* support batch axis concatenation when using dynamic shape

* opteller can't return true early, or some test will not be executed

* remove eigen deps dof float16

* add cstdlib header

* replace stdlib header by cmath

Quick fix to https://github.com/PaddlePaddle/Paddle/issues/13860

* init commit

* move eigen of bfloat16

* add complex header

* add dict/str/list supprot for DataLoader. test=develop

* fix engine key in trt int8 calibration

* fix unit test

* add serialize unittest

* fix element_op trt plugin serialize bug

* support duplicable op inout

* add costom concat op test

* Different machine generate different binary file, remove md5 check

* remove unnecessary functions

* heter bug

* format

* format

* add softmax_switch for softmax_with_cross_entropy_op, test=develop

* delete using EigenMatrix in softmax_with_cross_entropy_op.h, test=develop

* add REGISTER_OP_VERSION for softmax_switch attr of softmax_with_cross_entropy_op, test=develop

* remove the send/recv of tensor size, but users have to specify the shape of the received var explicitly.

* bugfix for test_cholesky_op

* bugfix for test_compare_op

* bugfix for lookup_table_op

* bugfix for affine_channel_op

* - introduced fix onednn 2.2 version

* - compilation fix

[Kunlun]Multi xpu dygraph performance optimization , add distributed.spawn support for multi xpu and some bug-fixes (#31130)