功能：本PR的目标是提高attention模块的计算性能。
为了减少框架层对op的调度开销，本PR通过在C++层手动实现attention模块，对外提供attention 大op；
为了减少防存开销，本PR采取了两种优化方法：
（1）在q,k,v计算时通过共享输入X，将该处的gemm，transpose和bias add从三次调用减少为一次；
（2）使用kernel融合优化技术，在不同cuda kernel之间通过寄存器传输数据；

* add viterbi decode cpu kernel

* add viterbi decoder api in paddle.text

* add a data buffer once to avoid create many small pieces of data buffer frequently

* fix viterbi max_seq_length bug

* fix seq_len=1 bug

* fix device context

* move split out of for loop

* remove INVERSE_SUB

* remove 2 GET_CAST_MASK

* remove 1 loop

* remove Functor

* add to_static deploy code

* use MAX_FUNC instead of ELE_MAX

* add MaxFunctor

* impl max_func

* remove MaxFunctor

* remove cast op

* use REGISTER_OP_WITHOUT_GRADIENT

* add viterbi cuda kernel

* add FIX_BLOCKDIM_CASE macro

* add MKL add, mul; add get data mask

* add arange mkl impl

* add CPU Argmax

* add cpu gather

* use EXECUTE_MKL_ELEMENT_BINARY_OP instead of some ADD, MUL

* use SameDimsBinaryOP instead of EXECUTE_MKL_ELEMENT_BINARY_OP

* use SAME_DIMS_ELEMENT_BINARY_OP

* add SimpleBroadcastBinaryOP

* use int instead of int64_t to accelerate

* optimize SimpleBroadcastBinaryOP

* optimize SimpleBroadcastBinaryOP

* optimize performance in both single thread and multithread situation

* remove useless line

* remove useless code

* add CREATE_TENSOR_BUFFER macro

* add INIT_REQUIRED_TENSOR macro

* add comment

* fix windows ci

* add viterbi unittest

* remove cuda add functor

* remove cuda equal

* remove a template function

* fix windows ci

* fix windows dtype

* remove some template instance

* remove useless header file

* remove some blockdim

* remove transpose impl

* accelerate cpu performance on single thread situation

* viterbi_decode->crf_decode

* rename crf params name

* add viterbi api test

* remove useless import

* add enable_static

* use viterbi decoder

* fix viterbi len=1

* fix  viterbi unittest

* remove useless comments

* reconstruct viterbi decode

* remove ADD,SUB,MUL structure

* fix coverage

* remove CREATE_TENSOR

* add name args

* crf.py->ops.py; with_start_stop_tag->include_start_end_tag

* update crf_decode en docs

* fix viterbi decode en docs

* fix some review comments

* add FIXED_BLOCK_DIM_CASE in cuda

* push_back->emplace_back

* crf_decode->viterbi_decode; include_start_end_tag->include_bos_eos_tag

* paddle.text.ops.viterbi_decode->paddle.text.viterbi_decode

* fix viterbi_decode en docs

* Fix a bug in ReadData, ReadDataBc and ReadDataReduce when NX != 1
* Update the implement of reduceAnyKernel according to kernel primitive api

 [Cherry-pick 2.2] Correct the return type of elementwise kernel to avoid many compiling warnings. (#35839) (#35868)

Cherry-pick #35839

* - REorder disabling caching

* - compilation fix

* - another compilation fix

* - another compilation fix

* - compilation fix

* - Fix

* - yet another compilation fix

* - suppresingly another compilation fix

* - lint

* - fix after review

* - fix

Implement FunctionTraits to support two kinds of elementwise functor and remove some old codes for broadcast. (#35688)

Revert "Implement FunctionTraits to support two kinds of elementwise functor and remove some old codes for broadcast. (#35487)" (#35686)

Implement FunctionTraits to support two kinds of elementwise functor and remove some old codes for broadcast. (#35487)

multiply supports bool  

* Add the extra flag for the some ops

* fix the compile problem in matmul extra

* add elementwise_pow_grad_npu

* fixed bug for CI

* deal with comments

* fixed bug for CI

* deal with comments

add axis check for elementwise op while the dimension of x is equal to the dimension of tensor (#35340)

* add elementwise max grad op for npu

* add elementwise max grad op for npu

* add elementwise max grad op for npu

* add elementwise max grad op for npu

* add elementwise max grad op for npu

[oneDNN] disable caching oneDNN primitives in  matmul v2, Reduce grad and elementwise_add grad, expand_v2 (#35132)

* - grad caching disabled of matmul_v1

- compilation fix

- compilation fix

* - reduction removed

* - Matmul v2 disabled caching

* Draft of further changes

* - workaround for reducegrad

* - fixes to UT

* - fix to compilation

* - another fix

* - fix

* - Added softmax without caching

* - Binary is no longer manually cached

* - Activation onednn caching removed

* - Removed manual caching of activation

* - modified UT

* - fix

* - fix

* - fixes to building

* - fix

* - fix

* - fix to UT

* - Faulty UT workaround

* - approval workaround

* - Fixes after review

* - compilation fixes

* - more lint fixes

* - more fixes after review

* - fixes after another round of review

* - hopefully compilation fix

- compilation fix

This reverts commit 0a5c99e8.

* - Added softmax without caching

* - Binary is no longer manually cached

* - Activation onednn caching removed

* - Removed manual caching of activation

* - modified UT

* - fix

* - fix

* - fixes to building

* - fix

* - fix

* - fix to UT

* - Faulty UT workaround

* - approval workaround

* - Fixes after review

* - compilation fixes

* - more lint fixes

* - more fixes after review

* - fixes after another round of review

* add broadcast supporting for elementwise_add

* add broadcast supporting for elementwise_add

* add more tests

* remove the redundant code

* update

* fix place error in unittest

* remove skip.If

* enhance error message when x or y is empty in elementwise_op

* format code

* format code

* - fix to #33282

* - Increased threshold for elementwise_mul_bf16 grad

* -disabled faulty UT

* - fix to approval

* refine ~npuOpRunner

* implement destructor and forbid copy

* use reference to avoid copy

* use const reference

* relax adam precision

* fix top_k

* modify complex template for elementwise ops

* modify mul, div grad struct

* add complex template for CudaShuffleDownSync CudaShuffleXorSync funcs and fix the bug when delete cuda<9000

* fix shuffle func args bug

* fix shuffle func args bug

* fix shuffle func args bug