On Wed, Jan 29, 2025 at 05:48:55PM +0530, Shaiq Wani wrote:
> In case some CPUs don't support AVX512. Enable AVX2 for them to
> get better per-core performance.
>
> The single queue model processes all packets in order while
> the split queue model separates packet data and metadata into
> different queues for parallel processing and improved performance
>
> Signed-off-by: Shaiq Wani <shaiq.w...@intel.com>
> ---
> doc/guides/rel_notes/release_25_03.rst | 7 +
> drivers/common/idpf/idpf_common_device.h | 1 +
> drivers/common/idpf/idpf_common_rxtx.h | 4 +
> drivers/common/idpf/idpf_common_rxtx_avx2.c | 224 ++++++++++++++++++++
> drivers/common/idpf/version.map | 1 +
> drivers/net/idpf/idpf_rxtx.c | 13 ++
> 6 files changed, 250 insertions(+)
>
> diff --git a/doc/guides/rel_notes/release_25_03.rst
> b/doc/guides/rel_notes/release_25_03.rst
> index 85986ffa61..a075bdf695 100644
> --- a/doc/guides/rel_notes/release_25_03.rst
> +++ b/doc/guides/rel_notes/release_25_03.rst
> @@ -63,6 +63,13 @@ New Features
> and even substantial part of its code.
> It can be viewed as an extension of rte_ring functionality.
>
> +* **Added support of AVX2 instructions on IDPF.**
> +
> + Support for AVX2 instructions in IDPF single queue RX and TX path
> + added.The single queue model processes all packets in order within
> + one RX queue, while the split queue model separates packet data and
> + metadata into different queues for parallel processing and improved
> performance.
> +
>
Good to see the release note update here. Please also merge in the other
doc updates into this patch too.
> Removed Items
> -------------
> diff --git a/drivers/common/idpf/idpf_common_device.h
> b/drivers/common/idpf/idpf_common_device.h
> index 734be1c88a..5f3e4a4fcf 100644
> --- a/drivers/common/idpf/idpf_common_device.h
> +++ b/drivers/common/idpf/idpf_common_device.h
> @@ -124,6 +124,7 @@ struct idpf_vport {
> bool rx_vec_allowed;
> bool tx_vec_allowed;
> bool rx_use_avx2;
> + bool tx_use_avx2;
> bool rx_use_avx512;
> bool tx_use_avx512;
>
> diff --git a/drivers/common/idpf/idpf_common_rxtx.h
> b/drivers/common/idpf/idpf_common_rxtx.h
> index f50cf5ef46..e19e1878f3 100644
> --- a/drivers/common/idpf/idpf_common_rxtx.h
> +++ b/drivers/common/idpf/idpf_common_rxtx.h
> @@ -306,5 +306,9 @@ __rte_internal
> uint16_t idpf_dp_singleq_recv_pkts_avx2(void *rx_queue,
> struct rte_mbuf **rx_pkts,
> uint16_t nb_pkts);
> +__rte_internal
> +uint16_t idpf_dp_singleq_xmit_pkts_avx2(void *tx_queue,
> + struct rte_mbuf **tx_pkts,
> + uint16_t nb_pkts);
>
> #endif /* _IDPF_COMMON_RXTX_H_ */
> diff --git a/drivers/common/idpf/idpf_common_rxtx_avx2.c
> b/drivers/common/idpf/idpf_common_rxtx_avx2.c
> index de76f01ff8..f82b6d7f4b 100644
> --- a/drivers/common/idpf/idpf_common_rxtx_avx2.c
> +++ b/drivers/common/idpf/idpf_common_rxtx_avx2.c
> @@ -483,3 +483,227 @@ idpf_dp_singleq_recv_pkts_avx2(void *rx_queue, struct
> rte_mbuf **rx_pkts, uint16
> {
> return _idpf_singleq_recv_raw_pkts_vec_avx2(rx_queue, rx_pkts, nb_pkts);
> }
> +static __rte_always_inline void
> +idpf_tx_backlog_entry(struct idpf_tx_entry *txep,
> + struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
> +{
> + int i;
> +
> + for (i = 0; i < (int)nb_pkts; ++i)
> + txep[i].mbuf = tx_pkts[i];
> +}
> +
I don't think it should block this patch - since other paths already use
idpf-specific versions - but in next release we must look to switch this
driver over to using the common tx (and rx) primitives now in
net/intel/common.
/Bruce
