remove safe_to_cpu, and re-wrote safe_new

theano/scan_module/scan_utils.py

@@ -30,24 +30,31 @@ import theano
 # Logging function for sending warning or info
 _logger = logging.getLogger('theano.scan_utils')
 
-
-def safe_new(x):
-    if isinstance(x, numpy.ndarray):
-        x = tensor.as_tensor_variable(x)
-    if cuda.cuda_available and isinstance(x.type, cuda.CudaNdarrayType):
-        return tensor.TensorType(
-            broadcastable = x.type.broadcastable
-            , dtype = config.floatX)()
+def safe_new(x, tag = ''):
+    """
+    Internal function that constructs a new variable from x with the same
+    type, but with a different name ( old name + tag). This function is used
+    by gradient, or the R-op to construct new variables for the inputs of
+    the inner graph such that there is no interference between the original
+    graph and the newly constructed graph.
+    """
+    if hasattr(x, 'name') and x.name is not None:
+        nw_name = x.name + tag
     else:
-        return x.type()
-
-def safe_to_cpu(x):
-    if isinstance(x, numpy.ndarray):
-        x = tensor.as_tensor_variable(x)
-    if cuda.cuda_available and isinstance(x.type, cuda.CudaNdarrayType):
-        return cuda.basic_ops.host_from_gpu(x)
+        nw_name = None
+    if isinstance(x.type, tensor.Constant):
+        return x.clone()
     else:
-        return x
+        try:
+            x = tensor.as_tensor_variable(x)
+        except TypeError:
+            # This could happend for example for random states, and I really
+            # want to avoid the convoluted logic that checks for cuda
+            # ndarrays
+            pass
+    nw_x = x.type()
+    nw_x.name = nw_name
+    return nw_x
 
 
 def traverse(out, x,x_copy, d):