[v3,3/3] lib/vhost: restrict pointer aliasing for packed vpmd

  Restrict pointer aliasing to allow the compiler to vectorize loop
more aggressively.

With this patch, a 9.6% improvement is observed in throughput for
the packed virtio-net PVP case, and a 2.8% improvement in throughput
for the packed virtio-user PVP case. All performance data are measured
on ThunderX-2 platform under 0.001% acceptable packet loss with 1 core
on both vhost and virtio side.

Signed-off-by: Joyce Kong <joyce.kong@arm.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
---
 drivers/net/virtio/virtio_rxtx_simple_neon.c |  5 +++--
 lib/librte_vhost/virtio_net.c                | 14 +++++++-------
 2 files changed, 10 insertions(+), 9 deletions(-)
  

On 7/10/20 4:38 AM, Joyce Kong wrote:
> Restrict pointer aliasing to allow the compiler to vectorize loop
> more aggressively.
> 
> With this patch, a 9.6% improvement is observed in throughput for
> the packed virtio-net PVP case, and a 2.8% improvement in throughput
> for the packed virtio-user PVP case. All performance data are measured
> on ThunderX-2 platform under 0.001% acceptable packet loss with 1 core
> on both vhost and virtio side.
> 
> Signed-off-by: Joyce Kong <joyce.kong@arm.com>
> Reviewed-by: Phil Yang <phil.yang@arm.com>
> ---
>  drivers/net/virtio/virtio_rxtx_simple_neon.c |  5 +++--
>  lib/librte_vhost/virtio_net.c                | 14 +++++++-------
>  2 files changed, 10 insertions(+), 9 deletions(-)
> 
> diff --git a/drivers/net/virtio/virtio_rxtx_simple_neon.c b/drivers/net/virtio/virtio_rxtx_simple_neon.c
> index a9b649814..02520fda8 100644
> --- a/drivers/net/virtio/virtio_rxtx_simple_neon.c
> +++ b/drivers/net/virtio/virtio_rxtx_simple_neon.c
> @@ -36,8 +36,9 @@
>   * - nb_pkts < RTE_VIRTIO_DESC_PER_LOOP, just return no packet
>   */
>  uint16_t
> -virtio_recv_pkts_vec(void *rx_queue, struct rte_mbuf
> -		**__rte_restrict rx_pkts, uint16_t nb_pkts)
> +virtio_recv_pkts_vec(void *rx_queue,
> +		struct rte_mbuf **__rte_restrict rx_pkts,
> +		uint16_t nb_pkts)
>  {
>  	struct virtnet_rx *rxvq = rx_queue;
>  	struct virtqueue *vq = rxvq->vq;
> diff --git a/lib/librte_vhost/virtio_net.c b/lib/librte_vhost/virtio_net.c
> index 236498f71..1d0be3dd4 100644
> --- a/lib/librte_vhost/virtio_net.c
> +++ b/lib/librte_vhost/virtio_net.c
> @@ -1353,8 +1353,8 @@ virtio_dev_rx_single_packed(struct virtio_net *dev,
>  
>  static __rte_noinline uint32_t
>  virtio_dev_rx_packed(struct virtio_net *dev,
> -		     struct vhost_virtqueue *vq,
> -		     struct rte_mbuf **pkts,
> +		     struct vhost_virtqueue *__rte_restrict vq,
> +		     struct rte_mbuf **__rte_restrict pkts,
>  		     uint32_t count)
>  {
>  	uint32_t pkt_idx = 0;
> @@ -1439,7 +1439,7 @@ virtio_dev_rx(struct virtio_net *dev, uint16_t queue_id,
>  
>  uint16_t
>  rte_vhost_enqueue_burst(int vid, uint16_t queue_id,
> -	struct rte_mbuf **pkts, uint16_t count)
> +	struct rte_mbuf **__rte_restrict pkts, uint16_t count)
>  {
>  	struct virtio_net *dev = get_device(vid);
>  
> @@ -2671,9 +2671,9 @@ free_zmbuf(struct vhost_virtqueue *vq)
>  
>  static __rte_noinline uint16_t
>  virtio_dev_tx_packed_zmbuf(struct virtio_net *dev,
> -			   struct vhost_virtqueue *vq,
> +			   struct vhost_virtqueue *__rte_restrict vq,
>  			   struct rte_mempool *mbuf_pool,
> -			   struct rte_mbuf **pkts,
> +			   struct rte_mbuf **__rte_restrict pkts,
>  			   uint32_t count)
>  {
>  	uint32_t pkt_idx = 0;
> @@ -2707,9 +2707,9 @@ virtio_dev_tx_packed_zmbuf(struct virtio_net *dev,
>  
>  static __rte_noinline uint16_t
>  virtio_dev_tx_packed(struct virtio_net *dev,
> -		     struct vhost_virtqueue *vq,
> +		     struct vhost_virtqueue *__rte_restrict vq,
>  		     struct rte_mempool *mbuf_pool,
> -		     struct rte_mbuf **pkts,
> +		     struct rte_mbuf **__rte_restrict pkts,
>  		     uint32_t count)
>  {
>  	uint32_t pkt_idx = 0;
> 

The vhost part looks good to me.

Acked-by: Adrián Moreno <amorenoz@redhat.com>