On Thu, Sep 17, 2015 at 09:29:45PM +0100, Zoltan Kiss wrote:
> It's better to have it in the cache as soon as possible. On my test setup it
> meant a 0.7 Mpps increase.
>
> Signed-off-by: Zoltan Kiss <zoltan.k...@linaro.org>
>
> diff --git a/lib/dpif-netdev.c b/lib/dpif-netdev.c
> index 72e5653..3312cc0 100644
> --- a/lib/dpif-netdev.c
> +++ b/lib/dpif-netdev.c
> @@ -3229,11 +3229,6 @@ emc_processing(struct dp_netdev_pmd_thread *pmd,
> struct dp_packet **packets,
> continue;
> }
>
> - if (i != cnt - 1) {
> - /* Prefetch next packet data */
> - OVS_PREFETCH(dp_packet_data(packets[i+1]));
> - }
> -
> miniflow_extract(packets[i], &key.mf);
> key.len = 0; /* Not computed yet. */
> key.hash = dpif_netdev_packet_get_rss_hash(packets[i], &key.mf);
> diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c
> index e4e3d2c..c3c7ec0 100644
> --- a/lib/netdev-dpdk.c
> +++ b/lib/netdev-dpdk.c
> @@ -1015,7 +1015,7 @@ netdev_dpdk_rxq_recv(struct netdev_rxq *rxq_, struct
> dp_packet **packets,
> struct netdev_rxq_dpdk *rx = netdev_rxq_dpdk_cast(rxq_);
> struct netdev *netdev = rx->up.netdev;
> struct netdev_dpdk *dev = netdev_dpdk_cast(netdev);
> - int nb_rx;
> + int nb_rx, i;
>
> /* There is only one tx queue for this core. Do not flush other
> * queues.
> @@ -1033,6 +1033,9 @@ netdev_dpdk_rxq_recv(struct netdev_rxq *rxq_, struct
> dp_packet **packets,
> return EAGAIN;
> }
>
> + for (i = 0; i < nb_rx; i++)
> + OVS_PREFETCH(dp_packet_data(packets[i]));
> +
Please use {} even for single statements to follow the coding style.
Do you know if removing the prefetch completely helps?
fbl
> *c = nb_rx;
>
> return 0;
>
> _______________________________________________
> dev mailing list
> dev@openvswitch.org
> http://openvswitch.org/mailman/listinfo/dev
_______________________________________________
dev mailing list
dev@openvswitch.org
http://openvswitch.org/mailman/listinfo/dev
