[XPU AMP] 1. xpu support gradient acc 2. xpu support create tensor in dygraph 3. xpu support update weight params in amp (#36439)

* autograd.functional passed pylint checker.

* autograd.functional: fix import errors.

* autograd.functional: fixed unit tests.

* autograd.functional minor format change

* [autograd.functional] Fixed vjp and jvp's v=None bug.

This reverts commit 0452f27c.

* fix the initializer of resnet unit op

* fix the initializer of resnet unit op

* remove wrong restrict

* remove master_param_out __restrict__

* update

* add resnext model
* add zh docs
* add unittest
* test performance
Co-authored-by: Ainavo <ainavo@163.com>
Co-authored-by: Npithygit <pyg20200403@163.com>
Co-authored-by: Ainavo <ainavo@163.com>
Co-authored-by: Npithygit <pyg20200403@163.com>

* fix no_grad context error in train mode when using save/load

* change net to train mode in test case

* Add CinnSubgraphSearchPass

* solve CI problem of subgraph order not same

* fix some bug by review advices

* ensure the independently of subgraph, that mean the subgraph should not have link to out-graph

* rename cinn_subgraph_search_pass to build_cinn_pass and delete paddle_to_cinn_pass

* add flag to control wheter append build cinn pass

* remove AppendPass at ParallelExecutorPassBuilder

* rename paddle_to_cinn_pass to build_cinn_pass in build_strategy and close test_run_from_cinn

* native commit for triple grad of sigmod

* Updated unittests files

* init functional jacobian api

* Updated trible_test func

* Updated gradient_checker & test_script

* finish test with dtype float32

* add float64 test case

* polish code

* use atol=1e-5 with dtype float64

* fix for ci

* set timeout for test_jacobian

* fix dygraph grad to support high differential

* polish API docstring

* Updated gradient checker and some related files

* fix double grad strip error for high differential

* fix double grad strip error for high differential

* Add Sigmoid triple grad tests

* fix dygraph double grad dtype error when calling for high differential senario

* Updated triple grad teses func

* Use np.random to initialize ddx

* Updated triple_grad_check func

* add todo for gradient checker and refine some comments

* remove additional code

* add test for warnging in backward.py

* add tanh triple grad

* format python code

* refine code
Co-authored-by: Nveyron95 <veyron_wu@163.com>
Co-authored-by: Nlevi131 <limaolin01@baidu.com>

* add sparse_embedding doc

* delete wrong space

* fix error for sample code

* fix error for doc compile

* delete __all__

* modify sample code

* support bert when exists matmul_v2

* update

* [NPU] Add density_prior_box op

* [NPU] Add density_prior_box op

This reverts commit bf748f24.

* merge momentum ops

* update

* add ut to improve coverage

* remove optimizer change

* fix error msg

* update ut

* add __restrict__ for CUDA

* update ut

* move merged_momentum_op to optimizer dir

* fix coverage

* add no_sync for parameters sync

* add pipeline for moe

* add memory_analysis

* fix has_none

* add pool2d convert test

* modify error

* modify error

* modify error

* modify error

* modify error

* modify error

* sparsity support

* fix BatchNorm for fp16

* update

* update

* update

* try make CI pass

* doc typo

* update doc string

* A leap of try for cudaLaunchCooperativeKernel

* fix bugs

* Totally replace the lar cuda kernel

* Fix bugs

* a test for lars merge

* Adding las_op_momentum infer_shape

* Fix codes

* use avg_numel instead of max_numel to acquire grid num

* modify unittest files about lars op

* Finally converge when merged-lars works

* fix ctest files

* add merged_operation kernel when cuda version is older than 11

* Fix code style

* fix ctest failure

* fix error

* fix all ctest error and change lars compute code of cpu

* fix bugs on v100.

* revert python modififation about lars

* revert python modification codes

Check detail PR description at https://github.com/PaddlePaddle/Paddle/pull/36116