[dpdk-dev] [PATCH v2] baseband/turbo_sw: offload cost measurement test

Mokhtar, Amr amr.mokhtar at intel.com
Tue Apr 24 21:09:09 CEST 2018



> -----Original Message-----
> From: De Lara Guarch, Pablo
> Sent: Tuesday 24 April 2018 18:45
> To: Chalupnik, KamilX <kamilx.chalupnik at intel.com>; dev at dpdk.org
> Cc: Mokhtar, Amr <amr.mokhtar at intel.com>; Chalupnik, KamilX
> <kamilx.chalupnik at intel.com>
> Subject: RE: [dpdk-dev] [PATCH v2] baseband/turbo_sw: offload cost
> measurement test
> 
> 
> 
> > -----Original Message-----
> > From: dev [mailto:dev-bounces at dpdk.org] On Behalf Of KamilX Chalupnik
> > Sent: Tuesday, April 17, 2018 3:27 PM
> > To: dev at dpdk.org
> > Cc: Mokhtar, Amr <amr.mokhtar at intel.com>; Chalupnik, KamilX
> > <kamilx.chalupnik at intel.com>
> > Subject: [dpdk-dev] [PATCH v2] baseband/turbo_sw: offload cost
> measurement
> > test
> >
> > New test created to measure offload cost.
> > Changes were introduced in API, turbo software driver and test
> application.
> >
> 
> Shouldn't this be generic to bbdev/baseband drivers in general and not just
> turbo?
> 

Yes, it is generic.
But the only driver we have right now is the turbo_sw driver. Future drivers
will have a similar support.

> > Signed-off-by: KamilX Chalupnik <kamilx.chalupnik at intel.com>
> 
> ...
> 
> > --- a/lib/librte_bbdev/rte_bbdev.h
> > +++ b/lib/librte_bbdev/rte_bbdev.h
> > @@ -239,6 +239,10 @@ struct rte_bbdev_stats {
> >  	uint64_t enqueue_err_count;
> >  	/** Total error count on operations dequeued */
> >  	uint64_t dequeue_err_count;
> > +#ifdef RTE_TEST_BBDEV
> > +	/** It stores turbo decoder/encoder working time. */
> > +	uint64_t turbo_perf_time;
> > +#endif
> 
> I don't think it is a good idea to use RTE_TEST_BBDEV here.
> This macro is used to enable/disable the compilation of the bbdev test app,
> so I think it should not be used in the API/PMDs.
> 
> Also, this looks too specific for the Turbo SW PMD to be exposed
> as a generic statistic.

Well, it should be generic. Probably 'turbo' is a bad comment and name.
It's intention is to feedback execution time/cycles back to test app in order to
collect the offload cost of the bbdev driver.
What is meant by the offload cost is the cycles consumed from the moment
of enqueue till the moment the request is put on the acceleration engine
inbound sw ring (software) or MMIO operation (hardware).


> 
> >  };
> >
> >  /**
> > --
> > 2.5.5
> >



More information about the dev mailing list