Do not constant-fold Alloc if it is the output

theano/sandbox/cuda/basic_ops.py

@@ -2005,13 +2005,18 @@ class GpuAlloc(Op):
         return (3,)
 
     def do_constant_folding(self, node):
-        if any([isinstance(client[0].op, (
+        for client in node.outputs[0].clients:
+            if client[0] == 'output':
+                # If the output is a constant, it will have to be deepcopied
+                # each time the function is called.  So we do not fold.
+                return False
+            elif (not isinstance(client[0], basestring)
+                    and isinstance(client[0].op, (
                         tensor.IncSubtensor,
                         tensor.AdvancedIncSubtensor1,
                         GpuIncSubtensor,
                         GpuAdvancedIncSubtensor1
-                        ))
-                for client in node.outputs[0].clients]):
+                        ))):
                 return False
         return True
 

theano/tensor/basic.py

@@ -2617,11 +2617,17 @@ class Alloc(gof.Op):
         return self.make_node(eval_points[0], *inputs[1:]).outputs
 
     def do_constant_folding(self, node):
-        if python_any([isinstance(client[0].op, (IncSubtensor,
+        for client in node.outputs[0].clients:
+            if client[0] == 'output':
+                # If the output is a constant, it will have to be deepcopied
+                # each time the function is called.  So we do not fold.
+                return False
+            elif (not isinstance(client[0], basestring)
+                    and isinstance(client[0].op, (
+                        IncSubtensor,
                         AdvancedIncSubtensor1,
                         AdvancedIncSubtensor,
-                                                 ))
-                       for client in node.outputs[0].clients]):
+                        ))):
                 return False
         return True