diff mbox series

[v2,18/22] net/ice: use vector SW ring for all vector paths

Message ID	20241203164132.2686558-19-bruce.richardson@intel.com (mailing list archive)
State	Superseded
Delegated to:	Thomas Monjalon
Headers	From: Bruce Richardson <bruce.richardson@intel.com> To: dev@dpdk.org Cc: Bruce Richardson <bruce.richardson@intel.com>, Anatoly Burakov <anatoly.burakov@intel.com>, Konstantin Ananyev <konstantin.v.ananyev@yandex.ru> Subject: [PATCH v2 18/22] net/ice: use vector SW ring for all vector paths Date: Tue, 3 Dec 2024 16:41:24 +0000 Message-ID: <20241203164132.2686558-19-bruce.richardson@intel.com> In-Reply-To: <20241203164132.2686558-1-bruce.richardson@intel.com> References: <20241122125418.2857301-1-bruce.richardson@intel.com> <20241203164132.2686558-1-bruce.richardson@intel.com> MIME-Version: 1.0 Content-Transfer-Encoding: 8bit Precedence: list Errors-To: dev-bounces@dpdk.org
Series	Reduce code duplication across Intel NIC drivers \| [v2,00/22] Reduce code duplication across Intel NIC drivers [v2,01/22] net/_common_intel: add pkt reassembly fn for intel drivers [v2,02/22] net/_common_intel: provide common Tx entry structures [v2,03/22] net/_common_intel: add Tx mbuf ring replenish fn [v2,04/22] drivers/net: align Tx queue struct field names [v2,05/22] drivers/net: add prefix for driver-specific structs [v2,06/22] net/_common_intel: merge ice and i40e Tx queue struct [v2,07/22] net/iavf: use common Tx queue structure [v2,08/22] net/ixgbe: convert Tx queue context cache field to ptr [v2,09/22] net/ixgbe: use common Tx queue structure [v2,10/22] net/_common_intel: pack Tx queue structure [v2,11/22] net/_common_intel: add post-Tx buffer free function [v2,12/22] net/_common_intel: add Tx buffer free fn for AVX-512 [v2,13/22] net/iavf: use common Tx free fn for AVX-512 [v2,14/22] net/ice: move Tx queue mbuf cleanup fn to common [v2,15/22] net/i40e: use common Tx queue mbuf cleanup fn [v2,16/22] net/ixgbe: use common Tx queue mbuf cleanup fn [v2,17/22] net/iavf: use common Tx queue mbuf cleanup fn [v2,18/22] net/ice: use vector SW ring for all vector paths [v2,19/22] net/i40e: use vector SW ring for all vector paths [v2,20/22] net/iavf: use vector SW ring for all vector paths [v2,21/22] net/_common_intel: remove unneeded code [v2,22/22] net/ixgbe: use common Tx backlog entry fn

Checks

Context	Check	Description
ci/checkpatch	success	coding style OK

Commit Message

Bruce Richardson Dec. 3, 2024, 4:41 p.m. UTC

The AVX-512 code path used a smaller SW ring structure only containing
the mbuf pointer, but no other fields. The other fields are only used in
the scalar code path, so update all vector driver code paths to use the
smaller, faster structure.

Signed-off-by: Bruce Richardson <bruce.richardson@intel.com>
---
 drivers/net/_common_intel/tx.h        |  7 +++++++
 drivers/net/ice/ice_rxtx.c            |  2 +-
 drivers/net/ice/ice_rxtx_vec_avx2.c   | 12 ++++++------
 drivers/net/ice/ice_rxtx_vec_avx512.c | 14 ++------------
 drivers/net/ice/ice_rxtx_vec_common.h |  6 ------
 drivers/net/ice/ice_rxtx_vec_sse.c    | 12 ++++++------
 6 files changed, 22 insertions(+), 31 deletions(-)

diff mbox series

Patch

diff --git a/drivers/net/_common_intel/tx.h b/drivers/net/_common_intel/tx.h
index 310b51adcf..aa42b9b49f 100644
--- a/drivers/net/_common_intel/tx.h
+++ b/drivers/net/_common_intel/tx.h
@@ -109,6 +109,13 @@  ci_tx_backlog_entry(struct ci_tx_entry *txep, struct rte_mbuf **tx_pkts, uint16_
 		txep[i].mbuf = tx_pkts[i];
 }
 
+static __rte_always_inline void
+ci_tx_backlog_entry_vec(struct ci_tx_entry_vec *txep, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
+{
+	for (uint16_t i = 0; i < nb_pkts; ++i)
+		txep[i].mbuf = tx_pkts[i];
+}
+
 #define IETH_VPMD_TX_MAX_FREE_BUF 64
 
 typedef int (*ci_desc_done_fn)(struct ci_tx_queue *txq, uint16_t idx);
diff --git a/drivers/net/ice/ice_rxtx.c b/drivers/net/ice/ice_rxtx.c
index ad0ddf6a88..77cb6688a7 100644
--- a/drivers/net/ice/ice_rxtx.c
+++ b/drivers/net/ice/ice_rxtx.c
@@ -825,7 +825,7 @@  ice_tx_queue_start(struct rte_eth_dev *dev, uint16_t tx_queue_id)
 
 	/* record what kind of descriptor cleanup we need on teardown */
 	txq->vector_tx = ad->tx_vec_allowed;
-	txq->vector_sw_ring = ad->tx_use_avx512;
+	txq->vector_sw_ring = txq->vector_tx;
 
 	dev->data->tx_queue_state[tx_queue_id] = RTE_ETH_QUEUE_STATE_STARTED;
 
diff --git a/drivers/net/ice/ice_rxtx_vec_avx2.c b/drivers/net/ice/ice_rxtx_vec_avx2.c
index 12ffa0fa9a..98bab322b4 100644
--- a/drivers/net/ice/ice_rxtx_vec_avx2.c
+++ b/drivers/net/ice/ice_rxtx_vec_avx2.c
@@ -858,7 +858,7 @@  ice_xmit_fixed_burst_vec_avx2(void *tx_queue, struct rte_mbuf **tx_pkts,
 {
 	struct ci_tx_queue *txq = (struct ci_tx_queue *)tx_queue;
 	volatile struct ice_tx_desc *txdp;
-	struct ci_tx_entry *txep;
+	struct ci_tx_entry_vec *txep;
 	uint16_t n, nb_commit, tx_id;
 	uint64_t flags = ICE_TD_CMD;
 	uint64_t rs = ICE_TX_DESC_CMD_RS | ICE_TD_CMD;
@@ -867,7 +867,7 @@  ice_xmit_fixed_burst_vec_avx2(void *tx_queue, struct rte_mbuf **tx_pkts,
 	nb_pkts = RTE_MIN(nb_pkts, txq->tx_rs_thresh);
 
 	if (txq->nb_tx_free < txq->tx_free_thresh)
-		ice_tx_free_bufs_vec(txq);
+		ci_tx_free_bufs_vec(txq, ice_tx_desc_done, false);
 
 	nb_commit = nb_pkts = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts);
 	if (unlikely(nb_pkts == 0))
@@ -875,13 +875,13 @@  ice_xmit_fixed_burst_vec_avx2(void *tx_queue, struct rte_mbuf **tx_pkts,
 
 	tx_id = txq->tx_tail;
 	txdp = &txq->ice_tx_ring[tx_id];
-	txep = &txq->sw_ring[tx_id];
+	txep = &txq->sw_ring_vec[tx_id];
 
 	txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts);
 
 	n = (uint16_t)(txq->nb_tx_desc - tx_id);
 	if (nb_commit >= n) {
-		ci_tx_backlog_entry(txep, tx_pkts, n);
+		ci_tx_backlog_entry_vec(txep, tx_pkts, n);
 
 		ice_vtx(txdp, tx_pkts, n - 1, flags, offload);
 		tx_pkts += (n - 1);
@@ -896,10 +896,10 @@  ice_xmit_fixed_burst_vec_avx2(void *tx_queue, struct rte_mbuf **tx_pkts,
 
 		/* avoid reach the end of ring */
 		txdp = &txq->ice_tx_ring[tx_id];
-		txep = &txq->sw_ring[tx_id];
+		txep = &txq->sw_ring_vec[tx_id];
 	}
 
-	ci_tx_backlog_entry(txep, tx_pkts, nb_commit);
+	ci_tx_backlog_entry_vec(txep, tx_pkts, nb_commit);
 
 	ice_vtx(txdp, tx_pkts, nb_commit, flags, offload);
 
diff --git a/drivers/net/ice/ice_rxtx_vec_avx512.c b/drivers/net/ice/ice_rxtx_vec_avx512.c
index f6ec593f96..481f784e34 100644
--- a/drivers/net/ice/ice_rxtx_vec_avx512.c
+++ b/drivers/net/ice/ice_rxtx_vec_avx512.c
@@ -924,16 +924,6 @@  ice_vtx(volatile struct ice_tx_desc *txdp, struct rte_mbuf **pkt,
 	}
 }
 
-static __rte_always_inline void
-ice_tx_backlog_entry_avx512(struct ci_tx_entry_vec *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];
-}
-
 static __rte_always_inline uint16_t
 ice_xmit_fixed_burst_vec_avx512(void *tx_queue, struct rte_mbuf **tx_pkts,
 				uint16_t nb_pkts, bool do_offload)
@@ -964,7 +954,7 @@  ice_xmit_fixed_burst_vec_avx512(void *tx_queue, struct rte_mbuf **tx_pkts,
 
 	n = (uint16_t)(txq->nb_tx_desc - tx_id);
 	if (nb_commit >= n) {
-		ice_tx_backlog_entry_avx512(txep, tx_pkts, n);
+		ci_tx_backlog_entry_vec(txep, tx_pkts, n);
 
 		ice_vtx(txdp, tx_pkts, n - 1, flags, do_offload);
 		tx_pkts += (n - 1);
@@ -982,7 +972,7 @@  ice_xmit_fixed_burst_vec_avx512(void *tx_queue, struct rte_mbuf **tx_pkts,
 		txep = (void *)txq->sw_ring;
 	}
 
-	ice_tx_backlog_entry_avx512(txep, tx_pkts, nb_commit);
+	ci_tx_backlog_entry_vec(txep, tx_pkts, nb_commit);
 
 	ice_vtx(txdp, tx_pkts, nb_commit, flags, do_offload);
 
diff --git a/drivers/net/ice/ice_rxtx_vec_common.h b/drivers/net/ice/ice_rxtx_vec_common.h
index 907828b675..aa709fb51c 100644
--- a/drivers/net/ice/ice_rxtx_vec_common.h
+++ b/drivers/net/ice/ice_rxtx_vec_common.h
@@ -20,12 +20,6 @@  ice_tx_desc_done(struct ci_tx_queue *txq, uint16_t idx)
 				rte_cpu_to_le_64(ICE_TX_DESC_DTYPE_DESC_DONE);
 }
 
-static __rte_always_inline int
-ice_tx_free_bufs_vec(struct ci_tx_queue *txq)
-{
-	return ci_tx_free_bufs(txq, ice_tx_desc_done);
-}
-
 static inline void
 _ice_rx_queue_release_mbufs_vec(struct ice_rx_queue *rxq)
 {
diff --git a/drivers/net/ice/ice_rxtx_vec_sse.c b/drivers/net/ice/ice_rxtx_vec_sse.c
index bff39c28d8..73e3e9eb54 100644
--- a/drivers/net/ice/ice_rxtx_vec_sse.c
+++ b/drivers/net/ice/ice_rxtx_vec_sse.c
@@ -699,7 +699,7 @@  ice_xmit_fixed_burst_vec(void *tx_queue, struct rte_mbuf **tx_pkts,
 {
 	struct ci_tx_queue *txq = (struct ci_tx_queue *)tx_queue;
 	volatile struct ice_tx_desc *txdp;
-	struct ci_tx_entry *txep;
+	struct ci_tx_entry_vec *txep;
 	uint16_t n, nb_commit, tx_id;
 	uint64_t flags = ICE_TD_CMD;
 	uint64_t rs = ICE_TX_DESC_CMD_RS | ICE_TD_CMD;
@@ -709,7 +709,7 @@  ice_xmit_fixed_burst_vec(void *tx_queue, struct rte_mbuf **tx_pkts,
 	nb_pkts = RTE_MIN(nb_pkts, txq->tx_rs_thresh);
 
 	if (txq->nb_tx_free < txq->tx_free_thresh)
-		ice_tx_free_bufs_vec(txq);
+		ci_tx_free_bufs_vec(txq, ice_tx_desc_done, false);
 
 	nb_pkts = (uint16_t)RTE_MIN(txq->nb_tx_free, nb_pkts);
 	nb_commit = nb_pkts;
@@ -718,13 +718,13 @@  ice_xmit_fixed_burst_vec(void *tx_queue, struct rte_mbuf **tx_pkts,
 
 	tx_id = txq->tx_tail;
 	txdp = &txq->ice_tx_ring[tx_id];
-	txep = &txq->sw_ring[tx_id];
+	txep = &txq->sw_ring_vec[tx_id];
 
 	txq->nb_tx_free = (uint16_t)(txq->nb_tx_free - nb_pkts);
 
 	n = (uint16_t)(txq->nb_tx_desc - tx_id);
 	if (nb_commit >= n) {
-		ci_tx_backlog_entry(txep, tx_pkts, n);
+		ci_tx_backlog_entry_vec(txep, tx_pkts, n);
 
 		for (i = 0; i < n - 1; ++i, ++tx_pkts, ++txdp)
 			ice_vtx1(txdp, *tx_pkts, flags);
@@ -738,10 +738,10 @@  ice_xmit_fixed_burst_vec(void *tx_queue, struct rte_mbuf **tx_pkts,
 
 		/* avoid reach the end of ring */
 		txdp = &txq->ice_tx_ring[tx_id];
-		txep = &txq->sw_ring[tx_id];
+		txep = &txq->sw_ring_vec[tx_id];
 	}
 
-	ci_tx_backlog_entry(txep, tx_pkts, nb_commit);
+	ci_tx_backlog_entry_vec(txep, tx_pkts, nb_commit);
 
 	ice_vtx(txdp, tx_pkts, nb_commit, flags);