* rm transfer_cast_op_pass

* rm header

* clean code

* fix ci

* fix ci

* fix ci 2

* merge recent changes

* fix setting pipline

* Add peak memory log for CI

* Change VLOG to std::cout

* Move print code to test_runner.py and paddle_gtest_main.cc

* Fix typo

* Fix conflicts

* Updata message format

* Fix CI errors

* Add FLAGS_enable_gpu_memory_usage_log

* Fix CI errors

* update xpu depends
*test=kunlun

* minor
*test=kunlun
Co-authored-by: Nroot <root@yq01-sys-hic-p40-0091.yq01.baidu.com>

* fix bug of std::move and others

* fix an compile error in debug mode

* fix wrong copy assignment operator
Signed-off-by: Ntiancaishaonvjituizi <452565578@qq.com>

* reformat
Signed-off-by: Ntiancaishaonvjituizi <452565578@qq.com>

* reformat
Signed-off-by: Ntiancaishaonvjituizi <452565578@qq.com>

* fix ArrayRef constructor following llvm

* fix format

* fix conflict with master

* Fix leading dimension setting error in fused_gemm_epilogue_grad_op.

* Add dyload to cuBlasLt functions.

* Added cublasLtMatmulAlgoGetHeuristic to improve performance.

* Added FLAGS_cublaslt_exhaustive_search_times to cublasLt epilogue

* Added UTs to FLAGS_cublaslt_exhaustive_search_times

* Added warmup runs in algo searching of Gemm epilogue.

* Update copyright and documents.

* Fixed error handling.

* changes for md in tensor

* ci fix

* Temporarily limited dims for test

* ci fix

* removed unnecessary includes

* added reviewers suggestions

* checkouted two files to avoid changing more than 19 in single PR

* minor fix

* reverted one file to reduce files changed to 19

cinn_launch_op: optimize the overhead of preparing variables before executing cinn compiled program (#41777)

* optimize preparation overhead before executing cinn compiled program

* update code notes

* fix flag annotation

* add a flag of auto-tune feature beforehand

* 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>

* [MLU] add mlu new profiler

* fix format

* regist elementwise_xxx

* support construct scalar using non-cpu tensor

* fix bugs when run unittest

* fix compile bugs

* fix bugs when run ci

* fix compile bugs

* fix bugs when move copy

* perfect unit test

* perfect unittest

* update according to comment

* add target dependency

* deal with conflict

* fix bugs when run unit test

* fix unit test bugs

* Add UT

- Added missed data_layout

- Added missing conversions

- NDHWC added

- NDHWC support in data_transform

- another fix

- condddate change

- fix

u- fix

- fix

- fix

- fix

- fix

- fix to hack

- compilation fix

- fix to automatic merge

* - reduced UT

* - fix

* - lint

* - fix to lint

* support multi layer and bidirection of lstm_grad, *test=kunlun

* support multi layer and bidirection of lstm_grad, *test=kunlun

This reverts commit fe214af2.

* support construct scalar using non-cpu tensor

* fix bugs when run unittest

* fix compile bugs

* fix bugs when run ci

* fix compile bugs

* fix bugs when move copy

* perfect unit test

* perfect unittest

* update according to comment

* add target dependency

* update sign op on xpu. test=kunlun

* fix typo. test=kunlun

* init commit no push

* collect comile errors

* bitwise UT

* fix compile problem

* cancel comments

* restore miss deletion

* fix compilation

* fix UT

* NO stash in multiple branch at the same times

* fix error

* combine .cu from gpu and kps

* replace gpu by kps

* fix by Chen-weihang

* Revert "Fix kps compile error in Junhui logic compare bitwise"

* fix backend test

* rm comments
Co-authored-by: NChen Weihang <chenweihang@baidu.com>

* fix_paddle_numel_check

* fix_paddle_numel_check

* 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>

* [ROCm] fix dcu error in device event base, test=develop

* fix, test=develop

* no

* maintain old profiler

* fix old dygraph record event

* ignore some failed test for KL2
*test=kunlun

* minor
*test=kunlun

* minor
*test=kunlun