Merge pull request #3253 from carriepl/cudnn_fixes

doc/library/sandbox/cuda/dnn.txt

@@ -53,7 +53,7 @@ To get an error if Theano can not use cuDNN, use this Theano flag:
    * ``small`` (default) : use a convolution implementation with small memory
      usage
    * ``none`` : use a slower implementation with minimal memory usage
-   * ``large`` : use a faster implementation with large memory usage
+   * ``large`` : use a sometimes faster implementation with large memory usage
    * ``fft`` : use the Fast Fourrier Transform implementation of convolution
      (very high memory usage)
    * ``guess_once`` : the first time a convolution is executed, the
@@ -91,6 +91,10 @@ To get an error if Theano can not use cuDNN, use this Theano flag:
      implementation selected every time the shapes of the inputs and kernels
      don't match the shapes from the last execution.
 
+   ``guess_*`` and ``time_*`` flag values take into account the amount of
+   available memory when selecting an implementation. This means that slower
+   implementations might be selected if not enough memory is available for the
+   faster implementations.
 
 .. note::
 

theano/sandbox/cuda/dnn.py

@@ -81,6 +81,13 @@ if ((err = cudnnCreate(&_handle)) != CUDNN_STATUS_SUCCESS) {
                                          " from one version, but we link with"
                                          " a different version %s" % str(v))
                     raise RuntimeError(dnn_available.msg)
+                if version() == -1:
+                    dnn_available.avail = False
+                    dnn_available.msg = (
+                        "CuDNN v1 detected. This version is no longer "
+                        "supported by Theano. Update your CuDNN installation "
+                        "to a more recent version")
+                    raise RuntimeError(dnn_available.msg)
                 if version() == (20, 20):
                     dnn_available.avail = False
                     dnn_available.msg = (