* Support two inputs of multihead attention named qk_multihead

* [Opt Performance] Optimize custom operator performance, reconstruct python API auto-gen, add cache and use const inference

* opt AutoGradMeta implementation

* remove profiler codes

* fix unit test

* change year, 2021->2023

* fix int64_t parse bug

* autogen softmax_with_cross_entropy

* fix error in softmax_with_cross_entropy version

* add strategy force_sequential_run

* remove flag

* fix

* fix

* fix

* fix

* fix

* fix

* fix

* fix

* fix

* adjust defalut tolerance of output and grad

* fix a bug in the grad of OpTest

* fix the type of setting defalut value in optest, both forward and
backward

* add defalut

* fix test_sum_op

* adjust tolerance

* fix the tolerance of eager

* add bf16 and fp16 to the activation tests

* remove some fixs

* fix activation

* fix fp16

* fix gelu

* fix the activation tests

* add bfloat16 specialization to singrad and cosgrad

* fix bugs

* fix bugs

* add unittest

* add skip

* add fp/bf to rrelu/rrelu_grad

* git add rrelu

* fix bugs

* add fp16 and bf16 support for instance_norm

* fix /= operator which not support bf16

* fix instance_norm_grad kernel and unittests.

* fix fp32 unittests.

* fix instance_norm_kernel and unittests.

* fix instance_norm_grad_kernel and unittest threshold.

* add fp16/bf16 for instance_norm_grad_grad op.

* add bf16 dtype check.

* fix conflicts.

* fix cpu support for fp32 op and fix type in instance_norm_grad_kernel.

* fix type in instance_norm_kernel.

* fix bf16 outputs in unittests and refine codes.

* fix dx computation.

* delete unuseful params and head including.

* add fp16/bf16 for static graph.

* fix device condiction for instance_norm op.

* fix instance_norm_grad_grad and bf16 op tests.

* fix op_test to support grad of bf16 can be compared with fp32.

* remove updates.

* add self-defined grad.

* mv WITH_ASCEND_CL

* mv WITH_ASCEND

* rollback

* remove WITH_ASCEND

* remove WITH_ASCEND

* add seed control

* fix bug

* add strategy force_sequential_run

* fix

* fix

* fix

* fix

* fix

* mv WITH_ASCEND_CL

* mv WITH_ASCEND

* rollback

* remove matrix_multiply unitest

* remove run_program

* remove FLAGS_USE_STANDALONE_EXECUTOR

* [AMP OP&Test] Fix the logic of calling infer_dtype func in op test

* add fp16

fix merge conflicts

* remove zeros

* remove zeros

* apply gcc12 to py3

* apply gcc12 to py3

* fluid api clear

* fluid api clean

* fluid api clean

* fix dy2s grad name parse

* pre-commit

* bug fix

* Fix grad/ error

* Format code

---------
Co-authored-by: N0x45f <wangzhen45@baidu.com>

* fix mkdir

* update

* [Test MV] standalone_executor

* update

as

* update

as

* update codestyle

* replace matmul with matmul_v2 in fuse passes

* Remove fusion logic from matmul

* removing fusion methods

* add proper name

* adjust namespaces

* clean attrs in python tests

* delete checkpoint and restore matmul version

* remove unused code

* matmul and reshape/transpose fuses migrated

* split MatmulOneDNN headers

* fuse activation and eltwise_add

* add fuse_activation

* matmul_transpose_reshape/reshape_transpose_matmul

* matmul + elementwise_add (fused)

* activation temporary modifciation

* restore matmul(v1) version 0

* merge newest develop

* remove depedency from other PR

* revert pbtxt

* remove placeholders from matmul_v2

* add description in OPMaker

* remove matmul_v2_op.h and all depedencies

* remove dims changing in base op

* add possibility to fuse already fused_matmul

* restart broken CI

* Empty-Commit

* revert matmul_utils.h

* codestyle

* adjust imports

* add pbtxt file

* 100% matmul unit tests coverage

* trigger CI with minimal changes to develop

* adjust changes to develop

* add fused_matmul op

* inherit base ops

* add "v2"

* move OPMaker

* Gradually add fused_matmul files

* second batch of fused_matmul changes

* split infershapes of matmul_v2 and fused_matmul

* merge code from other PR

* 2023

* inherit fused_matmul from matmul_v2

* Update paddle/phi/backends/onednn/onednn_reuse.h
Co-authored-by: NTomasz Socha <tomasz.socha@intel.com>

* Update paddle/phi/kernels/fusion/onednn/fused_matmul_kernel.cc
Co-authored-by: NTomasz Socha <tomasz.socha@intel.com>

* resolve conflicts

* codestyle

* simplify isgemmlinear

* 2023

* remove import

* reuse methods

* matmul_v2_mkldnn cleanup

* simplify ExecuteMatMulV1Grad

* matmul refactored

* fc

* SetOutMemDescWithLogicalLayoutFusesSupport

* matmul_v2

* alpha support

* group repetetive funcs

* matmul utils

* execute matmul methods

* restore registered kernel names

* split header and impl files

* remove double negatives

* reduce numer of modified files

* adjust ExecuteMatmul

* add scales for ut

* dates

* limit number of modified files

* fluid imports

* remove alpha

* codestyle

---------
Co-authored-by: NTomasz Socha <tomasz.socha@intel.com>

* add kernel functions

* update kernel functions

* update func parameters' name

* create codes for gpu device

* 调整文件位置

* fix include error

* remove dependent files to phi/

* restore fused_attention_op.cu

* fix dependence errors

* fix dependence errors

* fix include error

* fix all depandence errors[build success]

* remove useless include

* recover useless include

* use phi::ToNCCLDataType

* fix namespace

* update new register code

* fix error in fused_gemm_epilogue_utils

* fix error in FusedAttentionKernel parm

* finish fused_attention registe code[build success]

* add paddle::optional

* add sig file

* fix build error

* fix a include error

* update CMkaeList

* fix parameter sequence

* add include file

* update #if before include

* fix grammly error

* update codes for DropoutParam

* remove const cast

* trans some fluid api to phi api

* add #if

* update test code

* update test codes

* recover test codes

* trans fused_attention to fluid

* move #endif to end

* move #endif

* delete useless files

* use fused attention utils and recover random seed

* remove fluid include in phi

* feat: add relu composite rule

* feat: add relu composite rule, maximum op

* feat: add relu composite rule, maximum op

* feat: add relu composite rule, polish comments

* feat: add relu composite rule, polish comments

* feat: add relu composite rule, add python api of relu

* feat: add relu composite rule, commit hook

* fix: maximum type error & ban cinn test

* fix: maximum input sequence bugs

* resolve conflicts

* fix: code style bugs

* add: relu fp16 test

* feat: add rsqrt composite rule

* feat: add rsqrt composite rule

* resolve conflicts of composite rule

* fix: delete check eager

* feat: add roll grad composite rule

* fix minus shift

* fix test roll op

* [CINN] disable CINN test_mean_op unittest to pass CINN CI

* disable test_mean_op for pass ci

* fix flash attn

* fix another API

* rem is_compiled_with_npu

* rem nup related code

* make lint happy

* rem test

* remove some tests

* Update grad_scaler.py

* fix an error

* add fp16 and bf16 for eye and frame

* fix bug

* fix bug

* fix bug

* Update test_frame_op.py

fix code style

* fix bug

* fix bug