Programming the memory hierarchy revisited: supporting irregular parallelism in sequoia