* fix bug in cross_entropy in static mode

* fix ci-coverage

* add autogen code for clip_by_norm op

* bug fixed

* Optimize performance of unique kernel

* fix ci

* add autogen code support for squared_l2_norm_op

* Update ops.yaml

* add bf16 support and bf16/fp16 unittest for pool2d

* add include files

* dbg

* reformat

* reformat

* modify code according to review comment

* remove duplicate code

* remove dup code

* remove useless include

* dbg

* [DO NOT MERGE] adadelta lr support

* [DO NOT MERGE] gpu support

* [test] follow torch

* fix acc update order

* for ci

* [bug fix] update master para

* [bug fix] update test

* [bug fix] for ci test

* for ci

* fix xpu

* [adadelta fix] del fluid head file

* for ci

* del notes

* [AMP OP&Test] support bf16 for batchnorm

* codestyle

* Update batch_norm_grad_kernel.cu

* Update batch_norm_kernel.cu

* fix codestyle

* fix

* fix

* fix

* fix

* fix

* Update batch_norm_kernel.cc

* support auto generate static for randperm

* remove enforce in randperm infermeta

* add bfp16 test for isfinite

* fixed for ci

* deal with comments

* fixed test

* skip test in cpu

* deal with comments

* fixed for ci

* fixed testcase

* fixed for ci

* fixed for testcase

* fix error in generator/type_mapping.py

* support auto generate for op merged_momentum optimizer

* support auto generate for flatten (flatten_contiguous_range)

* add data_type for flatten_grad

* Add output defs for eigh kernel

* fix

* update

* update

* fix

* fix

* add autogen code support for reverse op

* bug fixed

* support auto generate for op momentum optimizer

* remove momentum_op.* and update signature

* fix dgc momentum op maker error

* autogen segment_pool

* delete legacy_dygraph about segment_pool

* add_fp_bf_for_flip_gaussian_random

* forget convert uint

* fix some error

* fix some error

* [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

* add autogen code bilinear_tensor_product

* [phi] rm cc file

* autogen softmax_with_cross_entropy

* fix error in softmax_with_cross_entropy version

* [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

* 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.

* fix divide zero bug for softmax_with_cross_entropy

* change the single test way

* can run but slow. the most important is that I do not know why it slow

* remove some useless commet

* change the copyright to correct

* remove some useless change

* if repeat_times == 1, we will not use BroadcastKernel