Questions remaining: here the import relation is really strange, so i use a new import here.

added op for pooling

small test fix

Add comments and code style fix

stdint should not be imported into OpenCL kernels

Generalize all [Inc]Subtensor optimization to advanced[inc]subtensor[1]

Modified where COp looks for the external implementation

Using DONE_SET method to make algorithm faster

Ynd dnn pooling

Clean up Code.
Better logic.