One more step when triing to reuse the output of GpuConvOp.

blas.py

@@ -184,6 +184,7 @@ class GpuConv(Op):
         print "out", out
         out[0] = cuda_ndarray.conv(img, kern, 
                 mode=self.border_mode,
+                out=out[0],
                 subsample=self.subsample,
                 logical_img_shape=self.logical_img_hw,
                 logical_kern_shape=self.logical_kern_hw,