Single stream parallelization of generalized LSTM-like RNNs on a GPU