* Add fused_scale_bias_relu_conv_bnstats op

* Review changes

* Fix no CUDNN Frontend build

* Fix PADDLE_ENFORCE format

* Fix PADDLE_ENFORCE CI error

* Rename kernel filename

* Refactor unittest to use paddle eager_op_test

* Fix padding bugs

* Review changes

* test=cuda117

* test=cuda117

* update

* repalce gflags header

* replace DEFINE_<type> with PD_DEFINE_<type>

* fix bug

* fix bug

* fix bug

* update cmake

* add :: before some paddle namespace

* fix link error

* fix CI-Py3

* allow commandline parse

* fix SetFlagsFromEnv

* fix bug

* fix bug

* fix CI-CINN

* fix CI-Coverage-build

* fix CI-Windows-build

* fix CI-Inference

* fix bug

* fix bug

* fix CI-CINN

* fix inference api test

* fix infer_ut test

* revert infer_ut gflags usage

* update

* fix inference

* remove flags export macro

* revert inference demo_ci gflags usage

* update

* update

* update

* update

* update

* update

* update

* update

* fix bug when turn on WITH_GFLAGS

* turn on WITH_GFLAGS

* fix bug when turn on WITH_GFLAGS

* fix bug when turn on WITH_GFLAGS

* update

* update and add unittest

* add unittest

* fix conflict

* rerun ci

* update

* resolve conflict

* Update CUDNN Frontend API to v0.9.1
- Remove old patches
- Remove workarounds that are no longer needed

* Fix test_switch_autotune

* 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>

* Try to increase the repeat of autotune and fix the setting of allow_tf32_cublas.

* Change the repeat of cublaslt to 10.

* Use FLAGS_cublaslt_exhaustive_search_times as repeats.

* Fix compiling error on CI.

* Polish the key and simplify codes.

* test,test=develop

* test,test=develop

* test,test=develop

* test,test=develop

* first commit

* restruct c++ interface to divide linear from matmulwithcublaslt

* finish building in cublaslt impl

* fix code bugs

* fix host cost

* add some changes

* [enforce.h Decouple gflags.h] Move gflags.h from enforce.h to enforce.cc

* Add gflags.h for other files

* Add gflags.h for other files

* Add gflags.h for blas_impl.hip.h

* Add gflags.h for miopen_helper.h

* optimization for fused linear op

* fix code format

* optimization for linear fused forward

* merge with develop

* fix bugs for gemm_ephilog

* package of cublaslt ephilogue type with enmu

* final fix before code reviewing

* fix missed fusedType typo

* fix code according to review suggestions

* fix windows ci error

* change location of MatmulPlanner

* add some changes for compiler error fix

---------

* first commit

* finish base work

* modification for good

* fix for cache setting and gather the algo and desc as one data for cache storage

* fix for cache setting and gather the algo and desc as one data for cache storage

* install pre-commit check

* implement of matmul using cublasLt instead of cublas

* Update matmul_kernel_impl_via_blasLt.h

---------
Co-authored-by: Nzhangbopd <1299246947@qq.com>
Co-authored-by: NBo Zhang <105368690+zhangbopd@users.noreply.github.com>
Co-authored-by: NLiu Yiqun <liuyiqun01@baidu.com>

Co-authored-by: Nzhangbopd <1299246947@qq.com>

* First Commit.

* add some codes

* add elementwise loader

* fix code styles

* merge with develop

* add some changes both in elementwise and transpose

* add init operation in broadcast kernel.

* change codes according to pr suggestions about transpose file

* fix error for op-benchmark ci

* fix according to ci

* Refactor conv_kernel and conv_grad_kernel to provide interface for CUDNNv8 implementation

* Fix macro

* Add implementation for conv_kernel and conv_grad_kernel

* Modification after rebase onto latest develop

* Modify plan cache to comply with the API of phi::autotune

* Refactor to reduce duplicate code

* Review fix:
- move functions in  conv_kernel_impl_v8.h and conv_grad_kernel_impl_v8.h to conv_kernel.cu and conv_grad_kernelk.cu
- add const specifier for input tensor
- add logging when plans fail to execute
- move CudnnConvBwdFilterV8 and CudnnConvBwdDataV8 to conv_cudnn_frontend.h

* - move plan building outside of cache

* Fix ROCM build

[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

* first commit

* transpose_kernel_optimization

* first complishment of transpose op

* second commit

* refine code logics of tranpose_kernel

* refine transpose kernel

* first commit

* fix DtoD copy bugs for hip

* refine code according to the PR advice

* change dim to int64_t type.

* fix some type error

Enable to record whether the conv algo is got by exhaustive search to fix autotune cache bug. (#47065)

* first commit

* clarify the quotes

* change code style format

* rerun for ci

* optimizer conv alog speed

* code polish

* remove useless code

* fix compile error

* fix cpu compile error

* not use cudnn alog t

* add search cache max number

* polish code

* fix cache test bug

* add groups data format to conv args

* fix cache test bug

* fix cudnn_deterministic bug

* fix test switch auto tune bug

* fix test swith autotune bug;

* fix conv cache bug

* fix cache test error

* fix cache test bug

* fix windows mac compile error

* fix workspace search error

* update cudnn cache

* fix cache test bug; test=develop

* fix autotune swith test error

* polish code

* oplish code

* 2nd part of transpose update

* add switch_auto_tune option.

* add some changes according to Ci

* refine the structure of auto_tune_base.

* merge develop changes

* reset the switch_set_range and change unittest of transpose auto-tune

* change the kernel auto-tune logits

* perfect copy

* deal with conflict

* deal with conflict

* fix compile bugs

* fix unittest bugs

* change code format

* deal with conflict

* modify code by review

* fix ce bugs

* fix ce bugs

* add lo

* perfect code format

* deal with conflicts

Transpose optimization with assitant of  Chengdu Supercomputing Center and auto_tune operation (#42704)

* change cudnn helper for auto-tune

* Add FLAGS_use_autotune to set the global status of autotune and change the order of choosing algorithm.

* Fix the bug in calculating and printing current step cache hit rate.

* Improve the autotune cache and fix unittest.

* Change the key from AlgorithmType to int64_t.

* Fix unittest for cpu-only env.

* change ChooseAlgoByWorkspace for heuristic mode
Co-authored-by: NLiu Yiqun <liuyiqun01@baidu.com>

* Using the maximum workspace_size of all alogirhms to limit the workspace size in exhaustive search mode.

* Use the system cudaMalloc and cudaFree to allocate workspace during searching.

* Enable switch of two kind of workspace setting methods.
Co-authored-by: NLiu Yiqun <liuyiqun01@baidu.com>

* switch autotune

* implement AutoTuneCache

* implement AutoTuneCache class

* add pybind api

* add dygraph test

* support static mode and eager mode and improve unittests

* rename the SwitchAutoTune Class and improve tests

* improve AutoTuneStatus and reduce the cost of tests

* for 1st time interface combine.

* modification with kernel factory

* first auto_tune version.

* first version.

* basic version

* add warm up step.

* a debug version.

* optimize the functionality of class auto_tuner.

* add some quotes for optimized auto_tuner class.

* add some quotes for optimized auto_tuner class.

* add namespace.

* modification according to the advices

* replace fluid header with phi header.

* replace fluid header with phi header.