[dpdk-dev] [PATCH] ether: add support for vtune task tracing

[Ananyev, Konstantin]

> -----Original Message-----
> From: Jerin Jacob [mailto:jerin.jacob at caviumnetworks.com]
> Sent: Friday, June 30, 2017 4:31 AM
> To: Kurakin, Ilia <ilia.kurakin at intel.com>
> Cc: dev at dpdk.org; Ananyev, Konstantin <konstantin.ananyev at intel.com>; Wiles, Keith <keith.wiles at intel.com>; Galanov, Dmitry
> <dmitry.galanov at intel.com>
> Subject: Re: [dpdk-dev] [PATCH] ether: add support for vtune task tracing
> 
> -----Original Message-----
> > Date: Tue, 27 Jun 2017 16:16:01 +0300
> > From: ilia.kurakin at intel.com
> > To: dev at dpdk.org
> > CC: konstantin.ananyev at intel.com, keith.wiles at intel.com,
> >  dmitry.galanov at intel.com, Ilia Kurakin <ilia.kurakin at intel.com>
> > Subject: [dpdk-dev] [PATCH] ether: add support for vtune task tracing
> > X-Mailer: git-send-email 2.7.4
> >
> > From: Ilia Kurakin <ilia.kurakin at intel.com>
> >
> > diff --git a/lib/librte_ether/rte_ethdev.h b/lib/librte_ether/rte_ethdev.h
> > index f6e6c74..ee7cc42 100644
> > --- a/lib/librte_ether/rte_ethdev.h
> > +++ b/lib/librte_ether/rte_ethdev.h
> > @@ -186,6 +186,10 @@ extern "C" {
> >  #include "rte_eth_ctrl.h"
> >  #include "rte_dev_info.h"
> >
> > +#ifdef RTE_ETHDEV_TRACE_WASTED_RX_ITERATIONS
> > +#include "rte_eth_itt.h"
> > +#endif
> > +
> >  struct rte_mbuf;
> >
> >  /**
> > @@ -2710,6 +2714,25 @@ rte_eth_rx_burst(uint8_t port_id, uint16_t queue_id,
> >  	int16_t nb_rx = (*dev->rx_pkt_burst)(dev->data->rx_queues[queue_id],
> >  			rx_pkts, nb_pkts);
> >
> > +#ifdef RTE_ETHDEV_TRACE_WASTED_RX_ITERATIONS
> 
> If we give the generic name like above then the solution should be generic
> and it should not be tightly coupled with itt.
> 
> Different architectures may have different profiling tools like x86 has
> itt or it is possible to have generic perf based plugin in future.
> 
> Considering the above points, IMO, we can not add code in
> rte_eth_rx_burst() for each profiling variants. 

+ 1

>I think, this code can
> go in lib/librte_ether/rte_ethdev_profile.c and based on specific
> conditional compilation flag(something like RTE_ETHDEV_TRACE_ITT_WASTED_RX_ITERATIONS)
> it can register a rx callback.
> or any other scheme without directly modifying rte_eth_rx_burst() for
> itt based profiling.

Or might be even better to have some wrapper function/macro on top of rx_burst():
rx_burst_profle(...) or so.
Then user can either replace rx_burst() with rx_burst_profile() by conditional compilation,
or just call rx_burst_profile() directly.

Konstantin

> 
> 
> > +	/* See rte_eth_itt.h to find comments on code below. */
> > +	if (unlikely(nb_rx == 0)) {
> > +		if (!itt_aux_data[port_id].queue_is_wasting_iters[queue_id]) {
> > +			__itt_task_begin(
> > +				itt_aux_data[port_id].wasted_iter_domains[queue_id],
> > +				__itt_null, __itt_null,
> > +				itt_aux_data[port_id].wasted_iter_handles[queue_id]);
> > +			itt_aux_data[port_id].queue_is_wasting_iters[queue_id] = 1;
> > +		}
> > +	} else {
> > +		if (unlikely(itt_aux_data[port_id].queue_is_wasting_iters[queue_id])) {
> > +			__itt_task_end(
> > +				itt_aux_data[port_id].wasted_iter_domains[queue_id]);
> > +			itt_aux_data[port_id].queue_is_wasting_iters[queue_id] = 0;
> > +		}
> > +	}
> > +#endif
> > +
> >  #ifdef RTE_ETHDEV_RXTX_CALLBACKS
> >  	struct rte_eth_rxtx_callback *cb = dev->post_rx_burst_cbs[queue_id];
> >
> > --
> > 2.7.4
> >
> >
> > --------------------------------------------------------------------
> > Joint Stock Company Intel A/O
> > Registered legal address: Krylatsky Hills Business Park,
> > 17 Krylatskaya Str., Bldg 4, Moscow 121614,
> > Russian Federation
> >
> > This e-mail and any attachments may contain confidential material for
> > the sole use of the intended recipient(s). Any review or distribution
> > by others is strictly prohibited. If you are not the intended
> > recipient, please contact the sender and delete all copies.
> >