Added a comment about possible optimization.

theano/tensor/basic.py

@@ -4908,6 +4908,8 @@ class AdvancedSubtensor1(Op):
             o = out[0]
         else:
             o = None
+
+        # I have read that using clip or wrap mode make it a lot faster when the output is provided
         out[0] = x.take(i, axis=0, out=o)
 
     def grad(self, inputs, grads):