Enable the use of clang to cross-compile for Android

Add im2col functor

Neon Depthwise Convolution Transpose Function

Add switch order layer for FCN model

add VarDesc design

Add squared_l2_distance_op

Rename `LargerThan` to `GreaterThan`

* add offset implementation to lod tensor design

add var desc proto

Update graph construction design doc

Change `Op::GetAttr` to `Op::Attr`

fix redeclaration of 'paddle::StatPtr __stat'

update the how to write op doc.

fix clang build and run error

Op refine for making build system more automatic

update the doc for how to write the operators.

Add cosine similarity operator.

make dim int to int64_t

update the doc of softmax_op.

Fix #3902