* Add bf16 matmul, fc, elementwise add and mul

* Correct unit test

* Fix wrong out dtype from helper.input_dtype

* add unittest

* remove disable_static in op_test

* fix param name typo

* 1.add xpu slice op 2. add xpu top_k op 3.modify xpu cast to new api

* 1.add xpu slice op 2. add xpu top_k op 3.modify xpu cast to new api

* add lstm, simple rnn op kernel

* fix the test_lstm for the rnn op

* change func name

* fix forward postprocess bug

* add gru forward, backward code

* remove unittest.skipIf; use a big rnn op instead of combination op

* fix input doesn't have gradient bug

* add eigen lstm forward, backward
Co-authored-by: Nwawltor <fangzeyang0904@hotmail.com>

* adjust kunlun header file
*test=kunlun

* update kunlun unittest
*test=kunlun

* update xpu unitest
* test = kunlun

* update xpu unittest
* test=kunlun

* update xpu unitest
* test=kunlun

[Dynamic-to-Static] Fix bug in loop_transformer: loop vars should contain the var from ancestor-for-node (#28735)

* Add multi_gru op and tests

* removed redundant disable_dygraph()

* modfied the timeout value for 5 ut

* modified timeout value for test_resnet

[Dynamic-to-Static ErrorMessage] Support dy2stat error message when call jit.save and polish error message (#28713)

* Support dy2stat error message when call jit.save;

* Polish dy2stat error message:
  (1) the original dygraph code is marked with (* user code *) ; 
  (2) "In user code:" -> "In transformed code:"

Add new operator log10

faster the compare ops dygraph model speed 

* add lars to fleet meta optimizer

* add lamb to proto

* add lamb to fleet meta optimizer

* fixed syntax bug

* fixed syntax bug

* fixed syntax error in lamb, add config setter of lamb in distributed_strategy

* trigger unitest to rerun

* add new unitest func for lamb

* revise unitest for lars and lamb

* revise dgc meta unitest

* revise lars document in distribute_strategy

* revise lars lamb document in distributed_strategy.py

* revise lars lamb document in distributed_strategy.py

* add weight decay exclude logic to lars

* restore optimzier.py

* restore optimizer.py as develop except lars

* add epsilon and exclude fn to distributed_sttrategy

* add lars epsilon

* revise unitest for fleet lars and lamb

* revise lars lamb unitest for CI coverage

* revise lars argument api

* revise lars argument api

* revise lars argument api

* revise api doc of lars

* fix op role

* add sharding save and add_sync_comm_for_test function

* add comm_analyse to utlis

* revise sharding_utils

* add sharding saving unittest

* revise sharding utils for unittest

* update interpolate, test=develop

* fix coverage, test=develop

add two paddle-2.0 apis: paddle.static.io.save_inference_model and paddle.static.io.load_inference_model (#28606)

* add two apis: paddle.static.io.save_inference_model and paddle.static.io.load_inference_mode, which are campatible with paddle.fluid.io.save_inference_model and paddle.fluid.io.load_inference_model respectively.

* add unittest for new save_inference_model and load_inference_model. test=develop

* enhance doc. test=develop

* add paddle.enable_static() to test_inference_model_io.py. test=develop

Fix a bug that used in PaddleGAN model which used `isinstance` in a for loop

* adapt pad const

* fix comment and rm fluid import

* rm stdout

* fix note

* fix output dtyp inconsistent with input

* refine code

fix the gradient bug for the topk v2 

* add WeightedRandomSampler. test=develop

* Add checkpoint to quantize

* Change bfloat16 option

* add mkldnn inplace op version test

* update mkldnn_inplace fuse pass

* update the inplace test

* fix test_multinomial

* fix test_multinomial add 0 prob

* DataLoader support not auto collate batch. test=develop

PaddleSeg uses tuple as parameter of transpose in dygraph code:
https://github.com/PaddlePaddle/PaddleSeg/blob/release/v0.7.0/dygraph/paddleseg/models/danet.py#L152

However, in dy2stat, static code doesn't support the perm as a tuple. This PR fixed it.

* Fix gradients with ignore_idx in softmax_with_cross_entropy.
test=develop

* Fix gradients with ignore_idx in softmax_with_cross_entropy on cpu.
Remove softmax_with_cross_entropy from op_threshold_white_list.
test=develop

* Fix test_softmax_cross_entropy_op.py.
test=develop

I found the unittest failed due to batch size. Maybe the reason is that our CI machine has limited memory. I decreased the batch size.

* update, test=develop

this is a preemptive measure, the unit test was failing occasionally with
cudnn internal error

* add onnx export module, test=develop

* add unit test for paddle.onnx.export

* adjust api & doc

* fix some typo

* skip_layernorm_op done

* add unittest

* slice op convertor support trt < 6

* skip_layernorm only work in ernie

As the title