Re: [dpdk-dev] [PATCH v3] net/i40e: add interface to choose latest vector path

["Li, Xiaoyun" <xiaoyun.li@intel.com>]

OK. Will send v4 later. Thanks.

> -----Original Message-----
> From: Zhang, Qi Z
> Sent: Wednesday, September 5, 2018 20:22
> To: Li, Xiaoyun <xiaoyun.li@intel.com>; Xing, Beilei <beilei.xing@intel.com>
> Cc: dev@dpdk.org; Yang, Zhiyong <zhiyong.yang@intel.com>; Richardson,
> Bruce <bruce.richardson@intel.com>; Hunt, David <david.hunt@intel.com>
> Subject: RE: [PATCH v3] net/i40e: add interface to choose latest vector path
> 
> Hi Xiaoyun:
> 
> > -----Original Message-----
> > From: Li, Xiaoyun
> > Sent: Tuesday, September 4, 2018 7:40 PM
> > To: Xing, Beilei <beilei.xing@intel.com>; Zhang, Qi Z
> > <qi.z.zhang@intel.com>
> > Cc: dev@dpdk.org; Yang, Zhiyong <zhiyong.yang@intel.com>; Richardson,
> > Bruce <bruce.richardson@intel.com>; Hunt, David
> > <david.hunt@intel.com>; Li, Xiaoyun <xiaoyun.li@intel.com>
> > Subject: [PATCH v3] net/i40e: add interface to choose latest vector
> > path
> >
> > Right now, vector path is limited to only use on later platform.
> > This patch adds a devarg enable-latest-vec to allow the users to use
> > the latest vector path that the platform supported. Namely, using AVX2
> > vector path on broadwell is possible.
> >
> > Signed-off-by: Xiaoyun Li <xiaoyun.li@intel.com>
> > ---
> > v3:
> >  * Polish the doc and commit log.
> > v2:
> >  * Correct the calling of the wrong function last time.
> >  * Fix seg fault bug.
> >
> >  doc/guides/nics/i40e.rst               |  8 ++++++
> >  doc/guides/rel_notes/release_18_11.rst |  4 +++
> >  drivers/net/i40e/i40e_ethdev.c         | 38
> > ++++++++++++++++++++++++++
> >  drivers/net/i40e/i40e_ethdev.h         |  1 +
> >  drivers/net/i40e/i40e_rxtx.c           | 27 ++++++++++++++++++
> >  5 files changed, 78 insertions(+)
> >
> > diff --git a/doc/guides/nics/i40e.rst b/doc/guides/nics/i40e.rst index
> > 65d87f869..6158e7c34 100644
> > --- a/doc/guides/nics/i40e.rst
> > +++ b/doc/guides/nics/i40e.rst
> > @@ -163,6 +163,14 @@ Runtime Config Options
> >    Currently hot-plugging of representor ports is not supported so all
> required
> >    representors must be specified on the creation of the PF.
> >
> > +- ``Enable latest vector`` (default ``disable``)
> > +
> > +  Vector path was limited to use only on later platform. But users
> > + may want the  latest vector path. For example, VPP users may want to
> > + use
> > + AVX2 vector path on HSW/BDW  because it can get better perf. So
> > + ``devargs`` parameter ``enable-latest-vec``  is introduced, for example::
> > +    -w 84:00.0,enable-latest-vec=1
> 
> 
> How about "use_latest_vec" or "use-lastest-vpmd"?
> 
> > +
> >  Driver compilation and testing
> >  ------------------------------
> >
> > diff --git a/doc/guides/rel_notes/release_18_11.rst
> > b/doc/guides/rel_notes/release_18_11.rst
> > index 3ae6b3f58..f8b0f3189 100644
> > --- a/doc/guides/rel_notes/release_18_11.rst
> > +++ b/doc/guides/rel_notes/release_18_11.rst
> > @@ -54,6 +54,10 @@ New Features
> >       Also, make sure to start the actual text at the margin.
> >       =========================================================
> >
> > +* **Added a devarg to eable the latest vector path.**
> > +  A new devarg ``enable-latest-vec`` was introduced to allow users to
> > +choose
> > +  the latest vector path that the platform supported. For example,
> > +VPP users
> > +  can use AVX2 vector path on BDW/HSW to get better performance.
> >
> >  API Changes
> >  -----------
> > diff --git a/drivers/net/i40e/i40e_ethdev.c
> > b/drivers/net/i40e/i40e_ethdev.c index 85a6a867f..16b5345fb 100644
> > --- a/drivers/net/i40e/i40e_ethdev.c
> > +++ b/drivers/net/i40e/i40e_ethdev.c
> > @@ -12513,6 +12513,44 @@ i40e_config_rss_filter(struct i40e_pf *pf,
> >  	return 0;
> >  }
> >
> > +#define ETH_I40E_ENABLE_LATEST_VEC	"enable-latest-vec"
> 
> This should be defined along with other exist devargs, please check
> ETH_I40E_SUPPORT_MULTI_DRIVER for reference Also it should be
> registered with RTE_PMD_REGISTER_PARAM_STRING.
> 
> > +
> > +bool
> > +i40e_parse_latest_vec(struct rte_eth_dev *dev) {
> > +	static const char *const valid_keys[] = {
> > +		ETH_I40E_ENABLE_LATEST_VEC, NULL};
> > +	int enable_latest_vec;
> > +	struct rte_kvargs *kvlist;
> > +
> > +	if (!dev->device->devargs)
> > +		return 0;
> > +
> > +	kvlist = rte_kvargs_parse(dev->device->devargs->args, valid_keys);
> > +	if (!kvlist)
> > +		return -EINVAL;
> > +
> > +	if (!rte_kvargs_count(kvlist, ETH_I40E_ENABLE_LATEST_VEC))
> > +		return 0;
> > +
> > +	if (rte_kvargs_count(kvlist, ETH_I40E_ENABLE_LATEST_VEC) > 1)
> > +		PMD_DRV_LOG(WARNING, "More than one argument \"%s\"
> and
> > only "
> > +			    "the first one is used !",
> > +			    ETH_I40E_ENABLE_LATEST_VEC);
> > +
> > +	enable_latest_vec = atoi((&kvlist->pairs[0])->value);
> > +
> > +	rte_kvargs_free(kvlist);
> > +
> > +	if (enable_latest_vec != 0 && enable_latest_vec != 1)
> > +		PMD_DRV_LOG(WARNING, "Value should be 0 or 1, set it as
> 1!");
> > +
> > +	if (enable_latest_vec)
> > +		return true;
> > +	else
> > +		return false;
> > +}
> 
> We call rte_kvargs_parse in different place for different parameter which is
> not necessary.
> it's better to merge them into one parse_devargs function at dev_init and
> then all corresponding field of i40e_adapter can be configured at the same
> place.
> Though this is not this patch's scope, but it's better to introduce a field like
> i40e_adapter->use_latest_vec and in i40e_parse_latest_vec, it just assign the
> value which could be used later.
> This will make things easy for future code clean and also it is not necessary to
> call i40e_parse_latest_vec multiple time in set_rx/tx_function.
> 
> > +
> >  RTE_INIT(i40e_init_log)
> >  {
> >  	i40e_logtype_init = rte_log_register("pmd.net.i40e.init");
> > diff --git a/drivers/net/i40e/i40e_ethdev.h
> > b/drivers/net/i40e/i40e_ethdev.h index 3fffe5a55..cdf68cd93 100644
> > --- a/drivers/net/i40e/i40e_ethdev.h
> > +++ b/drivers/net/i40e/i40e_ethdev.h
> > @@ -1243,6 +1243,7 @@ int i40e_config_rss_filter(struct i40e_pf *pf,
> >  		struct i40e_rte_flow_rss_conf *conf, bool add);  int
> > i40e_vf_representor_init(struct rte_eth_dev *ethdev, void
> > *init_params);  int i40e_vf_representor_uninit(struct rte_eth_dev
> > *ethdev);
> > +bool i40e_parse_latest_vec(struct rte_eth_dev *dev);
> >
> >  #define I40E_DEV_TO_PCI(eth_dev) \
> >  	RTE_DEV_TO_PCI((eth_dev)->device)
> > diff --git a/drivers/net/i40e/i40e_rxtx.c
> > b/drivers/net/i40e/i40e_rxtx.c index
> > 2a28ee348..75f8ec284 100644
> > --- a/drivers/net/i40e/i40e_rxtx.c
> > +++ b/drivers/net/i40e/i40e_rxtx.c
> > @@ -2960,6 +2960,15 @@ i40e_set_rx_function(struct rte_eth_dev *dev)
> >  			if
> (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F))
> >  				dev->rx_pkt_burst =
> >  					i40e_recv_scattered_pkts_vec_avx2;
> > +			/*
> > +			 * Give users chance to use the latest vector path
> > +			 * that the platform supported.
> > +			 */
> > +			if (i40e_parse_latest_vec(dev)) {
> > +				if
> (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2))
> > +					dev->rx_pkt_burst =
> > +					i40e_recv_scattered_pkts_vec_avx2;
> > +			}
> 
> 
> How about:
> 
> 			if (adapter->use_latest_vec)
> 				dev->pkt_burst = get_latest_rx_vec();
> 			else
> 				dev->pkt_burst = get_recommended_rx_vec();
> 
> this make code easy to understand and in future, if we have avx512 vpmd,
> it's easy for us to modify the code.
> 
> 
> >  #endif
> >  		} else {
> >  			PMD_INIT_LOG(DEBUG, "Using a Scattered with bulk
> "
> > @@ -2989,6 +2998,15 @@ i40e_set_rx_function(struct rte_eth_dev *dev)
> >  		 */
> >  		if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F))
> >  			dev->rx_pkt_burst = i40e_recv_pkts_vec_avx2;
> > +		/*
> > +		 * Give users chance to use the latest vector path
> > +		 * that the platform supported.
> > +		 */
> > +		if (i40e_parse_latest_vec(dev)) {
> > +			if (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2))
> > +				dev->rx_pkt_burst =
> > +					i40e_recv_pkts_vec_avx2;
> > +		}
> 
> Same as above
> >  #endif
> >  	} else if (ad->rx_bulk_alloc_allowed) {
> >  		PMD_INIT_LOG(DEBUG, "Rx Burst Bulk Alloc Preconditions
> are "
> > @@ -3083,6 +3101,15 @@ i40e_set_tx_function(struct rte_eth_dev *dev)
> >  			 */
> >  			if
> (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX512F))
> >  				dev->tx_pkt_burst =
> i40e_xmit_pkts_vec_avx2;
> > +			/*
> > +			 * Give users chance to use the latest vector path
> > +			 * that the platform supported.
> > +			 */
> > +			if (i40e_parse_latest_vec(dev)) {
> > +				if
> (rte_cpu_get_flag_enabled(RTE_CPUFLAG_AVX2))
> > +					dev->tx_pkt_burst =
> > +					i40e_xmit_pkts_vec_avx2;
> > +			}
> 
> Same as above
> 
> >  #endif
> >  		} else {
> >  			PMD_INIT_LOG(DEBUG, "Simple tx finally be used.");
> > --
> > 2.17.1
> 
> Regards
> Qi