* add fluid.device_guard to specify the device type for Op

* update ScopeBufferedSSAGraphExecutor&AsyncSSAGraphExecutor&ThreadedSSAGraphExecutor&FastThreadedSSAGraphExecutor&ParallelSSAGraphExecutor&ParallelExecutor for fetching unmerged results.

* add the unit test for fetch_unmerged.

* update ut for multi-card and multi-cpu.

* add the error message and the user suggestion in FetchOpHandle. test=develop

* user can call dataset.set_download_cmd to set its customized download cmd
* add UT to cover this scenario

* Add the codegen and auto fusion for sum Op  in fusion group

* support cond in clone, test=develop

* refine code, test=develop

* refine code, test=develop

* follow comments, test=develop

* refine code, test=develop

* Add two types of Metric Calculator: MultiTaskCalculator & CmatchRankCalculator.
* Add a config for DynamicAdjustChannelNum function to denote whether we will discard the remaining instances when they are not be distributed evenly.
* Remove CPU code in Pull/PushSparse and we will add it back when testing it fully.
* Fix some known issues: such as copying persistable vars after one epoch running.

* Add an interface of disabling FC padding
* fix bert regression
* polish fc padding interface
* recover pass function
* fix argument error
* fix mkldnn error

* add sync communicator and implement

* add python flag to control profile level test=develop

当一个模型中有多个fc_lstm子图的时候，且其中fc共用了同一个persistable的bias，此时不应该将bias节点删除，只将非persistable的节点去除即可。

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

test=develop

* fix conflict
test=develop

* fix comments
test=develop

test=develop

* add thread barrier for the compiled program

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.

* update

* update test=develop

* update compile set test=develop

* update compile set test=develop

* update test=develop

* update test=develop

* update test=develop

* update compile setting test=develop

* update compile setting test=develop

* update run demo test=develop

* update test=develop

* update test=develop

* fix test=develop

* update test=develop

* update test=develop

* update test=develop

* update test=develop

* update test=develop

* update test=develop

* update test=develop

* update test=develop

* update test=develop

* update format test=develop

* update format test=develop

* update style test=develop

* update style test=develop

* change style test=develop

* change style test=develop

* change style test=develop

* add dataset unittest test=develop

* update test=develop

* update for record test=develop

* udpate style for record test=develop

* update for record test=develop

* update for record test=develop

* update for record test=develop

* fix format test=develop

* update test=develop

* update test=develop

* update test=develop

* update test=develop

* update test=develop

* 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>

This PR makes assign op support LoDTensorArray and enable the loop_vars in
while_loop to support tuple or list.

* Add the first implememtation of fusion_group op #19621 (#3)

* 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

* Enable generating code for a given subgraph. #21126 (#4)

* Enable generating code for a given subgraph.

* Support sorting the subgraph.

* Remove the rearange of expressions because we use the sorted subgraph directly.

* Enable generating code for a subgraph which is composed of grad ops.

* Use expression information to check the accuracy in unittest.

* Separate load and store from computation expressions.
test=develop

* Improve the loading statements in generated codes.
test=develop

* Remove unused arguments from formal list.
test=develop

* Enable the detection of subgraph of grad ops.

* Generate code for detected subgraph in fusion_group_pass.

* Add an option in BuildStrategy to enable fusion_group_pass and add unittest.
test=develop

* Fix a bug when checking whether the shape of all inputs are the same.

* Add debug information.

* Remove subgraph_detector from inference/analysis to the common framework/ir directory. (#5)

test=develop

* Call subgraph_detector in fusion_group pass.
test=develop

* Disable fusion_group when WITH_GPU is OFF.
test=develop

* Refine all PADDLE_ENFORCE message.
test=develop

* Fix the case that some inputs are not defined in grad ops, and set op_role for fused op.
test=develop

* Follow review comments.
test=develop

* Add dequant scale squash

test=develop

* Correct dequant-scale squash test

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>

* fix copy table bug of lost some feasign
* test=develop

* add GeneralRoleMaker which is for general usage
* test=develop

* Enable quantize to reorder to nchw as well

* Correct FC MKL-DNN input dim requirements to accept 3D

* Improve DNNL FC format, error and 3D input handling

test=develop

* Improve error checking in FC

test=develop

* Improve PADDLE_ENFORCE messages in fc-related files

* Remove data layout attribute from obligatory pass args

test=develop

* Fix message in fc_mkldnn_pass to be logically correct

test=develop

* Implement a common python unittest to test the ir passes.
test=develop

* Save the results in np.array and support to startup on CPU.
test=develop

* Fix the unittest.
test=develop

* Add check_program to check whether the optimized program is different from the origin one.
test=develop

* Remove the inferface all_ops.
test=develop

* Add exception test in pass_test.
test=develop

* 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

faster build by reduce by-product, reduce linking library and fix compile warning of std=c++11 (#22164)

* add collective communication library in fleet to replace mpi
* test=develop