cache OpFromGrad sub grad computation. This remove duplicate op creation.

theano/compile/builders.py

@@ -84,8 +84,13 @@ class OpFromGraph(gof.Op):
         # compute the right numerical value for the gradients but
         # could fail to raise the disconnected inputs error in some
         # cases.
-        gs = G.grad(cost=None, known_grads=dict(zip(self.outputs, output_grads)),
+        if hasattr(self, "grad_ops"):
+            grad_ops = self.grad_ops
+        else:
+            gs = G.grad(cost=None,
+                        known_grads=dict(zip(self.outputs, output_grads)),
                         wrt=self.inputs, disconnected_inputs='ignore')
+
             grad_ops = []
             for g in gs:
                 if g is None:
@@ -96,6 +101,7 @@ class OpFromGraph(gof.Op):
                     grad_ops.append(OpFromGraph(self.inputs + output_grads,
                                                 [g],
                                                 on_unused_input='ignore'))
+            self.grad_ops = grad_ops
 
         return [go(*(inputs + output_grads)) for go in grad_ops]