Document the new flops interface.

doc/extending/op.txt

@@ -330,6 +330,16 @@ following methods:
    shape without computing the output itself, potentially sparing you
    a costly recomputation.
 
+.. function:: flops(inputs, outputs)
+
+   Optional.
+
+   It is only used to have more information printed by the memory
+   profiler.  It make it print the mega flops and giga flops per
+   second for each apply node. It take as inputs two list: one for the
+   inputs and one for the outputs. They contain one tuple with the
+   shape of the corresponding inputs/outputs.
+
 .. function:: make_thunk(node, storage_map, compute_map, no_recycling)
 
    TODO

doc/tutorial/extending_theano.txt

@@ -86,7 +86,10 @@ Op Contract
         def R_op(self, inputs, eval_points):
             pass
 
-        def infer_shape(node, (i0_shapes, ...))
+        def infer_shape(node, (i0_shapes, ...)):
+            pass
+
+        def flops(self, inputs, outputs):
             pass
 
 .. ../extending/op.txt
@@ -116,6 +119,11 @@ The :func:`infer_shape` method allows to infer the shape of some variable, somew
 middle of the computational graph without actually computing the outputs (when possible).
 This could be helpful if one only needs the shape of the output instead of the actual outputs.
 
+The :func:`flops` method allows to have the number of mega flops and
+giga flops per second printed by the memory profiler. It take as
+inputs two list: one for the inputs and one for the outputs. They
+contain one tuple with the shape of the corresponding inputs/outputs.
+
 The :func:`grad` method is required if you want to differentiate some cost whose expression
 includes your op.