* Refine adam op, test=develop

* Fuse kernels together to reduce cpu time.

* Refine paddle enforce, test=develop

* Remove some comments, test=develop

* Refine code,test=develop

* Refine cuda kernel, test=develop

* Refine code according to comments, test=develop

* Correct CPU gradients of the argsort op, form a network to test its forward and backward process, test=develop

* fix dynamic threshold error in test_argsort_op, test=develop

As the title

* register fp16 for assign op, test=develop

* add op test for fp16, test=develop

* add partial_sum_op, test=develop

* modify the Paddle Error Message, test=develop

* modify the Paddle Error Message, test=develop

* modify the bug for python3, test=develop

* modify the ut for ci, test=develop

* mv to contrib, test=develop

* use check_variable_and_dtype, test=develop

* fix ci, test=develop

* fix conflict, test=dvelop

* add partial concat, test=develop

* fix the conflict, test=develop

* fix the error, test=develop

* rm SSE4, test=develop

* add partial_concat, test=develop

* fix the grids and blocks, test=develop

* fix the Paddle_Enforce, test=develop

* fix the doc of op, test=develop

* fix the doc, test=develop

* fix the doc of the op, test=develop

* replace -1 with None, test=develop

test=develop

* add sync communicator and implement

1. Add PADDLE_ENFORCE to Check Sequence Length of RecurrentOp.
2. Also enrich PADDLE_ENFORCE error messages.

* improve the mul_mkldnn_op line coverage
test=develop

* remove fp32 mul mkldnn kernel
test=develop

* locally refactoring
test=develop

* change according to reviews
test=develop

* fix trt log
test=develop

* fix comments
test=develop

* Add TopK Op Grad CPU&GPU Kernel test=develop

* Add TopK Op Grad, modify grad op maker test=develop

* Add TopK Op Grad, modify grad op maker test=develop

* Add TopK Op Grad, modify PADDLE_ENFORCE test=develop

* Add TopK Op Grad, modify PADDLE_THROW test=develop

* Add TopK Op Grad, modify unittest test=develop

* fix ngraph top k op unittest test=develop

* update ops's unittest of elementwise_pow, elementwise_max, elementwise_min, scale and sqrt
1. update elementwise_pow, elementwise_max and scale's unitests with input data type (float32 -> float64)
2. fix bug that the elementwise_pow doesn't meet threshold requirements with tackling float64 data
3. remove sqrt from op_accuracy_white_list.py
4. update the unittests of elementwise_pow, elementwise_max and elementwise_min ops that their input data shape over 100
5. test=develop

* modify the writing style according suggestions
test=develop

* 1. optim multihead matmul: fuse three fc to multihtead matmul

test=develop

* fix conflict
test=develop

* fix comments
test=develop

* Add support for dynamic_decode(while) training. test=develop

* Fix assign_op and tensor_array_read_write_op after solving conflict. test=develop

* Fix test_rnn_decode_api.py. test=develop

* Refine docs for apis in rnn.py. test=develop

* Adjust outputs of dynamic_decode. test=develop

* Remove the force_cpu update in assign_op. test=develop

* Remove the force_cpu update in assign_op. test=develop

* Make RNNCell.get_initial_states support batch_dim_idx argument. test=develop

* Rename _create_array_outof_while as _create_array_out_of_while in rnn.py.
test=develop

* a test for Ernie QAT INT8 accuracy check

test=develop

* Remove NLP comparison test to split PRs

test=develop

* Fix typo and tabs, delete commented lines

test=develop

* re-combine the 2 PRs, test=develop
Co-authored-by: NMichał Gallus <sand3r@interia.eu>
Co-authored-by: Nbingyanghuang <33643817+bingyanghuang@users.noreply.github.com>

* support slice double grad, test=develop
* merge two doublegradopmaker to one doublegradopmaker,test=develop
* change the shape of slice_OP's unittest, test=develop

Refine PaddleBox Framework, Main functions: 
* Add MetricMsg util class, which can calculate metrics like AUC, bucket_error, COPC.
* Replace FeedPass with new interface: BeginFeedPass & EndFeedPass
* Refactor Pull/Push Sparse Function in box_wrapper.
* Use CUDA Kernel to copy keys and copy feasign between tensor and boxps struct.
* Cache copied keys in pull sparse in order to reuse it in push period.

* Refine code, fix select tile error,test=develop

* Refine element type and some comments, test=develop

* Refine comments and gpu utils, test=develop

* Remove some useless condition

* Refine floor and ceil, test=develop

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

支持不依赖nccl进行编译。[1/2]

多卡下，如果没有打开WITH_NCCL开关编译，多卡不能通信，则只能选择一张卡使用。
Co-authored-by: N石晓伟 <39303645+Shixiaowei02@users.noreply.github.com>

Compile without nccl deps. [1/2]
Co-authored-by: N石晓伟 <39303645+Shixiaowei02@users.noreply.github.com>

test=develop

Fix the  integer overflow problem in the op of sequence2batch, change the int32_t to size_t，
In the /paddle/fluid/operators/math/sequence2batch.h#L122.

* support weight quantization in post_training_quanzitaion, test=develop
* add test for weight quantization, test=develop

test=develop

* optimize interpolate op, test=develop

Correct the use of DeviceContext in unittest sequence_pooling_test and sequence_padding_test (#22456)

* Add log in memory::Copy for debug purpose.

* Change to use context in DeviceContextPool directly in sequence_pooling_test, instead to new one.

* Change to use context in DeviceContextPool directly in sequence_padding_test, instead to new one.
test=develop

* Change the type of second_dim from size_t to int64_t.
test=develop

cmake选项中添加了WITH_NCCL，显示指定是否编译NCCL的部分代码，WITH_NCCL默认打开，但如果WITH_GPU为OFF，则关闭WITH_NCCL

添加了PADDLE_WITH_NCCL定义

单机单卡能够关闭NCCL编译，多卡的话需要默认打开NCCL，如果关闭NCCL，则只能使用单卡
Co-authored-by: N石晓伟 <39303645+Shixiaowei02@users.noreply.github.com>

* Sigmoid bug fix, test=develop

* fix code format

test=develop
Co-authored-by: NManjunath Bhat <manjunathbhat9920@gmail.com>