* 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

* increase coverage

* add onednn tests to ctest

* remove fusion logic from base matmuls

* add sigmoid composite rule

* add python api

* fix code style.

* add check_prim=True

* add sigmoid fp16 unit test.

* fix code style.

* rm bf16 check_prim

* fix code style.

* find relevant testcase

* remove with_data_parallel in test_dist_base

* fix bugs

* fix test_minst_dgc_nccl

* fix typo

* fix test_dist_mnist_gradient_merge

* rm TestDistMnistNCCL2DGCMultiCards

* fix dist_mnist

* polish code

* fix different PaddleCloudRoleMaker

* support warnings for extra attrs and axis in non default value

* support extra_attr warnings and add test case

* fix dict name

* polish

* polish

* polish

* [phi] move jit kernels from fluid to phi

* [phi] fix paddle::phi err

* [phi] fix windows 'posix_memalign': identifier not found

* [phi] fix windows 'posix_memalign_free': identifier not found

* [phi] fix readme directory structure, fc_functor  paddle::platform

* fix issue accur in Mac OS m1 env

* use __arm64__ directive to distinguish m1 chip

* decouple memory copy

* fix ci bugs

* fix ci compile bugs

* fix rocm compile

* fix ci bugs

* comp gather_nd_grad

* fix

* test no cinn

* fix

* fix cinn

* patch on gloo/types.h

* fix patch

* change patch dir

* add patch

* [CustomDevice] fix process_group_custom api

* update

* update

* update

* update

* Check structed kernel for new executor static build

* Update code

* Ready for resnet50

* Move transfer_dtype to phi

* Ready for transformer

* Fix CI errors

* Fix layer_norm InferMeta

* Remove layer_norm infermeta 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

* add_trt_tile

* tile_trt

* fix performance drop in BF16 models

* fix test_cpu_quantize_squash_pass

* fix prim_op_test when python api outs is different with kernel sig

* add elementwise op prim test

* fix unit test

* add bfloat16 for full in static  prim api

* empty-commit

* close bf16 test

* polish elementwise tests

* fluid clean: remove parallel and parallel_helper api

* fix: fix the import path.

* fix DataParallel imports issue

* fix attrs copy error

* fix bert by fix slice error

* fix op test

* fix gcc12 error

* patch on device.cc

* fix gcc error while compiling gloo

* [dy2static] bugfix: make stop_gradient a cache key
1. make stop_gradient cache key in dy2static.

* fix ci errors

* fix ci error

* fix ci error

* fix ci error

* fluid clean: remove dygraph_utils._append_bias_in_dygraph

* fix func name and imports

* polish codes according #50813

* [getCurrentCUDAStream] Add C++ API getCurrentCUDAStream

* change get->Get

* wrap with macro

* use Get instead of get

* register fp16 and bf16 kernel for uniform_random

* fix compile

* support selected_rows

* add ut

* revert cpu

* fp16 test skip cpu