Add a theano flag for single stream mode.

theano/configdefaults.py

@@ -242,6 +242,14 @@ AddConfigVar('gpuarray.preallocate',
              FloatParam(0),
              in_c_key=False)
 
+AddConfigVar('gpuarray.single_stream',
+             """Switch between single stream mode or multi-stream
+             mode.  If your computation can't benefit from multiple
+             streams, single-stream is usually faster (by about 10%)
+             """,
+             BoolParam(True),
+             in_c_key=False)
+
 
 def safe_no_dnn_workmem(workmem):
     """

theano/gpuarray/__init__.py

@@ -62,7 +62,8 @@ def init_dev(dev, name=None):
     global pygpu_activated
     if dev not in init_dev.devmap:
         ctx = pygpu.init(dev,
-                         disable_alloc_cache=config.gpuarray.preallocate < 0)
+                         disable_alloc_cache=config.gpuarray.preallocate < 0,
+                         single_stream=config.gpuarray.single_stream)
         init_dev.devmap[dev] = ctx
         if config.gpuarray.preallocate > 0:
             MB = (1024 * 1024)