* support selectedrows allreduce in multi-cards dygraph, test=develop

* remove useless import modules in unittests, test=develop

* add nccl cmake to get nccl version, test=develop

* add if-condition to compiled correctly, test=develop

* add detail version parseing for old nccl, test=develop

* polish camke details, test=develop

* fix remove test cmake error, test=develop

* fix cmake condition, test=develop

* change unittest camke list, test=develop

* fix unittest cmake rule, test=develop, test=framep0

The random failure at Windows may due to some random gt_boxes can cause some numbers in YoloV3 to be negative thus access the invalid memory. This PR tries to solve it.

Add Yolov3 as ProgramTranslator Unit Test. The YoloV3 code is referred from PaddlePaddle/models/dygraph/yolov3

* add the support of device index for device_guard.

* add mock barrier all
test=develop

* fix
test=develop

* fix
test=develop

* fix
test=develop

* add queue_generator_op, dequeue_op, enqueue_op and ut, test=develop

* Fix the field length in LoD scenario
* Fix the missed lod info when copy tensor in dump field
* Add some log to make debug easy

* add amp_check_finite_and_scale op, test=develop

* add cpu kernel, test=develop

* use bool, test=develop

* follow comments, test=develop

* fix bilinear initializer, test=develop

* add test_bmn_model test=develop

* remove random test=develop

* support for iter & enumerate varbase, test=develop

* revert IsControlFlowVisitor change, test=develop

* Support return variable in only one of if body or else. 

* remove after_visit in IfElseTransformer.

* Modify the result of get_name_ids in test_ifelse_basic.py 

* Add unittest to test the new case. 

* Modify code according to reviews. 

* Support convert_while_loop. 

* Comment code that not supported 'if' in test_break_continue. 

* Convert int into tensor to support 'if' stmt in for/while loop. 

* Add unittest to test all cases of convert_logical_XX. 

* Add unittest to test all cases of convert_while_loop. 

* Fix bug in LogicalOpTransformer. test=develop

* add crypto helper for paddle, test=develop

* cryptopp.cmake bug fixed, test=develop

* remove debug build type, test=develop

* fixed CMakeLists for new target, test=develop

* fix CI bug, test=develop

* add cmake option flag DWITH_CRYPTO, test=develop

* add crypto api for python, test=develop

* Revert "add crypto api for python, test=develop"

This reverts commit 3a1cfa9d.

* Revert "Add crypto api (#24694)"

This reverts commit 5a7a517c.

* Revert "Revert "Add crypto api (#24694)""

This reverts commit f952b19f.

* fixed cryptopp cmake building error, test=develop

* change WITH_CRYPTO building option to OFF, test=develop

* â€fixed cipher test failed, test=develop

* "add crypto api for python, test=develop"

This reverts commit 83fb55c0.

* travis CI bug fixed, test=develop

* fixed test in python3

* test=develop

* fixed unittest, test=develop

* add histc operator, test=develop

* update english doc to 2.0 API, test=develop

* update API from histc to histogram, test=develop
Co-authored-by: Nroot <root@yq01-gpu-255-129-15-00.epc.baidu.com>

Enhance error message of checkpoint_notify_op, fake_init_op gen_nccl_id_op and listen_and_serv_op (#24554)

test=develop

The PR: https://github.com/PaddlePaddle/Paddle/pull/24651 seems causes new random failure of unit test test_parallel_executor_seresnext_base_cpu. The reason is that smaller batch size causes random optimization of neural network. I distinguished cpu/gpu batch size to fix the unittest.

* support nested input and output test=develop

* remove code of convert type(output) in unittest test=develop

* add warning test=develop

添加loss自适应的学习率衰减策略。

* Fix bug of different dtype in dygraph math_op_patch, test=develop

* support np.dtype and str, test=develop

* add unit test, test=develop

* [dy2stat] Support len

* add convert_call func

* refine code test=develop

* Revert "Fix test_bert on GPU (#24692)"

This reverts commit 62222bf4.

* fix random bug in bert unittest, test=develop

* Support AucRunner in PaddleBox
* update some code style

* remove pow in speed up in dequantize_log test=develop

* remove pow in speed up in dequantize_log test=develop

* fix unittest test=develop

* support NHWC in Pool2D, test=develop

* add unittest, test=develop

* fix unittest, test=develop

* fix typo, test=develop

* follow comments, test=develop

* refine comments, test=develop

DataLoader makes the data diff even if the data of reader is the same on CUDA place. This PR doesn't use DataLoader to pass the test. we will use DataLoader back after we fix it.

[Dy2Stat]Support to transform sequence assignments and multi-target assignments to normal assignments (#24643)

* test=develop update interpolate & upsample api

As discussed with QA, we will use p4 machine for unit test and the GPU on those machine may not have enough GPU, which can cause "test_parallel_executor_seresnext_base_gpu" failed. So I decrease the batch size.

* fix bug of varbase.__getitem__, test=develop

* fix bug of float and other type, test=develop

API/OP(sequence_first_step, sequence_last_step, sequence_mask, beam_search, beam_search_decode) error message enhancement (#24590)

In the past, the test_cond will fail with 2% probability and easy to re-produce.

Now I re-run 300 times and no failure occurs. The probability of still has the failure is (1 - 2%) ^ 300 ~= 0.00004. We can say the random failure disappears. Maybe someone fixed some bugs in PE.