* add half_async in the communicator
* fix DistributedStrategy

* remove unused inputs, test=develop

* remove unused inputs, test=develop

* update dtype, test=develop

* remove unused inputs, test=develop

* update op_use_default_grad_op_maker, tese=develop

* resolve conflicts, test=develop

* follow comments, test=develop

* update center_loss_grad, test=develop

* fix fsp_op, test=develop

* fix fsp grad op maker, test=develop

* update op_use_default_grad_op_maker.spec, test=develop

* test=develop, bug fix for sparse recorder

* add backward gradient computation for op argsort test=developo

* use pre-commit test=develop

* add bn and relu fuse pass

* add op attr assert and dtype assert

* fix some inputs&&outputs bugs for the fused op and pattern.

* add the unittest for fuse_bn_act_pass. test=develop

* use normative enforce statements. test=develop

* add the cpu test. test=develop

* add the support of batch_size=1 for the bn with relu op. test=develop

* add the error type for paddle throws. test=develop

* add fused_batch_norm_act and fused_batch_norm_act_grad to op_has_unsed_vars_white_list. test=develop

* test Optimizer in dygraph, test=develop

* add optest for Optimizer in dygraph, test=develop

* fix adagrad optimizer, test=develop

* fix dpsgd optimizer, test=develop

* fix test_optimizer.py, test=develop

* fix dpsgd optimizer, this op only support cpu, test=develop

* add optest for optimizer, test=develop

* add description for dpsgd, test=develop

* add rmsprop to white_list in unused_var_check.cc, test=develop

* polish code style, test=develop

* polish code style, test=develop

* delete seed attribute for DpsgdOptimizer, test=develop

* change testing to debugging, test=develop

stack's wait cost a lot of cpu time, use cuda kernel to do memory copy
will reduce cpu time.
Signed-off-by: Nzhaoyuchen <zhaoyuchen01@baidu.com>

1. Add a new input named batch_roi_nums for prroi_pool_op. batch_roi_nums includes the number of roi for each image in batch when rois is Tensor. This information is saved in rois's lod when rois is LoDTensor.
2. add grad check to prroi_pool_op and solve unnormal X grad diff in CPU.

windows conv_fusion failed as no kernel， explicit declare lambda
Signed-off-by: Nzhaoyuchen <zhaoyuchen01@baidu.com>

* add special way to add distribute vars， Update Pyramid hash op

* add erf op and python interface.

* add fp16 support for erf op.

* add unitests for erf op and its python interface.

* support elu activation double grad,test=develop

* delete the code commit in .cc,test=develop

* fix relu test unpass, test=develop

* add elu double grad kernel and unit test

* add caculate dX in elu double grad functor, test=develop

* update the commit code,test=develop

* add gelu plugin

* align trt bert with gpu

* add support for fused fc with relu,

* add unittest for bert trt

* add distributed_strategy

* register int/int64_t/float16 in  pow/square kernel,test=develop

* add abs/square/exp type,test=develop

* fix test_conv2d_ngraph for grad diff, test=develop

* register NoNeedBufferVarsInference for max_pool_grad_op, test=develop

* refine error message, test=develop

* fix numpy, test=develop

* disable test conv2d_ngraph_op, test=develop
Co-authored-by: NZhang Ting <709968123@qq.com>

* Add the dynamic load of nvrtc, and support runtime compiling of CUDA kernel using nvrtc.
test=develop

* Call CUDA driver api to launch the kernel compiled by nvrtc.
test=develop

* Disable for mac and windows.
test=develop

* Refine the codes to support manually specified num_threads and workload_per_thread.
test=develop

* Refine the CUDA kernel to support large dims.
test=develop

* Add DeviceCodePool to manage all device codes.

* Add the first implementation fusion_group op.

* Add unit-test for fusion_group op.

* Add the check of result.

* Add the check of nvrtc in unit-test.
test=develop

* Add comment to explain the inputs, outputs and features of fusion_group op.
test=develop

* Disable fusion_group op for mac and windows.
test=develop

* Make the compiling of device code return status instead of hanging up.
test=develop

* Add the check of whether there is CUDA driver library, and do not core dump when failing to call the CUDA driver API.

* Unify fusion_group_op's input and output names.
test=develop

* Add the check of CUDA driver library in unittest.
test=develop

* Refine the calling of PADDLE_ENFORCE.
test=develop

* test=develop fix generate_proposal_labesl op

* update error information about batch_norm_grad

* update bn,test=develop

* fix integer overflow in match_matrix test=develop

* fix integer overflow in match_matrix test=develop

* fix typo test=develop

* Refine multihead kernel, align block to 32

test=develop
Signed-off-by: Nzhaoyuchen <zhaoyuchen01@baidu.com>

* Refine log comments

test=develop
Signed-off-by: Nzhaoyuchen <zhaoyuchen01@baidu.com>

* add shuffle batch op, test=develop, test=document_preview

* fix size_t conflict and check_output test=develop, test=document_preview

* fix bug test=develop, test=document_preview

* add unittest of shuffle_batch layer test=develop, test=document_preview

* fix py coverage and op input type, test=develop, test=document_preview

* fix py coverage, test=develop

* fix en doc, test=develop

* move to contrib test=develop

* add unique_name test=develop

* invoke shuffle_batch in contrib.layers test=develop

* make reverse op support negative axis

* fix compile error in CUDA10 test=develop

* remove double in pad2d test=develop

* fix stat shape back in global auc scenario
* add UT to cover global auc