* Use MKLDNN for non-strided slice ops.

* Remove unused variables.

* Add utility function is_strided to graph_util.
Add datatype check for slice in cpu_assignment.

* Create blocked mkldnn memory desc for output when input format is blocked.

* Add slice converLayout fusion to cpu_post_layout_optimizations.

* Rename tensor_view.
Fix a bug.

* Rename a variable.