[dpdk-dev] [PATCH v1 1/2] net/virtio: restrict pointer aliasing for NEON vpmd

[Joyce Kong]

> -----Original Message-----
> From: Phil Yang <Phil.Yang at arm.com>
> Sent: Tuesday, June 23, 2020 5:06 PM
> To: Maxime Coquelin <maxime.coquelin at redhat.com>; Joyce Kong
> <Joyce.Kong at arm.com>; jerinj at marvell.com; zhihong.wang at intel.com;
> xiaolong.ye at intel.com; Honnappa Nagarahalli
> <Honnappa.Nagarahalli at arm.com>; Ruifeng Wang
> <Ruifeng.Wang at arm.com>
> Cc: dev at dpdk.org
> Subject: RE: [PATCH v1 1/2] net/virtio: restrict pointer aliasing for NEON
> vpmd
>
> > -----Original Message-----
> > From: Maxime Coquelin <maxime.coquelin at redhat.com>
> > Sent: Tuesday, June 23, 2020 4:48 PM
> > To: Joyce Kong <Joyce.Kong at arm.com>; jerinj at marvell.com;
> > zhihong.wang at intel.com; xiaolong.ye at intel.com; Honnappa Nagarahalli
> > <Honnappa.Nagarahalli at arm.com>; Phil Yang <Phil.Yang at arm.com>;
> Ruifeng
> > Wang <Ruifeng.Wang at arm.com>
> > Cc: dev at dpdk.org
> > Subject: Re: [PATCH v1 1/2] net/virtio: restrict pointer aliasing for
> > NEON vpmd
> >
> >
> >
> > On 6/11/20 5:32 AM, Joyce Kong wrote:
> > > Restrict pointer aliasing to allow the compiler to vectorize loops
> > > more aggressively.
> > >
> > > With this patch, a 9.6% improvement is observed in throughput for
> > > the virtio-net PVP case, and a 2.4% perf improvement in throughput
> > > for the virtio-user PVP case. All performance data are measured
> > > under the 0.001% acceptable packet loss with 2 cores on the vhost
> > > side.
> > >
> > > Signed-off-by: Joyce Kong <joyce.kong at arm.com>
> > > Reviewed-by: Phil Yang <phil.yang at arm.com>
> > > ---
> > >  drivers/net/virtio/virtio_rxtx_simple_neon.c | 4 ++--
> > >  1 file changed, 2 insertions(+), 2 deletions(-)
> >
> > Very nice, we should consider doing the same on other platforms.
> >
> > > diff --git a/drivers/net/virtio/virtio_rxtx_simple_neon.c
> > b/drivers/net/virtio/virtio_rxtx_simple_neon.c
> > > index 363e2b330..c08dd51fb 100644
> > > --- a/drivers/net/virtio/virtio_rxtx_simple_neon.c
> > > +++ b/drivers/net/virtio/virtio_rxtx_simple_neon.c
> > > @@ -36,8 +36,8 @@
> > >   * - nb_pkts < RTE_VIRTIO_DESC_PER_LOOP, just return no packet
> > >   */
> > >  uint16_t
> > > -virtio_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
> > > -uint16_t nb_pkts)
> > > +virtio_recv_pkts_vec(void *rx_queue, struct rte_mbuf **__restrict
> > > +rx_pkts, uint16_t nb_pkts)
> >
> > Is __restrict supported by all the compilers?
> > Wouldn't it be better to introduce a wrapper?
>
> +1 for this.
> In my understanding, the __restrict keyword is recognized in C at all language
> levels.
> However, the restrict keyword is recognized in C under compilation with c99.
> DPDK uses the restrict qualifier a lot, which might have some issues with
> some old compilers.
> So the wrapper will be useful.
>
> Thanks,
> Phil
>

Thanks for the suggestion, I shall introduce a wrapper to support all the compilers
in next version.

> >
> > >  {
> > >  struct virtnet_rx *rxvq = rx_queue;  struct virtqueue *vq =
> > > rxvq->vq;
> > >
>

IMPORTANT NOTICE: The contents of this email and any attachments are confidential and may also be privileged. If you are not the intended recipient, please notify the sender immediately and do not disclose the contents to any other person, use it for any purpose, or store or copy the information in any medium. Thank you.