* add variable_length_memory_efficient_attention
* update variable_length_memory_efficient_attention unittest
* update variable_length_mem_eff_attn's docs and unittest
* update variable_length_mem_eff_attn's docs
* Update test_variable_length_memory_efficient_attention.py
* Update variable_length_memory_efficient_attention.cu
* fix codestyle
* fix variable_length_fmha's docs and unittest
* fix variable_length_fmha's docs

* decompose tanh_triple_grad and add it into prim_white_list test=develop

* fix TanhTripleGradKernel bugs test=develop

* decompose tanh_triple_grad test=develop

* [XPU kernel] fix warpctc issue

* fix issue

* temporal hack to circumvent depthwise_conv2d precision issue

* reset test case

* onednn: rename macro to PADDLE_WITH_DNNL

* onednn: rename macro to CINN_WITH_DNNL

* [XPU] add pos_weight for sigmoid_cross_entropy_with_logits.

* update xdnn version.

* move decayed_adagrad_op to phi

* fix bug

* add mask

* add backword

* add enforce info

* update scale

* integrate code

* update enforce

* add enforce eq

* add error type

* update enforce

* add test_flash_attention

* Polish codes and fix compiling errors.

* Set num_splits to 0 for flash-attn with tensor mask.

* Fix the compiling error for non flash-attn case.

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

* 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.

* fix bug: feed_with_place should consider variable existence

* fix

* fix build scope

* change method to set feed var name

* remove feed_with_place to placeholder

* fix

* rename to data

* fix

* fix

* refacot aot

* update

* fix bugs

* remove some test

* fix bug

* fix bug

* fix bug

* fix bug

* update

* fix security bug

* fix security bug

* [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

[Inference] Replace groupNorm when data types are bf16 and fp16, and data format is NHWC implementation. (#55399)

* finish

* cpergroup odd

* fix bf16

* single channel

* code style

* jingdu duiqi

* add head_file

* add bf16 head file

* bf16 2

* bf16

* bf16 head

* bf16 compile

* py test

* bf16 compile

* bf16 compile

* unset py test

* nhwc

* test

* mean var

* bf16 success

* su

* ctest success

* use is_same_as

* is_same

* use is_same

* rtol

* gpu_stream

* del sigmod

* fix bfloat16 type

* use cuda_bf16_hpp

* use_cuda_arch

* bfloat162float2

* del inplace_tol

* del max_releative_tol

* temp store

* jingdu duiqi

* temp store

* plugin

* jingdu duiqi

* duiqi

* include cuda.h

* del half

* half single

* ci

* add const

* ci

* cudamemset

* del printf

* fp16 test

* add half compute

* del br16 ci

* del ci

* ci approve

* del fluid include

* fix security bug

* move prune_gate_by_capacity to phi

* fix

* fix registe info

* remove useless codes

* new ir support print op

* fix gpu bug

* fix bug

* update

* remove layout to string

* remove usless header

* polish code

* fix bug

* posolis code