- It provides a unified style to express graph and kernel for user.
- It provides a unified IR to represent graph and kernel for developer.
- It breaks the boundary between graph and kernel.
- It provides more opportunities to do compile optimization.

2.Fix core dump of exporting onnx model when device target is 'GPU'
3.Fix numbers of arguments and graph parameters check error
4.Fix log prefix of some files of gpu submodule is error

Merge pull request !1339 from JichenZhao/layernorm_mean_var_shape

Merge pull request !1133 from zhangbuxue/support_tensor_get_value_by_tensor_index

Merge pull request !1261 from changzherui/add_tensor_init

support tensor set value by tensor index

Merge pull request !1228 from liuxiao/UnsortedSegmentMin

Merge pull request !1099 from fary86/fix_bug_for_can_not_find_opadapter

Merge pull request !1062 from liuxiao/softplus-softplusgrad

Merge pull request !971 from Simson/push-to-opensource

fix DiagPart constant folding issue

fix argmin output type check

fix atan2 doc error

fix remove FusedBatchNorm and its grad

Merge pull request !75 from changzherui/syn-code428

Merge pull request !731 from fary86/remove_extra_empty_string_in_log

Merge pull request !494 from zhoufeng/cmake-sub

Merge pull request !691 from zhangbuxue/fix_bug_of_ops

Merge pull request !682 from liuxiao/fix-bug

Merge pull request !597 from ghzl/trans-const-to-variable-in-assign