* add use_hierarchical_allreduce

* support hierarchical allreduce for more cases

* fix test and doc

* add sparse SyncBatchNorm

* move average_accumulates op to phi kernel

* add xpu resnet_unit
*test=kunlun

* tmp
*test=kunlun

* add sparse api coalesce

* move _gradients to primapi and rename to grad

* modify jvp to call forward_grad in primitive mode

* add primapi unittest and remove some unused test cases.

* fix  circular import problem

* move paddle/autograd/functional into paddle/incubate.autograd/functional

* remove unused JacobianBatchLast class

* feat(): dynamic_to_static support for ipu.

* fix(): format fix.

* fix format

* fix cpplint error

* use phi::errors

* fix format

* fix format

* fix(): add api to restore patched function.

* fix(): identity_loss uses cpu place as expected kernel type.

* doc(): add IPU dy2static related docs.

* fix(): combine test cases.

* fix format

* fix comment

* fix format

* apply comment

* fix compiling

* fix(): align docs.

* fix(): fix identity_loss function docs.

* fix(): adjust mean and sum in identity_loss.

* fix(): minor docs.

* move API to paddle.incubate.identity_loss

* fix UT
Co-authored-by: Nzhaorui chen <zhaoruic@graphcore.ai>

* [Sparse] add SparseCsrTensor fused_attention kernel and API

* fix comment

* Add new attr of fused_multi_transformer

* fix format

* add note

* add in layer

* fixfixfixfix

* [UPDATE FLUID API] only reference in paddlerec

* change lr

* [UPDATE FLUID API] only reference in paddlerec

* update by reviews

* [Sparse]add SparseTensor mv kernel(csr*dense_vec->dence_vec, coo*dense_vec->dense_vec)

* fix CI

* enable Jacobian,Hessian supporting new autograd

* fix prim mode failed in PR-CI-Windows

* add forward_gradients api

* add forward_gradients api

* skip test_autograd_functional_prim in windows ci

* fix test_autograd_funciton_prim timeouot

* remove the block parameter in prim2orig method

* remove duplicate to_tensors code snippet # test=allcases

* rename eagerpylayer

* add new API/OP(csr->csr) of SparseTensor softmax

* fix comment

* fix to_sparse_csr

* Support optional residual add in fused_attention and fused_feedforward.

* Add checkpoint and add the check of add_residual when pre_layer_norm is false.

* Add TODO and change the python api to add add_residual argument.

* add some kernel(csr*dense->csr, dense*dense->csr) of SparseTensor matmul

* fix CI

* fix CI

* fix comment

* fix comment

* add nproc_per_node for DistributedFusedLamb

* fix nproc_per_node communicator bug

* fix ring_id = 1 init bug

* fix ci

* fix test_parallel_executor_mnist.py

* add use_master_acc_grad

* add ut

* use yapf to format all python file

* yapf exclude two unittests file for they rely on writing and reading file, and format will break them

* disable diff_py_file because too many diff files cause command following failed