[cherry-pick 2.3] Add fused_multi_transformer op to optimize transformer generation performance (#42311)

* Add fused_multi_transformer op to optimize transformer generation performance (#41814)

* fix fused_multi_transformer compile failed in cuda arch < sm53 (#42315)

* fix ci timeout

add mixed-precission support for ipu

cherry-pick from #41733

* add fused_seqpool_cvm op;test=develop

* update docs dtype(core.VarDesc.VarType)

* fix code style, test=document_fix

fix code style, test=document_fix
Co-authored-by: NChen Long <1300851984@qq.com>

* add adaround post-quant method

* refactor eager flags

* fix flags error when we switch from eager to dygraph

* fix ci problem

* fix ci

* fix ci

* merge develop and fix code style

* merge develop and fix code style

* fix op test error

* fix op test error

* fix op test error

* fix op test error

* fix op test error

* merge develop

* approve amp for intermediate_dygraph

* add amp_utils for intermediate_dygraph

* add amp needcast check for mlu & npu

* test unittest

* add SetGradNode for set_stop_gradient && add checktensor for GradientHooks

* refine code

* refien unittest of imperative_amp for new dygraph

* inplace api skip amp

* add test_imperative_qat_amp for intermediate amp

* refine code

* refine test_amp ci strategy

* refine unittest code

* refine amp_utils code

* refine amp getpromotetype for some special op

* refine unittest code

* Modify save_quant_model.py to support differnet input and output filenames

* Correct wrong order of arguments

* add some op for full_quantization

* eager, test=develop

* fix bug, test=develop

* eager, test=develop

* merge legacy to fluid

* eager, test=develop

* eager, test=develop

* Refactor TensorAdd func by template and remove gradient_accumulation in eager

* Remove needless target name

* eager, test=develop

* eager, test=develop

* Use overload instead of template

* Remove legacy code

* Remove legacy code

* selectedrows, test=develop

* Remove DataType test

* eager, test=develop

* eager, test=develop

* support gan, test=develop

* Using Tensor directly instead of using EagerTensor

* support gradient_accumulation

* make test_imperative_lod_tensor_to_selected_rows longer

* make test_imperative_lod_tensor_to_selected_rows longer

* refine code

* ptb, test=develop

* Rename all EagerTensor to Tensor

* Rename some EagerTensor to Tensor

* rename EagerTensor to EagerVariable

* eager, test=develop

* eager, test=develop

* eager, test=develop

* eager, test=develop

* add more test

* eager, test=develop

* Support copiable selected rows and merge develop

* save load, eager, test=develop

* save load, eager, test=develop

* refine, test=develop

* remove useless _set_value method

* refine, test=develop

* refine, test=develop

* revert static_runner, test=develop

* EagerTensor to Tensor, test=develop

* refine, test=develop

* refine, test=develop

* clear grad, test=develop

* merge, develop

* merge, develop

* merge, test=develop

* merge, test=develop

* Support quant and part of slice

* support legacy static save

* extend slim tests time

* remove imperative on inference

* remove imperative on inference

* merge develop

* fix typo

* fix typo

* split slice related code into 2 part for imperative and eager

* split slice from inference

* split slice from inference

* fix test_tensor_register_hook
Co-authored-by: NWang Huan <wanghuan29@baidu.com>
Co-authored-by: NWeilong Wu <veyron_wu@163.com>
Co-authored-by: Nwanghuancoder <wanghuancoder@163.com>

* Add mobilenetv3_large performance test

* Disable the BF16 test if the device does not support BF16 computations

* Change test timeout

* remove

* pass

* more pass

* add DistributedFusedLamb op

* polish code

* fix compile error

* compatible with pten changement

* fix rocm compile error

* improve converage

* update upstream/develop

* fix cast_with_ptr.h

* add FLAGS_distributed_lamb_divide_nranks_when_allreduce=1

* fix clip before allreduce

* add use_master_param_norm

* code polish

* fix bug

* fix ROCM ci

* mish unit tests

* code format

* remove unused imports

* code format

* remove hard-coded shape values

* remove timeouts

* remove timeouts v2

* restore timeouts

* rebuild matmul pass: trt and gpu_cpu

* rebuild matmul pass: trt and gpu_cpu

* rebuild matmul pass: trt and gpu_cpu

* rebuild matmul pass: trt and gpu_cpu

* amp list updated

* tests updated

* gray list updated

* amp list updated

* test updated

* Upadate pass in quant2_int8_mkldnn_pass

* Back to the previous scale_matmul order

* Change place of cpu_quantize_placement_pass

* base changes for mul reimplementation

* empty commit

* tmp save

* full implementation of mul bf16/fp32 fwd bwd

* CI fix

* CI rerun

* changed unity build cmake to avoid gpu issues

* removed mul mkldnn from unity build

* added skipping tests if not cpu_bf16

* CI fix

* CI fix

* CI fix

* fix conv act int8 scale

* add unit test for conv+hard_swish

* make post training quant API support dataloader

* Quantize nearest_interp and nearest_interp_v2

* Check if avx_core supported

* Add depthwise_conv2d to supported quantization list

* Fix scatter_op fp16 perf problem.

* Add scatter into black list.

* Add scatter into black list for dygraph.

* support FP16 for more ops

* add amp list tests

* refine reduce_mean_grad

* fix OP benchmark ci

* fix fp16 reduce_mean

* updat ut, but still have some problems

* remove mean/reduce_mean fp16 kernel

* support multi precision update for LAMB

* hide some api

* fix ci uts

* fix lamb output of dygraph

* remove some changes to some PR

* try to fix Py3 CI compile error

* fix test_imperative_optimizer, add lars ut, add layer_norm ut

* fix ut, fix format

* fix ut

* fix windows ci

* add map_matmul passes to quant2_int8_mkldnn_pass

* fix fc+act fuse (activation scale)

* ci fix, c++17 structured bindings not available

* fix ci static check

* reshape+transpose+matmul_v2

* in_name->input_name

* fix pr-ci-static-check

* fix single card 8 unittests in new executor

* fix

* fix

* Support sub graph quant-post