* 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

* 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

* A leap of try for cudaLaunchCooperativeKernel

* fix bugs

* Totally replace the lar cuda kernel

* Fix bugs

* fix code according to comments

* fix codes according to  review comments

* adding some function overload

* relocate the power operation.

* Create stateful OneDNNAXPYHandler object.

This makes it possible to call it multiple times without recreating the
oneDNN primitives every time.

* Prepare SGDOpKernel to reuse its implementation from OneDNN kernel.

* OneDNN SGD kernel.

* Update call to use new OneDNNAXPYHandler object api.

* Setup seed in proper place.

* Enable OneDNN kernel only for single case.

* For dense param and sparse grad.

* Small refactor.

* Enable oneDNN by op attr or by cmd line flag.

* Use int64_t type for number of elements.

* Support dense param and grad from OneDNN kernel.

* Enable SGD OneDNN kernel when use MP BF16 optimizer.

* Force non-copyable/movable OneDNNAXPYHandler.

* Reuse OneDNNAXPYHandler for spare tensors in SUM op.

* Fix SFINAE rules.

* Remove recording event inside AXPY.

* Get rid of internal primitive caching.

* Stop use PP cache mechanims to store mem and primitive obj.
* Handler obj store and reuse needed desc & prim

* Do not derive from MKLDNNHandlerT

* add layerwise learning rate for adamw

* fix format

* add unitest

* add NotImplementedError

* add gpu unitest

* update gpuplace

* sparse_momentum_op is used to save w@GRAD memory for gather_op when gather from a large parameter

* adamw support cuda

* adamw support cuda

* [npu]add rmsprop op

* add momentum_op_npu and test

* update

* fix hang

* copy found_inf to cpu in advance to improve performance

* add npu test

* add npu test

* refine code

* refine memcpy op

* fix adam

* adam add input SkipUpdate

* add unittest

* add npu unittest

* fix xpu compile

* remove param stream

* Add oneDNN AXPY handler.

* Add fallback for small tensors.

* Fix ifdefs

* Remove unnecessary namespace prefixes and add missing headers.

* Guard handler_axpy with proper ifdefs.

* Compilation of this function is possible only when Paddle is not build
with CUDA nor HIP.

* Move AXPY handler code to separate files.

* Use oneDNN AXPY handler in SGD op.

* Use axpy handler only when Paddle is built with oneDNN.

* Add test for SUM BF16 with big rows.

* Fix SFINAE rules for elementwise_add_to.

* Add test case for SGD with big rows.

* update

* update
Co-authored-by: NAdam Osewski <adam.osewski@intel.com>

* refine ~npuOpRunner

* implement destructor and forbid copy

* use reference to avoid copy

* use const reference

* relax adam precision

* fix top_k

* add use_global_beta_pow

* add use_global_beta_pow

* update npu kernel

* update python api

* refine code

* add ut for use_global_beta_pow

* fix npu kernel

* add ut for api

* add ut for exception

* add ut for save/load

* add input EpsilonTensor for adam

* update python api

* add unit test

* add npu test

* add more ut

* [NPU] support GarbageCollector for npu (#31874)

* support GarbageCollector for npu

* fix typo

* fix gather_grad

* disable NPUDefaultStreamGarbageCollector on NPU

* [NPU] support npu for memcpy op (#31808)

* support npu for memcpy op

* add ut

* fix ut

* fix typo

* 【NPU】fix bug of using temp vector (#31963)

* fix bug when beta1_pow on cpu (#31995)

* [NPU] support npu profiler (#31684)

* support npu profiler

* add python api

* fix bugs

* add wrapper for incomplete type

* update profile proto

* record npu wait

* add xpu placeholder

* fix adam (#32016)

* [NPU] enable async copy and  add wait before sync operation (#31956)

* enable async copy and  add wait before sync operation

* remove unneccessary wait

* add FillNpuTensorWithConstant

* refine

* fix fill_constant

* make TensorFromVector/TensorToVector sync

* [NPU] Support dataloader on npu place. (#31867)

* [NPU] Wait on NPUPlace (#32086)

* [NPU] fix cast op (#32121)

* fix npu kernel of cast op to handle casting to same dtype

* add comments

* [NPU] support cann 20.3 (#32044)

* fix compile problem on cann 20.3

* fix ut

* fix test_mul

* fix check_finite_and_scale

* fix lookup_table_v2_grad

* fix cmake

* support print op

* [NPU] Support npu save load (#31893)

* support save load for NPU

* add save load npu unittest

* support np.array transform in NPU

* fix errors

* delete dygraph in unittest

* add Wait

* fix unittest

* fix review comment

* fix unittest problem

* fix little problem

* change aclrtSynchronizeDevice to aclrtSynchronizeStream for better performance (#32196)

* change aclrtSynchronizeDevice to aclrtSynchronizeStream for better performace

* refine code

* fix NPUDeviceContext in all c++ unittest (#32198)

* fix NPUDeviceContext in all c++ unittest

* refine log
Co-authored-by: Npangyoki <pangyoki@126.com>

* [NPU] Remove TensorFromVector and avoid sync copy in npu op kernel for better performance (#31994)

* enable async copy and  add wait before sync operation

* remove unneccessary wait

* add FillNpuTensorWithConstant

* refine

* fix fill_constant

* change TensorFromVector to FillNpuTensorWithConstant

* fix ignored api

* delete extra unittest

* fix little error

* fix update_loss_scaling_op_npu and check_finite_and_unscale_op_npu

* change TensorCopySync to TensorCopy

* delete useless Wait and add StreamWait

* fix npu_stream error

* fix check_finite_and_unscale_op_npu TensorCopy

* only save stream wait

* fix NPUDeviceContext in all c++ unittest

* delete wait
Co-authored-by: Nzhiqiu <chenqiuliang@baidu.com>

* delete useless unittest file (#32206)

* Fix op test (#32231)

* fix conditional block (#32243)

* fix adam bug again (#32246)

* fix compile

* fix ut

* fix ut
Co-authored-by: Nliym27 <33742067+liym27@users.noreply.github.com>
Co-authored-by: Npangyoki <pangyoki@126.com>

* merge 31065

* Fix typo of selected_npus (#31230)

* merge 31249

* [NPU] Support npu op pow and pow grad (#31247)

* [NPU] Support npu op: (1) pow (2) pow_grad

* Support fp16

* Fix pow npu fp16 test (#31256)

* support list of list attribute for NPU (#31299)

* support list of list attribute for NPU

* fix compile problem

* fix reference

* [NPU] Support npu op: (1) slice (2) slice_grad (#31275)

* fix reading flags from env (#31329)

* merge 31347

* [NPU] Support npu op layer_norm and layer_norm_grad (#31310)

* init commit, add layer_norm npu kernel

* fix typo

* add unittest

* add unittest

* fix bug

* fix bug

* refine ut

* [NPU] add npu kernel for equal op (#31393)

* add npu kernel for equal op

* refine code

* add more ut

* update year

* [NPU] Support npu kernel for shape op  (#31427)

* add shape npu

* fix

* fix

* fix endif (#31431)

* Fix pow, use fillD instead of broadcast (#31433)

* Fix pow, refine code (#31440)

* fix cmake of cryptopp to avoid downloading every time (#31451)

* [NPU] squeeze and unsqueeze op for ascend (#31452)
Co-authored-by: Nroot <xiayanming@baidu.com>

* Support npu kernel for gather op (#31458)

* add gather npu op

* code review done

* update python new line

* precommit

* fix review

* del commit

* 【NPU】add scale op for npu (#31499)

* add scale npu

* fix

* fix

* Support TensorFormVector, TensorToVector of bool type (#31518)

* support TensorFormVector, TensorToVector of bool type

* add ut

* fix compile problem

* 【NPU】support npu kernel for fill_constant op (#31521)

* add fill_constant npu

* add fill_constant npu

* fix

* cherry-pick 31422, solve conflict

* 【NPU】Support npu kernel for matmul op (#31544)

* add matmulv2_npu

* add matmul

* add matmul

* [NPU] Support npu op elementwise_mul and elementwise_mul_grad (#31571)

* [NPU] Support npu op elementwise_max (#31574)

* 【NPU】add relu op for  npu (#31515)

* add relu npu

* fixed

* fix

* 【NPU】Suppert npu kernel for reshape2 op (#31524)

* add reshape2 npu

* add reshpe2

* [NPU] Support npu kernel for gather op fix bug (#31541)

* add gather npu op

* code review done

* update python new line

* precommit

* fix review

* del commit

* update gather_grad

* fix bug

* fix bug

* [NPU] Support npu kernel for amp_check_finite_and_unscale_npu op (#31457)

* Support npu kernel for amp_check_finite_and_unscale_npu op

* support EnforceNotMet exception

* fix exception bug

* modify python unittest

* precommit

* update c++ unittest

* fix review

* fix review

* [NPU] accuracy op (#31492)

* accuracy op

* fix license

* fix

* add test and fix bug

* [NPU] add Assign OP (#31561)

* add assign op

* add test assign npu test

* dele if def
Co-authored-by: Noyjxer <1728722986@qq.com>

* [NPU] fix npu op elementwise_mul_grad (#31592)

* 【NPU】Support npu op gelu and gelu_grad (#31530)

* Support npu op gelu and gelu_grad

* Support npu op gelu and gelu_grad

* [NPU] fix assgin cmake (#31595)

* fix gather_grad bug (#31607)

* [NPU] add range op (#31560)

* add range op

* fix codestyle; call GetSize directly
Co-authored-by: Noyjxer <1728722986@qq.com>

* 【NPU】Support npu op elementwise_div and elementwise_div_grad (#31573)

* Support npu op elementwise_div and elementwise_div_grad

* Support npu op elementwise_div and elementwise_div_grad

* Support npu op elementwise_div and elementwise_div_grad

* [NPU] Support npu op log, log_grad, sqrt, sqrt_grad, square, tanh and tanh_grad (#31600)

* [NPU] Support npu op logicalnot_op (#31534)

* [NPU] Support npu op elementwise_min (#31575)

* [NPU] Support npu op elementwise_pow (#31576)

* [NPU] Support npu op table_lookup_v2 and table_lookup_v2_grad (#31399)

* [npu] support npu kernel `table_lookup_v2`

* clean up

* +python test

* +cmake

* clean up

* remove int8 kernel
+ python unitest for fp16

* clean up

* [NPU] support npu kernel for `less_than` (#31327)

* [npu] support npu kernel for `less than`

* remove int* kernel

* cleanup

* [NPU] Support npu kernel scatter op (#31624)

* Support npu kernel scatter op

* Add more test

* [NPU] fix allocator min chunk size (#31632)

* [NPU] Support NPU kernel cast op (#31635)
Co-authored-by: Nfrankwhzhang <frankwhzhang@126.com>

* [NPU] add npu kernel for sgd (#31639)

* 【NPU】Support NPU kernel for reduce_sum op v2 (#31620)

* add reduce_sum

* fix broadcastd

* fix test

* fix

* add unsqueeze in reduce_sum

* add template

* add unittest for keep_dim

* test reduce_all
Co-authored-by: Nfrankwhzhang <frankwhzhang@126.com>

* [NPU] add npu kernel for adam (#31644)

* add npu kernel for adam

* refine code

* disable test

* modify atol

* 【NPU】Support npu kernel for mul op (#31584)

* add mul

* add test mul

* [NPU] add npu kernel for softmax_with_cross_entropy (#31656)

* init

* fix bugs

* [NPU] add npu kernel for mean Op (#31562)

* update mean op

* update mean op

* give a better test activation
Co-authored-by: Noyjxer <1728722986@qq.com>

* Revert "[NPU] add npu kernel for mean Op (#31562)" (#31665)

This reverts commit 468ac699.

* 【NPU】Add TensorCopy to NPU kernel for reduce_sum op  (#31667)

* update unittest

* add TensorCopy in npu grad kernel

* [NPU] Support npu op `expand` (#31405)

* [npu] support npu kernel  for `expand`

* [NPU] fix shape of dx in mul_grad (#31675)

* fix shape of dx

* refine code

* [NPU] add Increment op (#31563)

* add increment

* fix

* update test increment op inplace

* update increment op

* increment b = 2
Co-authored-by: Noyjxer <1728722986@qq.com>

* [NPU] add NPU add topk  (#31596)

* add topk op

* add cmake

* update topk npu op

* refactor func

* fix test not go npu TopKD bug

* NPUPlace(4) to NPUPlace(0)

* update comment
Co-authored-by: Noyjxer <1728722986@qq.com>

* [NPU] Support NPU kernel sum op (#31671)

* [NPU] npu support `transpose` (#31486)

* cherry-pick 31564, solve conflict

* [NPU] Fix bug: Fix calculation errors of pow grad npu kernel (#31699)

* [NPU] Support testing grad of NPU ops in OpTest (#31697)

* [NPU] Support NPU kernel of stack op (#31711)

* [NPU] Remove redundant ctest of top_k_op_npu_test (#31718)

* [NPU] fix reshape npu op kernel (#31726)

* rename npu op file

* fix reshape

* [NPU] change transpose to transpose2 (#31734)

* change transpose to transpose2

* fix bug

* [NPU] Support  mean npu kernel (#31729)

* [NPU] fix some bugs of npu op (#31739)

* fix softmax

* fix mean

* fix lookup_table_v2

* 【NPU】Fix npu kernel elementwise_div_grad  (#31753)

* [NPU] fix the grad kernel diff bug of gather op (#31757)

* fix gather grad kernel diff

* fix gather grad kernel diff

* fix gather review bug

* 【NPU】Fix reshape test & add grad test (#31776)

* fix

* fix

* [NPU] support fp16 for npu accuracy op (#31797)

* [NPU] support list of tensor input (#31801)

* support list of tensor as npu input

* add comment

* fix typo

* fix typo

* [NPU] add npu kernel for concat op (#31695)

* add npu kernel for concat op

* add npu kernel for concat op

* refine code

* update

* refine concat_grad

* [NPU] Support npu kernel for op elementwise_floordiv (#31822)

* [NPU] fix bug of lookup_table_v2_grad (#31834)

* [NPU] support default stream (#31510)

* [NPU] support mixed precision input for npu layer norm (#31847)

* support mixed precision input for npu layer norm

* fix layer_norm npu kernel
Co-authored-by: Nzhiqiu <chenqiuliang@baidu.com>

* 【NPU】Support npu kernel for update_loss_scaling op (#31830)

* add update_loss_scaling_npu NPU kernel

* change TensorFromVec to Memset

* fix compile problem (#31850)

* [NPU] support npu for conditional_block op (#31854)

* 【NPU】Add int dtype kernel for reshape2 op (#31864)

* fix

* fix

* [NPU] fix some op bugs (#31855)

* fix some op bugs

* fix some bugs

* follow comments

* fix log level

* add ut

* [NPU] support fp16 of input for api pow (#31871)

* [NPU] add npu kernel for truncated_gaussian_random op (#31654)

* init

* add todo

* add npu kernel for truncated_gaussian_random

* add sync

* fix concat_grad

* fix typo

* fix compile

* fix compile

* fix compile

* fix compile

* fix compile

* fix compile

* fix code style

* fix code style

* fix code

* Fix op test (#32231)

* fix conditional block (#32243)

* fix style code
Co-authored-by: Nxiayanming <41795079@qq.com>
Co-authored-by: NLeo Chen <chenqiuliang@baidu.com>
Co-authored-by: Nliym27 <33742067+liym27@users.noreply.github.com>
Co-authored-by: NReventon_L <luyuxiang1994@qq.com>
Co-authored-by: Nroot <xiayanming@baidu.com>
Co-authored-by: Noyjxer <1728722986@qq.com>
Co-authored-by: Nyinhaofeng <66763551+yinhaofeng@users.noreply.github.com>
Co-authored-by: NOleNet <olenet@126.com>
Co-authored-by: NMeiyim <chen_xuyi@outlook.com>
Co-authored-by: Noyxuan-11 <963650125@qq.com>
Co-authored-by: Npangyoki <pangyoki@126.com>

* Initial draft for SGD BG16 kernel.

* Unit tests for SGD with BF16 data type.

* Add VLOG message to SGD BF16 op CPU kernel.

* Enhance error messages and error types.

* Refactor SGD op kernels to leverage some common code.

* Make easier to add new kerne invoke code.

* Fix SGD op kernel for sparse grad.

* Unify quotes style.

* Fix error for ROCM compilation.

* Use specialized PADDLE_ENFORCE_xx functions.

* lamb_op_xpu;test=kunlun

* modify lamb_op_xpu.cc;test=kunlun

* delete atol lamb_op_xpu; test=kunlun

* update xpu.cmake;test=kunlun

* test_error 1e-5,lamb_op_xpu;test=kunlun

* error1e-5,lamb_op_xpu,test=kunlun

* delete atol lamb_xpu;test=kunlun

* modify atol,lamb_op_xpy;test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu, XPUOptest;test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu,modify xpu_cmake; test=kunlun

* lamb_op_xpu;test=kunlun

* lamb_op_xpu,modify xpucmake;test=kunlun

* add rmsprop_op_xpu test=kunlun

* modified rmsprop_op_xpu error code. test=kunlun

* Modify the calculation logic of LambOptimizer

* 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.

* fix momentum op register

* Test compilation time with less parallel count, notest, test=windows_ci

* optimize rules of Unity Build, notest, test=windows_ci, test=windows_op

* limit parallel counts used only on GPU, test=develop

* remove limit of argument /m:8 on Windows, test=develop

* Compiling operator libraries with Unity Build on Windows CPU.

* Compiling operator libraries with Unity Build on Windows GPU, no_test, test=windows_ci

* Add option in windows ci script, no_test, test=windows_ci

* Optimize parallel compiling, test=develop

* remove limit of parallel compile and skip some ops in UB, test=develop

* remove changes of header file, test=develop

* remove changes of header file, test=develop

* fix test_eye_op unittest failed, test=develop

* Compiling operator libraries with Unity Build on Linux, test=develop

* set default WITH_UNITY_BUILD=OFF, test=develop

* Move unity build rules into a single file and add comment, test=develop

* optimize parallel compilation, test=develop

* fix undefined reference error on coverage ci, test=develop

* add the weight decay func for the momentum op

* Add the multi_precision function in Momentum Optimizer.

* Make sure that the initial value of master weights are same with the fp16 weights.

* add static loss scaling.

* add the rescale_grad function in the pure fp16 training.

* use the original momentum updating method.

* Polish some codes, such as variable names.

* add docstring for apis.

* update the var creation details of _create_master_weight.

* not modify codes about imperative momentum updating.

* Fix the error of test_dist_sparse_tensor_load_momentum UT.

* add unit test for multi precision fp16 training.

* add more unit tests for CI.

* Use lower threshold values for allclose comparing in test_multi_precision_fp16_train UT.

* For CI Coverage Checking.

* refactor momentum op to combine weight_decay (scale op and sum op)

* lookup_table_xpu op report errors;test=kunlun

* add adam xpu op;test=kunlun

* reset lookup

* change adam wrong;test=kunlun