Don't zeros output before reusing it.

theano/tensor/type.py

@@ -636,10 +636,10 @@ theano.compile.register_shape_i_c_code(
         TensorType,
         """
         if(!%(oname)s)
-            %(oname)s=(PyArrayObject*)PyArray_ZEROS(0, NULL, NPY_INT64, 0);
+            %(oname)s=(PyArrayObject*)PyArray_EMPTY(0, NULL, NPY_INT64, 0);
         ((npy_int64*)PyArray_DATA(%(oname)s))[0]=PyArray_DIMS(%(iname)s)[%(i)s];
         """,
-        version=1)
+        version=2)
 
 # Register TensorType C code for DeepCopyOp
 theano.compile.register_deep_copy_op_c_code(