init . ARM INT8 support

See merge request inference/paddlelite!33

make all ARMContext share the same DeviceInfo, and export SetRunMode to allow users to set the thread num

fix comments

Merge branch 'incubate/lite' of http://10.87.145.36/inference/paddlelite into hongming/refine_arm_context

test=develop

speedup and split ci

See merge request inference/paddlelite!32

add scale eliminate pass

See merge request inference/paddlelite!18

1. add fake_quant fake_dequant op
2. add quant_dequant fuse pass
3. fix bug for passes for arm
4. softmax axis problem

add mobilenet_v2 test

See merge request inference/paddlelite!31

test=develop

add GoogleNet realted kernels unitest

See merge request inference/paddlelite!27

support clang compile on arm

See merge request inference/paddlelite!29

test=develop

lite/enhance model

See merge request inference/paddlelite!16

test=develop

test=develop

[LITE][ARM] Add transpose, transpose2 operator, kernel of arm cpu. test=develop

See merge request inference/paddlelite!24

add elementwise_add_activation fuse pass & op

See merge request inference/paddlelite!21

fix split op

See merge request inference/paddlelite!22

test=develop

Tangjian/lite/cmake

See merge request inference/paddlelite!19

[LITE][ARM] Add concat kernel of arm cpu. test=develop

See merge request inference/paddlelite!17

add dropout arm kernel and unit test

See merge request inference/paddlelite!15

make lite CI compile faster

See merge request inference/paddlelite!12