Make some reduction print there acc_dtype.

theano/tensor/elemwise.py

@@ -1836,6 +1836,20 @@ class CAReduceDtype(CAReduce):
         assert op.acc_dtype is not None
         return CAReduce.make_node(op, input)
 
+    def __str__(self):
+        name = self.__class__.__name__
+        if self.__class__.__name__ == "CAReduceDtype":
+            name = "ReduceDtype{%s}" % self.scalar_op,
+        axis = ""
+        if self.axis is not None:
+            axis = ", ".join(str(x) for x in self.axis)
+            axis = "axis=[%s], " % axis
+        return "%s{%sacc_dtype=%s}" % (
+            name,
+            axis,
+            str(self.acc_dtype)
+        )
+
 
 class Sum(CAReduceDtype):
     """
@@ -1908,12 +1922,6 @@ class Sum(CAReduceDtype):
             return [None]
         return self(*eval_points, **dict(return_list=True))
 
-    def __str__(self):
-        if self.axis is None:
-            return "Sum"
-        else:
-            return "Sum{%s}" % ", ".join(map(str, self.axis))
-
 
 class Prod(CAReduceDtype):
     """
@@ -2067,12 +2075,6 @@ class Prod(CAReduceDtype):
 
             return [final_grad]
 
-    def __str__(self):
-        if self.axis is None:
-            return "Prod"
-        else:
-            return "Prod{%s}" % ", ".join(map(str, self.axis))
-
     def c_code_cache_version(self):
         return (1,)
 
@@ -2113,9 +2115,3 @@ class ProdWithoutZeros(CAReduceDtype):
     def __init__(self, axis=None, dtype=None, acc_dtype=None):
         CAReduceDtype.__init__(self, mul_without_zeros, axis=axis,
                                dtype=dtype, acc_dtype=acc_dtype)
-
-    def __str__(self):
-        if self.axis is None:
-            return "ProdWithoutZeros"
-        else:
-            return "ProdWithoutZeros{%s}" % ", ".join(map(str, self.axis))