On 11/21/2014 12:18 PM, Alexandru Badicioiu wrote:
Hi, the scheduling tests in odp_example were discussed some time ago and there was an agreement, at least for FSL and TI platforms, that fair scheduling assumed by the following loop: for (i = 0; i < QUEUE_ROUNDS; i++) { buf = odp_schedule_one(&queue, ODP_SCHED_WAIT);if (odp_queue_enq(queue, buf)) { ODP_ERR(" [%i] Queue enqueue failed.\n", thr); return -1; } } for an ATOMIC queue doesn't make sense as the behavior of an ATOMIC hardware queue is to be scheduled to the same core as long as there are packets in the queue and the core has room to dequeue them (other platforms please confirm or infirm). On my specific platform I can force this with a particular HW configuration, but this configuration effectively disables the use of POLL queues. I think we need scheduling tests for the most general case (SYNC_NONE and no assumption about how many buffers are scheduled to a particular core).
I agree. Tests should not assume fair scheduling. _______________________________________________ lng-odp mailing list lng-odp@lists.linaro.org http://lists.linaro.org/mailman/listinfo/lng-odp
