* fix elementwise_pow bug on integer, test=develop

* use llrint to support elementwise_pow_grad, test=develop

* add some tests, test=develop

* revert grad functor, test=develop

* patch math method for varbase using auto-generated op functions, test=develop

* clean code that handles batch_size, test=develop

* follow comments, test=develop

* follow comments, test=develop

* code clean, test=develop

* add backward unittest test=develop

* add sub-branch in test_backward.py test=develop

* refine code, add comment test=develop

* reconstruct TestBackward Class test=develop

* fix typo of comment test=develop

* update op test framework

affine_grid, label_smooth, spectral_norm, warpctc,
nearest_interp, data_norm, match_matrix_tensor,
var_conv_2d, fused_embedding_seq_pool

test=develop

sequence_expand_as, squared_l2_distance, gather_nd,
center_loss, rank_loss, conv_shift, spp,
modified_huber_loss, smooth_l1_loss, multiplex,
sequence_softmax, nce, huber_loss, group_norm,
kldiv_loss, hinge_loss, expand_as

test=develop

* Add reshape int8 op

test=develop

* Change test to CPUPlace

test=develop

* Correct tests

test=develop

* fix test for PY3 test=develop

* reduce file changes test=develop

sequence_unpad, expand, pad, pad_constant_like,
norm, bilinear_tensor_product, flatten2, im2sequence,
unpool, cos_sim, strided_slice, flatten, elementwise_min,
abs, acos

test=develop

* add fake init for the trainer, fix large memory hold in the trainer
* do not merge recv vars from a remote endpoint, test=develop
* add recv and save op, merge slice var in one op, save memory
* remove hsigmoid with pull sparse, test=develop

* fix input shape of op tests for elementwise_sub,
gather, pad2d, transpose, softmax, scale,
elementwise_max, hierarchical_sigmoid, reshape2,
sign, squeeze, reduce_sum, sum, squeeze2,
unsqueeze, unsqueeze2, cast, reverse

test=develop

* fix cast, elementwise_mul, gather, scale, sign,
softmax, transpose

test=develop

* change several tests to use the right parent class, test=develop

* add dtype for TestLoDTensorAndSelectedRowsOp, test=develop

* add no_need_buffer_slots interface to pybind

* fix ci bug for deleting data files, test=develop

* update, test=develop

* refine dygraph dataloader & polish related code, test=develop

* refine code based review comment, test=develop

* add seed op

* fix master patch when slot is dense
* test=develop

* add op function generator, test=develop

* add unittest, test=develop

* follow comments, test=develop

* fix windows compilation problem, test=develop

* support float64 for GradClipByGlobalNorm in dygraph, test=develop

* fix comment for GradClipByGlobalNorm, test=develop

* update benchmark for int8v2, QAT1, QAT2 accuracy and performance
test=document_fix

* change according to reviews
test=develop test=document_fix

* improve some descriptions and some models
test=develop test=document_fix

* update models benchmark data
test=develop test=document_fix

* update int8v2 and qat2 performance
test=develop test=document_fix

As the title

* fix python API tests that do not need to inherit OpTest, test=develop

* fix fp16 cases that will only be enabled in GPU mode, test=develop

* remove TestSoftmaxFP16Op from test cases of softmax_mkldnn_op, test=develop

* fix tests so that the cases are only created in GPU mode, test=develop

* add shell scripts. test=develop

* rename test_pybind_inference to test_pybind_interface and print repeat process in check_op_desc.py. test=develop

* add approval RD. test=develop

Add basic while_loop

* refine init function, test=develop

* add tests, test=develop

* remove extern, which may cause symbol error in gcc-4.8, test=develop

change input data type and decrease max_relative_error value in test_check_grad for grop_nom_op test test=develop (#21608)

Add tests to use dy/dx to make sure the gradient values calculated by the control flow backward is correct. Also fixed bugs detected by those tests.

Fix bugs:

1. Unlike sum_op, optimizer ops don't allow uninitialized input tensor. But in conditional_block_grad_op, since the conditional_block may not run, the output gradient tensor may be uninitialized, which will cause the optimizer op error. To fix it, we should let optimizer ops support uninitialized input like sum_op or assign the uninitialized gradient to 0 when the conditional_block_grad_op doesn't run. I found there are about 10+ optimizer ops. **To be simpler, I just assign output gradient of the conditional_block_grad_op to 0 in this PR**. But it can be further explored whether we can make optimizer ops like sum_op to support uninitialized input tensor because theoretically we can speed up without the assigning in conditional_block_grad_op.

2. Infer parameter shapes during append_backward. I didn't know that all our parameters are in global block. When op_desc is inferring shapes at the sub-block, it may not know the shape of gradients of parameters whose shape information is at global block. I fixed it by inferring shapes of gradients from forward var.

This PR also did some code clean up:
1. Print the var name when sgd_op catches shape error so that it is easier to debug
2. Fix a typo: dicta -> dict

* Add a single_process_multi_thread transpiler.
* Add some UTs.
* Fix some API description.