Update doc

doc/library/tensor/nnet/conv.txt

@@ -25,26 +25,30 @@
 .. note::
 
     As October 20, 2014, the default GPU image convolution
-    changed. Now, if `cuDNN <https://developer.nvidia.com/cuDNN>`_ is
-    available and the GPU selected is supported by it. This give
-    faster GPU convolution without using more memory then the legacy
-    convolution.
+    changed. Here is the algo:
 
-    - If can use cuDNN, use it.
+    - If we can use `cuDNN <https://developer.nvidia.com/cuDNN>`_, use it.
     - If not, use gemm version (slower then cuDNN, use more memory).
 
-    - If the user don't want the extra memory of the gemm version,
-      they can enable the legacy code that is even slower, but don't
-      use extra memory.
+    If the user don't want the extra memory of the gemm version, they
+    can enable the legacy code that is even slower, but don't use
+    extra memory. For this, use the Theano flag
+    ``optimizer_excluding=conv_gemm``.
+
+    There is no reason to use the legacy code or the gemm version if
+    cuDNN is available.
+
+    2 other options:
 
     - There is also the fft version that is the fastest in some cases,
       but use even more memory. It don't support striding to remove
-      computation and have some shape restriction.
+      computation and have some shapes restriction.
 
     - There is also the cuda_convnet convolution in Pylearn2. It use a
       different memory layout, have shapes restriction, but don't use
       extra memory and is faster then the legacy convolution.
 
+
 TODO: Give examples on how to use these things! They are pretty complicated.
 
 - Convolution operators implemented:

doc/tutorial/multi_cores.txt

+.. _tut_multi_cores:
+
 =============================
 Multi cores support in Theano
 =============================