* Fix bug of reduce_sum op. When input.numel() > INT32_MAX, its result
is wrong.

* Remove climits.

* Fix bug of hybrid parallel strategy with recompute using bf16.

* Fix bug of recompute_hybrid ctx.amp_dtype

* Fix bug of amp_dtype.

* Fix bug of auto_cast.

* * add comm context for device context

* add broadcast phi operator kernel and api

* add broadcast support dtype, update ut

* fix broadcast bfloat16 type

* fix ut

* update test_collective_broadcast_api timeout to 300

* add visit_place to phi/core/utils

* remove reference to operator.h in phi/kernels/funcs/math_function.h

* update data type from framework.proto to phi

* fix enforce error in fluid

* [cinn] add elementwise_pow backward

* [cinn] update unnitest

* [cinn] update by comments

* [cinn] for ci

* [cinn] for ci

* [cinn] for ci

* [cinn] for ci

* [cinn] for ci

* add output defs for qr kernel

* Revert "add output defs for qr kernel"

This reverts commit 2704d4053219418245180564aa5fb2f3f19d1af5.

* remove op from blacklist

* remove with_data_parallel in collective_optimizer

* add comm op

* fix collective optimizer

* remove check_err_log=True

* add isfinite

* add isnan & isinf

* add xpu isNaN

* mean fp16

* fp16

* [AMP OP&Test] mean append bf/fp16

* means append more bf16 uts

* format class name

* fix ci

* fix for windows

* fix issue

* fix redundancy

* fix redund

* fix elewise_max ut bf16 numeric delta

* remove use func

* move entry_attr

* remove evaluator

* fix evaluator

* modify optimizer tests

* lint

* Optimize tcp store initialized

* fix: use phi::errors

---------
Co-authored-by: Wen Sun <syl1887415157@126.com>

* tile_op

* fix bfloat16 x

* 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

* modify the 'import' position

* rollback

* move 'from paddle.fluid.io import save_inference_model' to head of api.py

* add output defs

* add output defs for kps

* support transpose test on prim and cinn

* fix transpose logic and polish eager_op_test

* move io

* fix ci bugs

* fix ci bugs

* fix py3 bugs

* fix example code

* fix example code

* fix text

* fix text

* deal with ci bugs

* perfect code according comment

* delete import batch

* [benchmark] add cmake graph

* [benchmark] add cmake graph

* [benchmark] set gpu_graph off defaultly