This PR hotfixed the `test_cond.py` in CUDA 11.2

The reason of the bug is that the `fill_constant` op returns wrong value in the modified test case `test_extremely_simple_net_with_op_in_condition`, SWEs can use `layers.Print(a)` and `layers.Print(b)` in the test case to reproduce it and they can see the `fill_constant` returns something `e-50` instead of `1.23` and `1.25`

This PR hotfixed the bug by comparing `b` value instead of actual number, which makes sure the `cond` logic is right. **However, the PR didn't fix `fill_constant`**. We would let the SWEs who are working here to find the op bug and fix it.

* Fix test_cudnn_norm_conv and test_cudnn_bn_add_relu in CUDA11.2

* no throw in V100 for some cases

* remove useless lod copy

* fix test failed

* revert meta change

* revert tensor change

* add Tensor support for sub-functions of colorjitter
* add UT

* enable graph-engine to return all id

* change vector's dimension

* change vector's dimension

* enlarge returned ids dimensions

* Extend python einsum interface to make einsum_v2 support multi-operands and switch it to default.

* add opt_einsum dependence

* add yaml and support eager model

* fix by code review

* [Eager] Remove enable_legacy_dygraph setting

* Add more tests

* add double yaml

* add inline func

* [Dy2Stat]Fix losting pre/post hook from outermost layer while jit.save

* fix kwargs

* fix unittest

* back fl

* delete ssl cert

* .

* make warning

* .

* unittest paral degree

* solve unittest

* heter & multi cloud commm ready

* .

* .

* arm_brpc compile

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* only output is ok

* base is ok

* .

* .

* .

* .

* .

* .

* .

* .

* add switch server bin

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* adapt brpc ssl

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* .

* fix heter_server & heter_client

* .

* .

* int->int64_t

* .

* safe map in multithread

* fix heter unitest

* .

* fix code_style

* .

* opt attr eaque perf

* opt attr select code

* fix one hot infermeta

* polish get attr impl

* fix tests failed

* add testcases

* Suppport more scenes for fused_fast_ln

* fix

* Refactor Quantization

* Refactor Dequantization

* Classy solution

* Style I

* Style II

* Style III

* Use VLOG(4) for debug info

* Style IV

* add gradient merge for DistributedFusedLamb

* use master acc gradient

* fix CI ut

* polish

* remove math_function_impl.h change

* fix test_update_loss_scaling_op.py

* try to fix XPU/NPU CI

* add gm ut

* fix PIL sample mode deprecated warning

* compatible with old pil version

* [KP] fix bug when phi kernel is *_raw

* modify the static graph

* delete useless comment

* delete the phi multiply kernel case

* add VLOG(3) message

* add VLOG(3) message

* fix static graph error in phi

* fix bug in tranform model

* modify the comment

* delete useless code

* fix CI bug

* fix CI bug

* test op_test test=allcases

* fix

* avoid copy many same file

* fix for win

* test PYTHONPATH

* change path adding way

* fix win

* use old way

* use old way test=allcase

* use old way test=allcase

* set device id of Place() to get GPUContext needed by LimitGridDim in ElemwiseGradBroadcast

* fix code style