Optimization to remove upcast in local_cast_cast

theano/tensor/opt.py

@@ -2204,7 +2204,7 @@ def local_cast_cast(node):
 
     when those contrain:
     dtype1 == dtype2
-    TODO: the base dtype is the same (int, uint, float, complex)
+    OR the base dtype is the same (int, uint, float, complex)
           and the first cast cause an upcast.
 
     """
@@ -2216,10 +2216,51 @@ def local_cast_cast(node):
             not isinstance(x.owner.op, T.Elemwise) or
             not isinstance(x.owner.op.scalar_op, scalar.Cast)):
         return
-    if node.op.scalar_op.o_type == x.owner.op.scalar_op.o_type:
+
+    type1 = x.owner.op.scalar_op.o_type
+    type2 = node.op.scalar_op.o_type
+    base = x.owner.inputs[0]
+
+    if type1 == type2:
         # We don't need to copy over any stack traces here
         return [x]
 
+    if(upcast(base.dtype, type1.dtype)):
+        # Checking for further redundancy. Eg: int8 -> int32 -> int8
+        if(type2.dtype == base.dtype):
+            return x.owner.inputs
+        else:
+            # Apply the second cast only
+            v = node.op(base)
+            return [v]
+
+def upcast(type1, type2):
+    '''
+    Given two data types (as strings), check if converting to 
+    type2 from type1 constitutes an upcast.
+    '''
+    upcast_pairs = (
+        ('int8','int16'),('int8','int32'),('int8','int64'),('int16','int32'),('int16','int64'),('int32','int64'),
+        ('uint8','uint16'),('uint8','uint32'),('uint8','uint64'),('uint16','uint32'),('uint16','uint64'),('uint32','uint64'),
+        ('float16','float32'),('float16','float32'),('float16','float64'),('float32','float64'),
+        ('complex64','complex128'),
+
+        ('uint8','int16'),('uint8','int32'),('uint8','int64'),('uint16','int32'),('uint16','int64'),('uint32','int64'),
+        ('int8','float16'),('int8','float32'),('int8','float64'),('int16','float32'),('int16','float64'),('int32','float64'),
+        ('uint8','float16'),('uint8','float32'),('uint8','float64'),('uint16','float32'),('uint16','float64'),('uint32','float64'),
+
+        ('int8','complex64'),('int16','complex64'),
+        ('uint8','complex64'),('uint16','complex64'),
+        ('float32','complex64'),
+        ('int8','complex128'),('int16','complex128'),('int32','complex128'),
+        ('uint8','complex128'),('uint16','complex128'),('uint32','complex128'),
+        ('float32','complex128'),('float64','complex128')
+    )
+
+    for pair in upcast_pairs:
+        if(type1 == pair[0] and type2 == pair[1]):
+            return True
+    return False
 
 @register_canonicalize
 @register_specialize