Added some doc

doc/library/sandbox/cuda/dnn.txt

@@ -40,6 +40,14 @@ To get an error if Theano can not use cuDNN, use this Theano flag:
     currently does not allow all options supported by cuDNN ops. So it is
     possible that you will need to call them manually.
 
+.. note::
+
+    CUDNN R1 and R2 tell that for the following 2 operations, the
+    reproducibility is not guarantee: `cudnnConvolutionBackwardFilter`
+    and `cudnnConvolutionBackwardData`. Those correspond to the
+    gradient again the weights and the gradient of the
+    convolution. They are used sometimes in the forward when they give
+    speed up.
 
 Functions
 =========

doc/library/typed_list.txt

@@ -8,6 +8,12 @@
 
     This is not in the released version 0.6.0, but will be in the next release (0.7 or 0.6.1).
 
+.. note::
+
+    This work, but isn't well integrated with the test of Theano. If
+    you want speed, you are probably better by padding to a dense
+    tensor.
+
 This is a type that represents a list in Theano. All elements must have
 the same Theano type. Here is an example::
 

doc/tutorial/printing_drawing.txt

@@ -14,6 +14,15 @@ compilation.  These two functions print expression graphs in different ways:
 Theano also provides :func:`pydotprint` that creates a *png* image of the function.
 You can read about them in :ref:`libdoc_printing`.
 
+.. note::
+
+    For all printing of Theano function, sometimes they can be hard to
+    read.  To help on that, you can disable some Theano optimization
+    by using that Theano flag:
+    ``optimizer_excluding=fusion:inplace``. Do not use that during
+    real job execution, as this make the graph slower and use more
+    memory.
+
 Consider again the logistic regression but notice the additional printing instuctions. 
 The following output depicts the pre- and post- compilation graphs.