* add FP16 for multinomial

* fix input data

* update code

* fix FP16

* fix code

* stride slice fp16 and bf16 unitest

* fix code style

* add self.dtype

[CodeStyle][C416][C417] rewrite unnecessary comprehension with function call and use generator instead of map (#52140)

* codestyle c416 c417

* fix error

* fix inc

* unify all C4 rules into one

* fix inc

---------
Co-authored-by: NSigureMo <sigure.qaq@gmail.com>

* pad3d add unittests of fp16 and bf16

* pad3d add unittests of fp16 and bf16

* fix cuda place

* fix random to uniform

* fix class name

* fix fp16 max relative error to 1.5e-3

* add dytpe register for onednn

* add pad uint16 check of common.py

* remove check_eager

* test_check_grad --> test_check_grad_normal

* regist output type for GraphSampleNeighbors and GroupNorm

* Update return type

* fix return type

* update

* fix detail

* fix bg

* add utest

* add utest

* add fuse adamw pass

* fix some bugs

* fix CIbug

* change chunk_size

* fix CI bug

* rm test_fused_adam_op.py

* fix CI bugs

* fix fuse_adamw_op_pass.cc

* change code style

* fix CI bug

* fix ut bug and use_adamw_op_pass.cc

* fix test_fuse_adamw_pass.py

* fix CI bug

* remove fluid

* fix ci bug

* fix CI bug

* fix generate_kernels.py in CUDA 12.0

* fix attrs bug

* add overflow check in memory efficient attention

* fix ci compile error

* fix ci compile error

* fix int8 support for full kernel

* fix ut.

* add unittest for conv2d/depthwise_conv2d/conv2d_transpose

* add bf16 for DWConv and ConvTranspose

* fix unitest of conv2d_transpose

* modify DWConv2d op and unittest

* fix unittest of conv2d_transpose_bf16

* modify unittest name according to review

* modify atol of DWConv2D unittest

* support assign op

* support assign infer_var_type

* change code according to review

* change code according to review

* only save 'get_infer_var_type_func'

* rest file mode

* unbind support bool dtype

* replace np.array_equal

instance_norm_grad kernel. Fix bugs that the data type of input is different from output in reduce_sum kernel. test=kunlun

* fix_gcc12_error

* fix gcc12 error

* fix gcc12 error

* extract Op and OPMaker to .h

* extend pattern for fused_op

* set "with_residual" default to false

* adjust fuse passes

* remove fc+eltwise flag

* fused_output_scale

* activation attrs

* remove extra attrs

* fix int8/bf16 unit tests

* simplify RecomputeOutputDims

* remove unused method

* Add description for attributes

* add extra check

* adjust op compats

* update quantize test

* fix protobuf parsing error

* fix int8 performance

* fused elementwises

* merge develop

* remove activation

* restore activation for existing add/sub ops

* fix mea compile error

* support 2-D bias

* add inline to avoid compile error

* polish codes

[Fix Bug] fix get_new_shape and get_new_data_from_tensor not support fallback to CPU on custom device (#52002)

* add all_reduce, reduce kernel and api

* fix all_reduce reduce ut

fix reduce op maker conflict

fix merge conflicts

* fix conflicts, rename ReduceOp->ReduceBaseOp in reduce_ops

rename allreduce op, to remove

* fix code format

fix comments

* modify test_collective_reduce_api ut timeout

* fix PR-CI-Build

fix comments: format phi operator

* decouple memory copy

* fix ci bugs

* fix ci compile bugs

* fix rocm compile

* fix ci bugs

* decouple memory

* deal with conflict

* fix xpu compile bugs

* fix xpu bugs

* deal with xpu bugs

* fix cmake bugs

* fix windows bugs

* fix ci bugs

* fix ci bugs

* delete redundance code

* add code for pybind

* fix py3 bugs

* fix ci bugs

* fix momentum dtype infer

* fix momentum datatype

* fix on cpu

* add momentum

* untracked files

* kthvalue perf

* remove unused files

* fix isnan

* fix isnan2

* fix bug

* try to fix rocm error

* first version, notest

* return final rst, notest

* use infinity() instead of max

* ut structure

* start up of ut

* generate lse

* update

* add depense

* reconstruct cmake

* move file

* add memory efficient attention and fix blasimpl

* update

* update cmake

* add namespace

* update cmake

* use .cu

* update for pad3d

* bug fix

* bug fix

* update

* bug fix

* update enforce

* add test case

* merge the lse pad

* fix kernel_fn of backward

* fix PADDLE_ENFORCE_EQ and phi_api

* fix PADDLE_ENFORCE

* fix PADDLE_ENFORCE

* rerun coverage

* fix memory efficient attention test

* rerun ci

* add cuda version condition

* add cuda version condition

* delete WIP test

* replace PADDLE_ENFORCE

* edit the namespace of datatype in multiple.cc

* rerun

* rerun

---------
Co-authored-by: Nliuyuang <liuyuang@baidu.com>

* remove fluid deps in fused_linear_param_grad_add_kernel

* fix compile error

* fix ut error

* follow comments