Add cuDNN version information.

theano/sandbox/cuda/cudnn_helper.h

@@ -3,7 +3,7 @@
 
 #include <cudnn.h>
 
-static inline const char *cudnnGetErrorString(cudnnStatus_t err) {
+inline const char *cudnnGetErrorString(cudnnStatus_t err) {
   switch (err) {
   case CUDNN_STATUS_SUCCESS:
     return "The operation completed successfully.";
@@ -28,4 +28,14 @@ static inline const char *cudnnGetErrorString(cudnnStatus_t err) {
   }
 }
 
+static inline const int cudnnVersionMacro(){
+#ifdef CUDNN_VERSION
+  return CUDNN_VERSION;
+#else
+//CUDNN_VERSION undefined, you probably use cuDNN R1 version
+  return -1;
+#endif
+
+}
+
 #endif

theano/sandbox/cuda/dnn.py

@@ -3,7 +3,7 @@ import os
 import theano
 from theano import Apply, gof, tensor
 from theano.gof import Optimizer, local_optimizer
-from theano.gof.type import CDataType
+from theano.gof.type import CDataType, Generic
 from theano.compat import PY3
 from theano.tensor.nnet import SoftmaxGrad
 from theano.sandbox.cuda.type import CudaNdarrayType
@@ -121,6 +121,39 @@ if ((err = cudnnCreate(&_handle)) != CUDNN_STATUS_SUCCESS) {
 }""" % (error_out,)]
 
 
+class DnnVersion(DnnBase):
+    def c_compiler(self):
+        return NVCC_compiler
+
+    def make_node(self):
+        return Apply(self, [], [Generic()()])
+
+    def c_code(self, node, name, inputs, outputs, sub):
+        o = outputs[0]
+        return """
+        %(o)s = PyInt_FromLong(cudnnVersionMacro());
+        """ % locals()
+
+    def do_constant_folding(self, node):
+        # Needed as we do not want to cache this information.
+        return False
+
+    def c_code_cache_version(self):
+        # Not needed, but make it clear that we do not want to cache this.
+        return None
+
+
+def version():
+    """
+    return the current cuDNN version we compile with.
+
+    This only check the header version, the the library we link with.
+    """
+    f = theano.function([], DnnVersion()(),
+                        theano.Mode(optimizer=None))
+    return f()
+
+
 class GpuDnnConvDesc(GpuOp):
     """This Op builds a convolution descriptor for use in the other
     convolution operations.

theano/sandbox/cuda/tests/test_dnn.py

@@ -192,3 +192,9 @@ def test_dnn_tag():
         assert cuda.dnn.dnn_available()
         assert any([isinstance(n.op, cuda.dnn.GpuDnnPool)
                     for n in f.maker.fgraph.toposort()])
+
+
+def test_version():
+    if not cuda.dnn.dnn_available():
+        raise SkipTest(cuda.dnn.dnn_available.msg)
+    assert isinstance(cuda.dnn.version(), int)