* multihead matmul add fp16

* fix windows error

* fix rocm error

* fix rocm error

* [Eager] call final_state_slice under eager mode

* rm useless comments

* use eager final state instead intermidiate state

* update fill_constant yaml

* update fill_constant yaml

* modify wrapped_infermeta_gen logic to fix special case

* fix slice in manipulation

* use fill_constant_

* modify slice infermeta

* rm final_state_conv2d

* use final_state_slice

* use final_state_slice only

* polish slice, use final state

* add paddle_throw for SplitInferMeta

* rm fill_constant_ temply

* recover array_equal, not allclose

* recover original code

* [Phi] Move Qr to the Phi

* [Phi] Regiter the cpu grad kernel for qr

* [Phi] Share the cuda kernels to lstsq

* [Phi] Remove some improper inlcude files

* [Phi] Modify codes based on the reviews

* [Phi] Remove unecessary files and add the cuda_only comment

* [Phi] Remove the unecessary include file

* [Phi] Remove qr_op.cu and lstsq_op.cu

* manuel_fused_gemm_epilogue

* polish and rename, pt* -> phi*

* fix code format

* [XPU] fp16 for layer_norm op. test=kunlun

* Skip inplace for coalesce_tensor_op outputs

* Fix typos

* Add UTs

* Fix typos

* add yaml and unittest

* update yaml

* update backward yaml and unittest

* update yaml

* add Yolov3LossGradInferMeta

* update yolov3_loss_op.cc

* fix bug

* code format

* support beam_search operator on xpu. test=kunlun

* support beam_search operator on xpu. test=kunlun

* support beam_search operator on xpu. test=kunlun

* support beam_search operator on xpu. test=kunlun

* support beam_search operator on xpu. test=kunlun

* Refactor build_op_downstream_map for standalone executor

* Add some comments

* remove cudaDeviceContext

* remove more template

* fix rocm compile

* remove alias name CUDADeviceContext

* fix compile

* fix tests

* revert changes

Co-authored-by: seemingwang <zsasuke@qq.com>
Co-authored-by: NDesmonDay <908660116@qq.com>
Co-authored-by: Nseemingwang <seemingwang@users.noreply.github.com>
Co-authored-by: NThunderbrook <a754913769@163.com>
Co-authored-by: Nxuewujiao <105861147+xuewujiao@users.noreply.github.com>
Co-authored-by: Nroot <root@yq01-sys-hic-k8s-v100-box-a225-0693.yq01.baidu.com>
Co-authored-by: NThunderbrook <52529258+Thunderbrook@users.noreply.github.com>
Co-authored-by: Nroot <root@yq01-inf-hic-k8s-a100-ab2-0009.yq01.baidu.com>
Co-authored-by: Nhuwei02 <53012141+huwei02@users.noreply.github.com>
Co-authored-by: Nyaoxuefeng <yaoxuefeng@baidu.com>
Co-authored-by: Nlxsbupt <luoxsbupt@163.com>
Co-authored-by: Nmiaoli06 <106585574+miaoli06@users.noreply.github.com>
Co-authored-by: Nroot <root@yq01-inf-hic-k8s-a100-ab2-0008.yq01.baidu.com>
Co-authored-by: Nchao9527 <33347532+chao9527@users.noreply.github.com>
Co-authored-by: Nqingshui <qshuihu@gmail.com>
Co-authored-by: Nyangjunchao <yangjunchao@baidu.com>

* Polish PEFuntion to speed up JitLayer

* Polish PEFunction code

* Fix comments

* update unstack_op

* update unstack_op

* update unstack_op

* fix unstack test

* update unstack

* update with remote

* fix unstack_test.py

* temp_save_change_nms_op

* add nms test

* update nms fix

* update unstack_op

* temp save change

* finish fix nms_op

* pass nms test

* fix CI

* fix ops test

* save change

* fix code style

* fix code style

* fix ci and codestyle

* fix ci
Co-authored-by: NShiningZhang <zhang_liang1991@126.com>

* infer context fix place error.

* update

* update

* - fix

* - another fix

* lint

* update code format

* add ymal and test

* update for comments

* add varlen_token_prune plugin, pass, convert

* Add kernel declarations

* Copy kernel implementation code

* Transfer implementation code

* Register new kernels

* Remove old kernels

* Fix code style

* Fix bugs

* mutable_data->HostAlloc

* Transfer infermeta

* Add yaml and update python api

* Add PADDLE_WITH_HIP check

* Update unittests

* Add kernel declarations

* Copy kernel implementation code

* Transfer kernel implementation code

* Register new kernels

* Remove old kernels

* Add lu_unpack_sig

* Fix bugs

* Fix bugs

* Fix bugs

* Optimize directory structure

* Add output checks

* Update include files

* lu_impl.h->lu_kernel_impl.h

* Transfer infermeta

* Add yaml and update python api

* Add check_eager
Co-authored-by: NBobholamovic <linmanhui@baidu.com>

* phi_prior_box

* add float[] support

* phi_prior_box_optest

* update

* remove cudaDeviceContext

* remove more template

* fix rocm compile

* migrate lstsq op

* update

* fix bugs for CIs

* update

* fix bugs

* add uts

* update

* update

* update

* fix bugs of jip

* fix bugs of hip

* update

* update according to review

* update

* update

* update

* update

* add some fp16 op for kunlun resnet50 model
*test=kunlun

* tmp
*test=kunlun

* add FLAGS_enable_api_kernel_fallback

* deal with more cases

* add ut for coverage

* - Unit tests to be debugged

- fix

- refactor

- diagnostic

- more diagnostic

- fix

- Fix number two

- fix

- fix

- fix

- alpha added

- more fixes

- compilation fix

- removed diagnostic code

- cosmetic fixes

* lint

* skip cast trt convert when input dtype is bool

* fixed bug for pass & engine

* fixed bug for benchmark GPT-3

* add tuner & profiler

* add algorithms & config

* init

* move CUDAStream to phi

* fix compilation

* merge develop

* add stream_owned_ member

* split cuda_stream.h

* fix cpu compile

* fix constructor

* fix bug

* fix windows compile

* fix inference test_levit

* fix windows tests

* Add kernel declarations

* Copy kernel implementation code

* Transfer implementation code

* Register new kernels

* Remove old kernels

* Fix code style

* Fix bugs

* mutable_data->HostAlloc

* Transfer infermeta

* Add yaml and update python api

* Add PADDLE_WITH_HIP check

* Update unittests

* Fix bugs

* Fix bugs

* Optimize directory structure

* Add output checks

* lu_impl.h->lu_kernel_impl.h
Co-authored-by: NBobholamovic <linmanhui@baidu.com>

* [Phi] Add yaml for assign_value

* [Phi] Fix the bug of the assign api and modify the unittest

* [Phi] Fix the bug when the tensor does not have the backend info

* [Phi] Replace the functional-style cast init by the brace-init

* [Phi] Cast the data explicitly

* fused_fc_elementwise_layernorm support fp16

* fused_fc_elementwise_layernorm support double