Hi,

The concern I had with the above testing scenario is that IMO this scenario
does not match perfectly with an application usecase. The testing for
ORDERED queue in my opinion could be,

1. Dequeue work W1-W10 from ORDERED queue Q1
2. Enqueue work W1-W5 to ORDERED queue Q2
3. Enqueue work W6-W10 to ORDERED queue Q3

Now if Q2 and Q3 have same priority, the scheduler on calling
odp_schedule() should dispatch work W1 and W6 to different cores as they
both now from different queues with same priority.

Yes. I agree we need to define the semantics of ORDERED and ATOMIC queue
more clearly.

Regards,
Bala

On 24 November 2014 14:40, Taras Kondratiuk <[email protected]>
wrote:

> On 11/23/2014 03:55 AM, Bill Fischofer wrote:
>
>> The semantics of ordered queues still need to be fully (and rigorously)
>> defined.  Otherwise it's impossible to ensure that different
>> implementations will yield the same results.  Once we get past the
>> "sunny day" tests, its the job of the test writer to be devious in
>> trying to trick implementations into doing something that the spec says
>> they shouldn't do.  So Ciprian's scenario is a good one.
>>
>
> It is not clear how to interpret results of this test scenario, because
> exact behavior is not specified.
>
_______________________________________________
lng-odp mailing list
[email protected]
http://lists.linaro.org/mailman/listinfo/lng-odp

Reply via email to