* elementwise_add_grad Op optimization  (#29575)

* optimize for long width for elementwise (#29602)

* refine (#29622)

* delete the code for fp16 optimization because it is not faster than common template code (#29715)

* fix the shape choose of vectorize for cuda

* optimization for fp16 elementwise add (#29744)

* Fix the compiler error for half type (#29799)

* refine the compiler error for half2 operation (#29816)

* fix the compiler error when gcc4 cuda9.0 (#29997)

* Support pure fp16 training for AMP API. (#29544)

* add cast ops before and after unsupported fp16 ops.

* Keep partial net in FP32 pattern.

* Support check_finite_and_unscale and update_loss_scaling for FP16 calculation mode.

* Add fp16 support for adam op.

* add multi precision attr for adam.

* Fix the bug of test_multi_precision_fp16_train UT.

* Code format for CI.

* Fix the redefine error about MPTypeTrait on windows.

* fix bugs of the _create_accumulators func in Momentum.

* fix bug when inserting post cast op.

* Add the update_loss_scaling op in allow_set of UnusedVarCheck.

* Update for ci coverage.

* Add some doc for OptimizerWithMixedPrecision.

* Fix the code style.

* Imporve the doc of `amp_init`.

* Change for fp16 testing if users have the infer program defined in separate way.

* Remove tensor copy in the update_loss_scaling op. (#29426)

* remove tensor copy in the update_loss_scaling op

* not use thrust.

* fix some cuda memory access error.

* fix tensor shape bug

* fix op_num

* clean code

* Quantization supports 2.0 APIs

* Fix the error of save_quantized_model

* Optimization grad merge performance (#29784)

* [fleet] combine amp and gradient merge, test=develop (#30086)

* fix assign_op_xpu concat_op_xpu warining (#30120)
Co-authored-by: Nliuyuhui <liuyuhui@baidu.com>

att, cherry-pick of #30219

* fix dynamic to static error

* delete paddle.nn.functional.assign

[Cherry-Pick 2.0] In creation.assgin, reuse implamention code of layers.tensor.assign to avoid maintain two code (#30227) (#30236)

cherry-pick #30227

[cherry-pick] [Dy2Stat] Don't convert to paddle.shape if var_x.shape is not negetive #29965 (#30235)

* [Cherry-Pick 2.0] [Dy2Stat] Don't convert to paddle.shape if var_x.shape is not negetive (#29965)

1. When x is Variable, call nn.shape(x) only in following cases:
 1）The shape of x is used in control flow condition.
 2）The dim to be used is negetive
2. When x is Variable, but x.shape or x.shape[idx] doesn't contain negetive value, don't convert to paddle.shape()

* [Cherry-Pick 2.0] [Dy2Stat] Use Paddle2.0 api paddle.tensor.array_* (#30156)

* Optimizer trans momentum (#29597)

* merge amp related function in Momentum from paddle.fluid.contrib.optimizer into paddle.optimizer.

* Add unittest for 2.0  Momentum API.

* fix some bugs in weight_decay.

* add alias for fluid.contrib.mixed_precision (#29562)

* add alias for fluid.contrib.mixed_precision

* add static.amp into setup.pu.in (#29621)

* add static.amp into setup.pu.in

* add unittest for api

* fix a bug in multi_precision_fp16 unittest. (#29756)

[cherry-pick 2.0] Fix bug: In dynamic mode, if start or end is negetive, __getitem__  return wrong result(#30003) (#30146)

1. when slice_item is a slice:
 1) the start of __getitem__ should be std::max(start, 0) if slice
 2) the start of __getitem__ should be std::min(end, dim)
2. when slice_item is an integer, it should be in [-dim_len, dim_len)
3. Fix error message to use accurate data

1. Type of index: int, slice(step must be 1).

2. Type of value:
 (1) int32, int64, float32, bool;
 (2) numpy.array(int32, int64, float32, bool);<Note: float64 is not supported>
 (3) paddle.Tensor(int32, int64, float32, float64, bool);

* fix beam search bug

* add dygraph unittest

* update dynamic_decode argument doc

* add warning info for state which has no lengths attribute

* simplify prepared op impl to improve performance

* fix kunlun compile error

* continue fix kunlun compile error

* only transform diff place when dtype diff

* fix failed unittests

* remove useless file

* polish impl by review comment

* Add Lookahead and ModelAverage Optimizer (#30004)

* test=develop, add model_average and lookahead

* Improve Index select cuda kernel (#30139)

* test=develop, add index_select_cuda kernel

* fix syncbn convet

* add unittest

* Simplify the options of spawn based on fleetrun (#30144)

* Simplify the options of spawn based on fleetrun

* polish details

* polish doc details

* cleanup enum test=develop (#29294)
Co-authored-by: Ngongweibao <weibao.gong@gmail.com>

* Support storage of large parameters (#29988)

* Support storage of large parameters

* Reduce the complexity of the unittest

* Reduce the complexity of the unittest,commented out unittest for

* add unittest for static.save/load

* Increase the timeout threshold of 'test_static_save_load'

* Increase the timeout threshold of 'test_static_save_load'

* Increase the timeout threshold of 'test_static_save_load' and 'test_paddle_save_load'

* Increase the timeout threshold of 'test_static_save_load' and 'test_paddle_save_load'

* Extend the timeout for the (#30151)

* Improve performance of elementwise_add grad op (#29187)

* pass stop_gradient for cast op

* improve performance of elementwise_add grad

* use tensor copy async

* dygraph branch

* fix dygraph branch

* add ut

* make gelu fp16 computing more robust (#29484)

* Add fast path for dropout when p == 0  (#29553)

* add fast path for p == 0 in dropout

* add ut

* Layer norm fp16 (#29169)

* add fp16 for layer_norm op

* revert layernorm api

* fix forward

* fix forward

* fix backward for layernorm with fp16

* fix unit test for layernorm with fp16

* fix with_mkldnn compile error for layernorm with fp16

* 1. revert to PADDLE_ENFORCE_NOT_NULL, 2. change static_cast<float> to static_cast<U>

* fix with_mkldnn compile error for layernorm with fp16

* fix with_mkldnn compile error for layernorm with fp16
Co-authored-by: Nzhiqiu <chenqiuliang@baidu.com>

* fix layer_norm accuracy (#29434)

* Layernorm opt (#29522)

* layernorm fw opt

* layernorm bw opt

* fix typo, test=develop

* remove const dim3 for windows CI compatibility

* merge develop
Co-authored-by: Nzlsh80826 <zlsh80826@gmail.com>

* Fix compile problem when cuda_arch < 6000 (#29576)

* fix compile problem when cuda_arch < 6000

* refine code

* refine code
Co-authored-by: Nzhiqiu <chenqiuliang@baidu.com>
Co-authored-by: Nzlsh80826 <zlsh80826@gmail.com>

Change-Id: Ia5279b0cbb6a5b3970aff66e9510e0d85efa70ce

* fix bn docs (#30096)

* add attribute for batch_norm (#29950)

* add attribute for batch_norm

* fix log test=release/2.0

* fix ut test=develop

* add fp16 check into max and avg pool (#29479)

* Add ReserveSpace in dygraph batch_norm. (#29221)

* Add ReserveSpace in dygraph batch_norm.

* Add unittest for reservespace

* add float16 into adaptive_avg_pool2d check list. (#29547)

4 APIs: array_length, array_read, array_write, create_array，cherry-pick #29565

* add topo aware

* resource.h

* topo aware

* format

* [cherry-pick 2.0] Fix unitest test_slice (#29740)

Before this commit, test_slice use old api `dygraph_to_static_func` to use Dynamic-t-Static and use Executor explicitly，which is not recommended to users.
After fixed, use recommended API `paddle.jit.to_static` to replace `dygraph_to_static_func`, which won't trigger the random exception on coverage CI.

* [cherry-pick 2.0][Dy2Stat] Support grammar: for ele in var[idx] (#29541)

Support to transformfor ele in var stms in which var is a slice of Tensor.

* [cherry-pick 2.0][Dy2Stat] Fix bug for loop: a variable is used and created in loop, but used before created (#29769)

* fix ininite scale values (#29386)

* Support dygraph quant model (#29927)

* Avoid the scale to be infinity in quant2_int8_mkldnn_pass, test=develop
* support quantized model for paddle2.0 dygraph, test=develop
Co-authored-by: NWojciech Uss <wojciech.uss@intel.com>

Set FLAGS_selected_gpus for spawn.

When the child process starts, it will inherit the configuration of the main process and set the FLAGS once, but the environment variable has not been set at this time, which leads to the FLAGS_selected_gpus is keep same with mainprocess(usually empty), so manually update the flags here.

注：增加了一个单测，又移除了，单测打印显示CI机器nvidia-smi只有两张卡，需要大于两张卡才能测这个问题

* support deepcopy for Layer/Tensor/Paramerbase

* fix some code

* add distributed.split, test=develop

* update, test=develop

* update, test=develop (#29559)

* Disable gloo by default (#29805)

* update, test=develop

* update, test=develop

test=develop

* add alias for upsample, test=develop

* add alias for upsample

* fix example