corrected accidental paste

theano/sandbox/cuda/basic_ops.py

@@ -2163,7 +2163,7 @@ class GpuCAReduce(GpuOp):
                                              "A[i0 * sA0 + blockIdx.x * sA1 + i2 * sA2 + i3 * sA3]",
                                              {})
             reduce_init = self._assign_init("A[blockIdx.x * sA1]")
-            print >> sio, """join
+            print >> sio, """
             static __global__ void kernel_reduce_1011_%(nodename)s(
                     const unsigned int d0,
                     const unsigned int d1,