test=develop

* optimize quant_dequant_fuse_pass, test=develop

* update, test=develop

* update, test=develop

* fix bug for accessing the removed node, test=develop

* set the bias of int8 conv as float, test=develop

* support weight quantization, test=develop

* up, test=develop

* up, test=develop

* up, test=develop

1. add conv_5x5s2_dw to support any padding
2. add 1x1s2pooling impl
3. fix conv dw 3x3 s1p01 bug

* Support bitman backend

* temporarily remove x86 fuse test=develop

* remove useless logs test=develop

[LITE][CORE][NPU][XPU] Recording the data type when mutable_data() is called, and supporting the type inference from the tensor for the op bridges (#2735)

temporarily remove cuda fc fuse because we don't support cuda fc now

* Enhance fc_fuse_pass to enable fusing relu to fc_op
test=develop

* restrict fusing relu in x86
test=develop

X86 and cuda compile simutaneously cmake ..  -DCMAKE_BUILD_TYPE=RelWithDebInfo  -DWITH_MKL=ON           -DLITE_WITH_CUDA=ON           -DWITH_MKLDNN=OFF           -DLITE_WITH_X86=ON           -DLITE_WITH_PROFILE=OFF          -DWITH_LITE=OFF           -DLITE_WITH_LIGHT_WEIGHT_FRAMEWORK=OFF           -DWITH_PYTHON=OFF           -DWITH_TESTING=ON           -DLITE_WITH_ARM=OFF           -DLITE_ON_TINY_PUBLISH=OFF           -DCUDNN_ROOT=/usr/local/cudnn/           -DLITE_BUILD_EXTRA=ON (#2708)

x86 and cuda compile simutaneously

* Fix the compiling error which occurs when specify the ddk_root path and build for huawei NPU.

* Refine the registration of op bridges and make it similar to the registration of op and kernel.

* Refine the interfaces of the graph and node for op bridges, and support creating constant and data node automatically according to the attribute 'persistable' of the target tensor.

* Add the unit test of the scale and softmax op bridge for NPU.

Optimize the execution of RuntimeProgram by saving the bool whether the op is feed/fetch op. (#2703)

test=develop

add clear for tensor

* Support multiple types for XPU and NPU op bridges

* Add lookup_table, gather, slice, stack and scale op bridges for supporting BERT

* Fix the definition of lookup_table kernel for X86

add sequence_pool_concat fuse pass

add fuse kernel

add var_conv_2d + relu fuse pass

* [XPU] add transpose bridge and unit test

test=develop

* [ARM] change global pooling choose kernel policy, test=develop

* Support Mask RCNN2 (#2588)

* add cv image process

* fix arm liunx build error

* add LITE_WITH_CV defien to make cv, test=develop

* fix cv format, annd add describe in utils/cv

* delete some Meaningless comments, test=develop

* set LITE_WITH_CV=OFF in build.sh, test=develop

* delete cv_enum.h in utils/cv, push the contents in cv_ennum.h to paddle_image_preprocess.h, test=develop

* according to reviews to redefine paddle_image_preprocess.h, test=develop

* add detailed note of flipParam, test=develop

* fix format in paddle_image_preprocess.h, test=develop

* fix error when build x86. test=develop

lite_with_X86 does not contain lite_with_cv

* fix cmake error in llite/CMakeLists.txt, missing mkdir cxx, test=develop

* according to review change, test=develop

* chang grb to rgb, test=develop

* add elemetnwise mul constant elimination and deconv+relu, deconv+batchnorm fusion, test=develop

* fix format, test=develop

test=develop

* [LITE][OPENCL] Add depthwise_conv_3x3 opencl kernel

* [LITE][OPENCL] Add depthwise_conv_3x3 opencl kernel. test=develop

* [LITE][OPENCL] Add Pool opencl kernel. test=develop

* update profiler, test=develop

* warm up times of profiler, test=develop

[LITE][NPU][XPU] Refine subgraph pass, and support NPU/XPU model generation at execution time (#2576)

* add 3 layout for opencl image. test=develop

For multiple-use parameters, only copy once

修改了选kernel的逻辑，默认从模型文件中读取出lod_tensor的data type，在static_kernel_pick pass中如果kernel输入输出的类型与读取的data type完全一致，则选择该Kernel的概率增大。

- 增加 从模型文件__model__读取lod_tensor的data type到cpp::vardesc

- program中增加unordered_map<string, type>字段，并在 Program::PrepareWorkspace中对该字段赋值

- 修改了node.h文件，将const Type* 更改为Type*，并在SSAGraph::Build过程中为符合条件的type*赋值

- static_kernel_pick_pass中添加新规则，如果kernel的输入类型输出类型与__model__中存储的类型的一致，则score*=2。

- 支持模型中用到sequence_reverse_float kernel（输入输出均为float）和sequence_reverse_int64 kernel（输入输出均为int64），能够根据输入输出type选kernel

* add arm split lod tensor, test=develop

* add arm merge lod tensor, test=develop

* update split merge lod tensor, test=develop

* add reduce_prob op, test=develop

* support mask_rcnn succeed, test=develop

update cuda kernels to run content-dnn model

* init resnet cuda int8 support
test=develop

* refine cuda unit test
test=develop

* add the forgeted file.
test=develop