Hi Xiaoyun:
> -----Original Message-----
> From: Li, Xiaoyun
> Sent: Tuesday, September 4, 2018 7:40 PM
> To: Xing, Beilei <beilei.x...@intel.com>; Zhang, Qi Z <qi.z.zh...@intel.com>
> Cc: dev@dpdk.org; Yang, Zhiyong <zhiyong.y...@intel.com>; Richardson,
> Bruce <bruce.richard...@intel.com>; Hunt, David <david.h...@intel.com>; Li,
> Xiaoyun <xiaoyun...@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...@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
