_api(s)
```
Dy2stat will change it to
```
s = tensor.shape
s__STATIC_CONVERT_VAR_SHAPE_SUFFIX = shape(tensor)
...
y = paddle.some_api(choose_shape_attr_or_api(s, s__STATIC_CONVERT_VAR_SHAPE_SUFFIX ))
```
In this case, the code is consistent with origin dygraph meaning and it fixed the change after assign bug.

**Code Key Note**

To help reviewers, the key change of this PR is changing `self.name_to_var_shape` from "mapping name to shape node" to "mapping name to its STATIC_CONVERT_VAR_SHAPE_SUFFIX name", then if a variable name has the SUFFIX, we can choose to use attribute shape or shape api. Other changes go with the key change.

**Consideration**
The issue of this PR is that we store extra static `shape` API result, will it harms the speed of Dy2stat? In some cases it will, but we argue that the benefit would be greater than the cost.

1. The extra calling to static `shape` API will happen when coder assign among shape variables. Take the following dygraph code as an instance:
```
s1 = tensor.shape
s2 = s1
s3 = s2
...
```
Then we called extra static `shape` APIs again and again, however users seldom write code like this.

2. If the shape variable is used a lot, for example:
```
s = tensor.shape
y1 = paddle.some_api1(s)
y2 = paddle.some_api2(s)
y3 = paddle.some_api3(s)
```
Our old logic will create 3 shape APIs but now just 1. This is more common user code pattern. In fact, if reviewers take a look at the current unit test in this PR, you could see the op numbers decrease after this PR. So we argue that this PR can also improve speed in this code pattern.

* fix dist fleet ctr ut

Change-Id: I59bf5123c7bd47bd0e8f1ca2a26295257597c0f5

* fix dist fleet ctr ut

Change-Id: Iafcdd172364be47fe67b753774ce09af050bcbce

* Update CMakeLists.txt

add squeeze_op/unsqueeze_op on kunlun;fix conv op and parallel executor;optimize lookup_table op (#31056)

* add squeeze_op/unsqueeze_op on kunlun; fix conv op and parallel executor on kunlun; optimize lookup_table op on kunlun

* update squeeze/unsqueeze op

* test=develop, save/load, shrink
Co-authored-by: NseiriosPlus <tangwei12@baidu.com>

* export paddle.static.normalize_program method. test=develop

* fix ut coverage.test=develop

* [static setitem] support the index step > 1. tensor_a[::3] = value

* [static setitem] support the index step < 0. Eg: tensor_a[::-3] = value

* [static setitem] support the index is Tensor. eg: tensor_a[tensor_3:0:-1] = value

* Add op version.

As the title, when slice_node like 1:3 being passed to idx of convert_var_shape, it will cause syntax error because a function cannot take this as argument. This PR fixed it.

* - added Reshape grad bf16

* - Added reshape grad bf16

* - cosmetics in py

* fix dataloader collate return list mix tensor and numpy array. test=develop

* add  parameter in roi_align op

* Add conv transpose BF16

* Share function GetWeightsTz

* Adjust to review and fix op compatibility

* Add bias to unique handler name

* Remove errors related to paddle enforce

* Add conv2d_transpose to bf16 list and kernel refator

Refine fake_interface Error Message

Dy2stat didn't support tuple as iteration variable in the past. This PR added there main cases:

       1). Non-enumerate case: for var1, var2 in var|var.numpy() will be re-written as:
          for FOR_ITER_TUPLE_PREFIX_x in var | var.numpy():
            var1 = FOR_ITER_TUPLE_PREFIX_x[0]
            var2 = FOR_ITER_TUPLE_PREFIX_x[1]
        2). Enumerate out tuple case: for t in enumerate(var|var.numpy) will be rewritten as:
          for FOR_ITER_TUPLE_INDEX_PREFIX_x, FOR_ITER_TUPLE_PREFIX_x in enumerate(var|var.numpy):
            t = (FOR_ITER_TUPLE_INDEX_PREFIX_x, FOR_ITER_TUPLE_PREFIX_x)
        3). Enumerate inner tuple case: for i, (var1, (var2, va3)) in enumerate(var|var.numpy()) will
        be re-written as:
          for i, FOR_ITER_TUPLE_PREFIX_x in var | var.numpy():
            var1 = FOR_ITER_TUPLE_PREFIX_x[0]
            var2 = FOR_ITER_TUPLE_PREFIX_x[1][0]
            var3 = FOR_ITER_TUPLE_PREFIX_x[1][1]

fix the broadcast for the large second input 

As the title

* fix trt plugin clone and initialize bugs

* fix unit test error

* enable trt in ci py3

* update unittest timeout

* Add cache for Transformer encoder.

* Bug fixed.

* add unittests for transformer encoder.