Merge pull request #1273 from nouiz/mixed

theano/sandbox/cuda/blas.py

@@ -582,6 +582,8 @@ class GpuConv(GpuOp):
         self.__dict__.update(d)
         if not hasattr(self, "imshp"):
             self.imshp = None
+        if not hasattr(self, "max_threads_dim0"):
+            self.max_threads_dim0 = None
 
     def __hash__(self):
         # don't use hash(self.version) as hash(-1)==-2 and
@@ -639,11 +641,6 @@ class GpuConv(GpuOp):
         return super(GpuConv, node_.op).make_thunk(node_, storage_map,
                                                    compute_map, no_recycling)
 
-    def __setstate__(self, d):
-        self.__dict__.update(d)
-        if not hasattr(self, "max_threads_dim0"):
-            self.max_threads_dim0 = None
-
     def c_compile_args(self):
         nb = 0
         if self.kshp is not None:
@@ -760,7 +757,8 @@ class GpuDownsampleFactorMax(GpuOp):
         int dims[4], xdim2, xdim3;
         if (%(x)s->nd != 4)
         {
-            PyErr_SetString(PyExc_ValueError, "rank error");
+            PyErr_SetString(PyExc_ValueError,
+                            "GpuDownsampleFactorMax: rank error");
             %(fail)s;
         }
         xdim2 = CudaNdarray_HOST_DIMS(%(x)s)[2];
@@ -796,6 +794,7 @@ class GpuDownsampleFactorMax(GpuOp):
                 Py_XDECREF(%(z)s);
                 %(z)s = NULL;
                 PyErr_SetString(PyExc_ValueError,
+                                "GpuDownsampleFactorMax:"
                                 "Was not able to allocate output!");
                 %(fail)s;
             }