[dpdk-dev] [PATCH 7/7] net/mlx5: remove inefficient prefetching

[Nelio Laranjeiro]

Prefetching completion queue entries is inefficient because too few CPU
cycles are spent before their use, which results into cache misses anyway.

Signed-off-by: Nelio Laranjeiro <nelio.laranjeiro at 6wind.com>
Acked-by: Adrien Mazarguil <adrien.mazarguil at 6wind.com>
---
 drivers/net/mlx5/mlx5_rxtx.c | 21 ---------------------
 1 file changed, 21 deletions(-)

diff --git a/drivers/net/mlx5/mlx5_rxtx.c b/drivers/net/mlx5/mlx5_rxtx.c
index 4b8c197..9f74fd4 100644
--- a/drivers/net/mlx5/mlx5_rxtx.c
+++ b/drivers/net/mlx5/mlx5_rxtx.c
@@ -348,23 +348,6 @@ mlx5_tx_dbrec(struct txq *txq)
 }
 
 /**
- * Prefetch a CQE.
- *
- * @param txq
- *   Pointer to TX queue structure.
- * @param cqe_ci
- *   CQE consumer index.
- */
-static inline void
-tx_prefetch_cqe(struct txq *txq, uint16_t ci)
-{
-	volatile struct mlx5_cqe *cqe;
-
-	cqe = &(*txq->cqes)[ci & ((1 << txq->cqe_n) - 1)];
-	rte_prefetch0(cqe);
-}
-
-/**
  * DPDK callback for TX.
  *
  * @param dpdk_txq
@@ -395,8 +378,6 @@ mlx5_tx_burst(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	if (unlikely(!pkts_n))
 		return 0;
 	/* Prefetch first packet cacheline. */
-	tx_prefetch_cqe(txq, txq->cq_ci);
-	tx_prefetch_cqe(txq, txq->cq_ci + 1);
 	rte_prefetch0(*pkts);
 	/* Start processing. */
 	txq_complete(txq);
@@ -733,7 +714,6 @@ mlx5_tx_burst_mpw(void *dpdk_txq, struct rte_mbuf **pkts, uint16_t pkts_n)
 	if (unlikely(!pkts_n))
 		return 0;
 	/* Prefetch first packet cacheline. */
-	tx_prefetch_cqe(txq, txq->cq_ci);
 	rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci));
 	rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci + 1));
 	/* Start processing. */
@@ -938,7 +918,6 @@ mlx5_tx_burst_mpw_inline(void *dpdk_txq, struct rte_mbuf **pkts,
 	if (unlikely(!pkts_n))
 		return 0;
 	/* Prefetch first packet cacheline. */
-	tx_prefetch_cqe(txq, txq->cq_ci);
 	rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci));
 	rte_prefetch0(tx_mlx5_wqe(txq, txq->wqe_ci + 1));
 	/* Start processing. */
-- 
2.1.4