Annotate ConvOp and GpuConv with direction hint and use this to select algorithm in conv_dnn optimizer

Remove warning caused by AdvancedSubtensor.infer_shape.

This just remove warning and to not implement the case where the
warning was generated.

Make automatic insertions of dnn_conv() use the optimal cuDNN code path

padding support for dnn_conv

Fixes a crash of casting bytes to Python string

The crash happened when cuDNN isn't used and the error message was raised. However, the message was a 'bytes' object, not a string. Casting the bytes to string solves the issue.

Fix ifelse crash and gpu support

fix gh-2262. Fix crash cause by using cudnn when we shouldn't

Support GpuCumsum on 3D array.

Slightly clean up registration of GPU convolution optimizers

Documentation for sliding windows 

Cache opt

Fix cuDNN detection

Add CUDA include dir so cuda.h is found; also add CUDA library dir to mimic configuration when compiling CUDA ops