Do not enforce that if the input is sparse, we need the grad to return
sparse.

I do not force the grad to be the same format. This would require
conversion.  I do not think we always enforce this already, so I do
not want to enforce that now.

Also make theano.grad test the output is good.

Fix crash at import

Revert "Fixed dependencies for theano.sandbox.cuda import"

new citation policy

Fixed dependencies for theano.sandbox.cuda import

Opt

Make automatic insertions of dnn_conv() use the optimal cuDNN code path (2)

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

Allow opt registered in name EquilibriumDB to do not be enable by the EquilibrumDB name and use that for local_remove_all_assert opt.

Make opt local_adv_sub1_adv_inc_sub1 add assert to don't loose error and document how to get rid of them