* Work for fa-2 padded fwd. Code to be cleaned.

* Work for fa2 unpadded fwd.

* Work for padded-bwd, dk get small diff on np.random.seed(0)

* Anyway I pass paddle's utest, except return softmax without dropout.

* Clean code.

* Modify interface.

* Clean code and add some check.

* Easy compile for dev.

* Fix ci.

* Fix ci-build.

* Add std c++17 option again.

* Limit max job when compiling fa2.

* Remove const_cast

* Add fwd params, to be cleaned.

* Clean code.

* Add bwd params.

* Clean code.

* Add enforce.

* Use v2.0.4

* Pass RNG state to fa2 capi

* Fix review.

* Add assert

* Skip compile for sm less than 80.

* [clang-tidy] modernize-avoid-c-arrays

* rollback

* [clang-tidy] fix

* close modernize-avoid-c-arrays

* fix PHI_DEFINE_string; add PHI_DEFINE_bool NOLINT

* fix PHI_DEFINE_string

* fix next_h_state and parity err

* fix win32

* fix cuda_graph

* fix accuracy_kernel

* fix math_function

* fix fused_softmax_mask_kernel.cu load_data and warp_reduce; rollback concat_and_split_functor ins_addr

* fix fused_dropout_add_grad_kernel

* fix

* rollback cu

* rollback concat_and_split_functor.cu

* rollback

* [ROCM] fix cupti, hipcub

* update

* update

* [XPU] support xpu runtime profiler: follow up

* fix compile issue

* [kunlun] avoid compile issue in non-xpu env

also rename macro WITH_XPU_XPTI to WITH_XPTI

* move get_xpti_dependency.sh to tools/xpu

* move get_xpti_dependency.sh to tools/xpu

* call get_xpti_dependency.sh only in need

* [kunlun] support xpu runtime profiler

* fix cmake error

* add libxpti.so to paddle package

* fix for style check

* sync change in setup.py and python/setup.py.in

* remove libxpti.so from paddle output dir in this PR

* create phi so

* fix ci bugs

* fix py3 bugs

* add file

* fix py3 bugs

* fix windows bugs

* perfect so

* fix py3 bugs

* delete all static target in phi

* fix windows bugs

* fix py3 bugs

* fix ci bugs

* fix windows bugs

* fix bugs: gflags can't be linked by dynamic and static lib

* fix bugs that can not load 3rd party

* fix ci bugs

* fix compile bugs

* fix py3 bugs

* fix conflict

* fix xpu bugs

* fix mac compile bugs

* fix psgpu bugs

* fix inference failed

* deal with conflict

* fix LIBRARY_PATH bug

* fix windows bugs

* fix onednn error

* fix windows compile bugs

* fix windows compile bugs

* fix test_cuda_graph_static_mode_error aborted

* fix windows bugs

* fix mac-python3 error

* fix hip compile bugs

* change mode to static

* change to static mode

* fix ci bugs

* fix py3 bugs

* fix windows bugs

* fix bugs

* add static flag

* add PADDLE_API

* change position of PADDLE_API

* fix windows bugs

* change mode to dynamic lib

* fix windows static bugs

* deal with conflict

* fix windows unit bug

* fix coverage

* deal with conflict

* fix windows-inference

* fix py3 bugs

* fix bugs when compile type_info

* fix compile bugs

* fix py3 bugs

* fix windows bugs

* fix windows openblas

* fix xpu bugs

* fix enforce_test in windows

* update code according comment

* fix windows cmake bug

* fix windows bugs

* fix windows bugs

* delete cinn unittest

* fix cinn bugs

---------
Co-authored-by: lzydev <1528794076@qq.com>

* Reorganize the forward codes of flash-attention.

* Fix forward.

* Remove some noused codes.

* Simplify codes and fix backward.

* Change all LOG(INFO) to VLOG and fix the backward.

* add scale for AF2 flash_attn, much thanks to xreki and shaojie for debug these codes

* decrease the effect of debug print on performance

* Unify the initialize of flashattn arguments.

* Rewirte the reshape of temp_mask and temp_bias.

* API support use_flash_attn.

* Fix compiling error on CI.

* Try to crop the flash-attention lib.

* Correct the condition of whether can use flash-attn.

* Remove the softmax_out argument.

* Remove is_causal.

* Polish codes.

* Fix qkv_transpose_out's shape and scaling of Q * K.

* Update commit of flash-attention.

---------
Co-authored-by: NLiu Yiqun <liuyiqun01@baidu.com>

* add flags for phi

* fix compile bugs

* fix ci bugs

* fix inference bugs

* fix cinn' bugs

* fix cinn bugs

* perfect code according comment

* fix ci bugs

* fix ci bugs

* flash attn

* seed

* almost

* softmax

* fix workspace

* add unitest; linux only

* fix setup

* fix datatype include

* fix setup typo

* fix def scope

* new error api

* use paddle fork

* fix attr bug; complete ut

* update flash hash

* fix rng reset

* fix offset

* fix comments

* add warp transducer code

* fix codestyle

* add double complex<float> complex<double> dtype support for syevj_batched

* fix use_syevj flag for precision loss when input dtype of syevj_batch is complex128 in some case

* optimize eigh in different case

* fix missing ; bug

* fix use_syevj bug

* fix use_cusolver_syevj_batched flag

* rm "paddle/fluid/platform/complex.h" in phi

* fix codestyle with pre-commit

[PHI Decoupling] remove dependency on "paddle/fluid/platform/errors.h" and "paddle/fluid/platform/fast_divmod.h" in phi. (#47815)

* rm "paddle/fluid/platform/errors.h" in phi

* rm "paddle/fluid/platform/fast_divmod.h" in phi

* Add build option for CUDNN Frontend API

* Fix review comments

* Change namespace for cudnn_frontend.h

* Delay TensorRT registry
* Add unused define
* Fix TensorRT test
* fix function to reference
* Update trt_plugin.h

* clean glog header in public header

* move marco pos

* [Sparse]add SparseTensor mv kernel(csr*dense_vec->dence_vec, coo*dense_vec->dense_vec)

* fix CI

* add some kernel(csr*dense->csr, dense*dense->csr) of SparseTensor matmul

* fix CI

* fix CI

* fix comment

* fix comment

* Refactor port.h

* Remove some unnecessary code

* Fix CI errors