Merge pull request #5518 from abergeron/fix_scan_gpuarray

theano/gpuarray/opt.py

@@ -193,7 +193,8 @@ def op_lifter(OP, cuda_only=False):
                 context_name = None
                 # We replace if any input is a host_from_gpu
                 for i in node.inputs:
-                    if i.owner and i.owner.op == host_from_gpu:
+                    if (i.owner and i.owner.op == host_from_gpu and
+                            move_to_gpu(i)):
                         context_name = i.owner.inputs[0].type.context_name
                         replace = True
                         break
@@ -922,7 +923,7 @@ def local_gpua_lazy_ifelse(op, context_name, inputs, outputs):
     c = inputs[0]
     inps = []
     for v in inputs[1:]:
-        if isinstance(v.type, tensor.TensorType):
+        if isinstance(v.type, tensor.TensorType) and move_to_gpu(v):
             inps.append(as_gpuarray_variable(v, context_name))
         else:
             inps.append(v)

theano/scan_module/scan_op.py

@@ -619,7 +619,8 @@ class Scan(PureOp):
         # in this case is just a int saying how many steps of this output we
         # need to store. This input does not have the same dtype, nor is it the same
         # type of tensor as the output, it is always a scalar int.
-        new_inputs += self.outer_nitsot(inputs)
+        new_inputs += [as_tensor_variable(ons)
+                       for ons in self.outer_nitsot(inputs)]
         for inner_nonseq, _outer_nonseq in zip(self.inner_non_seqs(self.inputs),
                                                self.outer_non_seqs(inputs)):
             outer_nonseq = format(_outer_nonseq, as_var=inner_nonseq)