Balancing locality and concurrency: solving sparse triangular systems on GPUs