[dpdk-dev] [PATCH] eventdev: add producer enqueue hint

Van Haaren, Harry harry.van.haaren at intel.com
Tue Jun 27 10:44:34 CEST 2017


> From: Jerin Jacob [mailto:jerin.jacob at caviumnetworks.com]
> Sent: Tuesday, June 27, 2017 9:08 AM
> To: Eads, Gage <gage.eads at intel.com>
> Cc: dev at dpdk.org; Richardson, Bruce <bruce.richardson at intel.com>; Van Haaren, Harry
> <harry.van.haaren at intel.com>; hemant.agrawal at nxp.com; nipun.gupta at nxp.com; Vangati,
> Narender <narender.vangati at intel.com>; Rao, Nikhil <nikhil.rao at intel.com>
> Subject: Re: [dpdk-dev] [PATCH] eventdev: add producer enqueue hint

<snip some patch code>

> > >  void
> > > diff --git a/lib/librte_eventdev/rte_eventdev.h
> > > b/lib/librte_eventdev/rte_eventdev.h
> > > index a248fe90e..1c1a46593 100644
> > > --- a/lib/librte_eventdev/rte_eventdev.h
> > > +++ b/lib/librte_eventdev/rte_eventdev.h
> > > @@ -933,7 +933,15 @@ struct rte_event {
> > >  			 * and is undefined on dequeue.
> > >  			 * @see RTE_EVENT_OP_NEW, (RTE_EVENT_OP_*)
> > >  			 */
> > > -			uint8_t rsvd:4;
> > > +			uint8_t all_op_new:1;
> > > +			/**< Valid only with event enqueue operation - This hint
> > > +			 * indicates that the enqueue request has only the
> > > +			 * events with op == RTE_EVENT_OP_NEW.
> > > +			 * The event producer, typically use this pattern to
> > > +			 * inject the events to eventdev.
> > > +			 * @see RTE_EVENT_OP_NEW
> > > rte_event_enqueue_burst()
> > > +			 */
> > > +			uint8_t rsvd:3;
> > >  			/**< Reserved for future use */
> > >  			uint8_t sched_type:2;
> > >  			/**< Scheduler synchronization type
> > > (RTE_SCHED_TYPE_*)
> > > --
> > > 2.13.1
> >
> > I slightly prefer the parallel enqueue API -- I can see folks making the mistake of
> setting all_op_new without setting the op to RTE_EVENT_OP_NEW, and later adding a
> "forward-only" enqueue API could be interesting for the sw PMD -- but this looks fine to
> me. Curious if others have any thoughts.
> 
> If forward-only parallel enqueue API interesting for the SW PMD then I
> can drop this one and introduce forward-only API. Let me know if others
> have any thoughts?


To make sure I understand correctly, the "parallel API" idea is to add a new function pointer per-PMD, and dedicate it to enqueueing a burst of packets with the same OP? So the end result would be function(s) in the public API like this:

rte_event_enqueue_burst_new(port, new_events, n_events);
rte_event_enqueue_burst_forward(port, new_events, n_events);

Given these are a "specialization" of the generic enqueue_burst() function, the PMD is not obliged to implement them. If they are NULL, the eventdev.c infrastructure can just point the burst_new() and burst_forward() to the generic enqueue without any performance delta?

The cost is some added code in the public header and infrastructure.
The gain is that we don't overload the current API with new behavior. 


Assuming my description of the parallel proposal above is correct, +1 for the parallel function approach. I like APIs that "do what they say on the tin" :)


More information about the dev mailing list