[v2] net/octeontx2: perf improvement to Tx vector func
diff mbox series

Message ID 20200111135742.7782-1-pbhagavatula@marvell.com
State Accepted, archived
Delegated to: Jerin Jacob
Headers show
Series
  • [v2] net/octeontx2: perf improvement to Tx vector func
Related show

Checks

Context Check Description
ci/Intel-compilation success Compilation OK
ci/iol-intel-Performance success Performance Testing PASS
ci/travis-robot success Travis build: passed
ci/iol-mellanox-Performance success Performance Testing PASS
ci/iol-nxp-Performance success Performance Testing PASS
ci/iol-testing success Testing PASS
ci/checkpatch success coding style OK

Commit Message

Pavan Nikhilesh Bhagavatula Jan. 11, 2020, 1:57 p.m. UTC
From: Pavan Nikhilesh <pbhagavatula@marvell.com>

Use scalar loads instead of vector loads for fields
that don't need any vector operations.

Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
 drivers/net/octeontx2/otx2_tx.c | 15 +++++----------
 1 file changed, 5 insertions(+), 10 deletions(-)

Comments

Jerin Jacob Jan. 14, 2020, 6:27 a.m. UTC | #1
On Sat, Jan 11, 2020 at 7:27 PM <pbhagavatula@marvell.com> wrote:
>
> From: Pavan Nikhilesh <pbhagavatula@marvell.com>
>
> Use scalar loads instead of vector loads for fields
> that don't need any vector operations.
>
> Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>


Acked-by: Jerin Jacob <jerinj@marvell.com>
Applied to dpdk-next-net-mrvl/master. Thanks

> ---
>  drivers/net/octeontx2/otx2_tx.c | 15 +++++----------
>  1 file changed, 5 insertions(+), 10 deletions(-)
>
> diff --git a/drivers/net/octeontx2/otx2_tx.c b/drivers/net/octeontx2/otx2_tx.c
> index fa533000e..9f5926560 100644
> --- a/drivers/net/octeontx2/otx2_tx.c
> +++ b/drivers/net/octeontx2/otx2_tx.c
> @@ -112,7 +112,6 @@ nix_xmit_pkts_vector(void *tx_queue, struct rte_mbuf **tx_pkts,
>         uint64x2_t ltypes01, ltypes23;
>         uint64x2_t xtmp128, ytmp128;
>         uint64x2_t xmask01, xmask23;
> -       uint64x2_t mbuf01, mbuf23;
>         uint64x2_t cmd00, cmd01;
>         uint64x2_t cmd10, cmd11;
>         uint64x2_t cmd20, cmd21;
> @@ -137,9 +136,6 @@ nix_xmit_pkts_vector(void *tx_queue, struct rte_mbuf **tx_pkts,
>         sgdesc23_w0 = sgdesc01_w0;
>
>         for (i = 0; i < pkts; i += NIX_DESCS_PER_LOOP) {
> -               mbuf01 = vld1q_u64((uint64_t *)tx_pkts);
> -               mbuf23 = vld1q_u64((uint64_t *)(tx_pkts + 2));
> -
>                 /* Clear lower 32bit of SEND_HDR_W0 and SEND_SG_W0 */
>                 senddesc01_w0 = vbicq_u64(senddesc01_w0,
>                                           vdupq_n_u64(0xFFFFFFFF));
> @@ -149,13 +145,11 @@ nix_xmit_pkts_vector(void *tx_queue, struct rte_mbuf **tx_pkts,
>                 senddesc23_w0 = senddesc01_w0;
>                 sgdesc23_w0 = sgdesc01_w0;
>
> -               tx_pkts = tx_pkts + NIX_DESCS_PER_LOOP;
> -
>                 /* Move mbufs to iova */
> -               mbuf0 = (uint64_t *)vgetq_lane_u64(mbuf01, 0);
> -               mbuf1 = (uint64_t *)vgetq_lane_u64(mbuf01, 1);
> -               mbuf2 = (uint64_t *)vgetq_lane_u64(mbuf23, 0);
> -               mbuf3 = (uint64_t *)vgetq_lane_u64(mbuf23, 1);
> +               mbuf0 = (uint64_t *)tx_pkts[0];
> +               mbuf1 = (uint64_t *)tx_pkts[1];
> +               mbuf2 = (uint64_t *)tx_pkts[2];
> +               mbuf3 = (uint64_t *)tx_pkts[3];
>
>                 mbuf0 = (uint64_t *)((uintptr_t)mbuf0 +
>                                      offsetof(struct rte_mbuf, buf_iova));
> @@ -927,6 +921,7 @@ nix_xmit_pkts_vector(void *tx_queue, struct rte_mbuf **tx_pkts,
>                         lmt_status = otx2_lmt_submit(io_addr);
>
>                 } while (lmt_status == 0);
> +               tx_pkts = tx_pkts + NIX_DESCS_PER_LOOP;
>         }
>
>         return pkts;
> --
> 2.17.1
>

Patch
diff mbox series

diff --git a/drivers/net/octeontx2/otx2_tx.c b/drivers/net/octeontx2/otx2_tx.c
index fa533000e..9f5926560 100644
--- a/drivers/net/octeontx2/otx2_tx.c
+++ b/drivers/net/octeontx2/otx2_tx.c
@@ -112,7 +112,6 @@  nix_xmit_pkts_vector(void *tx_queue, struct rte_mbuf **tx_pkts,
 	uint64x2_t ltypes01, ltypes23;
 	uint64x2_t xtmp128, ytmp128;
 	uint64x2_t xmask01, xmask23;
-	uint64x2_t mbuf01, mbuf23;
 	uint64x2_t cmd00, cmd01;
 	uint64x2_t cmd10, cmd11;
 	uint64x2_t cmd20, cmd21;
@@ -137,9 +136,6 @@  nix_xmit_pkts_vector(void *tx_queue, struct rte_mbuf **tx_pkts,
 	sgdesc23_w0 = sgdesc01_w0;
 
 	for (i = 0; i < pkts; i += NIX_DESCS_PER_LOOP) {
-		mbuf01 = vld1q_u64((uint64_t *)tx_pkts);
-		mbuf23 = vld1q_u64((uint64_t *)(tx_pkts + 2));
-
 		/* Clear lower 32bit of SEND_HDR_W0 and SEND_SG_W0 */
 		senddesc01_w0 = vbicq_u64(senddesc01_w0,
 					  vdupq_n_u64(0xFFFFFFFF));
@@ -149,13 +145,11 @@  nix_xmit_pkts_vector(void *tx_queue, struct rte_mbuf **tx_pkts,
 		senddesc23_w0 = senddesc01_w0;
 		sgdesc23_w0 = sgdesc01_w0;
 
-		tx_pkts = tx_pkts + NIX_DESCS_PER_LOOP;
-
 		/* Move mbufs to iova */
-		mbuf0 = (uint64_t *)vgetq_lane_u64(mbuf01, 0);
-		mbuf1 = (uint64_t *)vgetq_lane_u64(mbuf01, 1);
-		mbuf2 = (uint64_t *)vgetq_lane_u64(mbuf23, 0);
-		mbuf3 = (uint64_t *)vgetq_lane_u64(mbuf23, 1);
+		mbuf0 = (uint64_t *)tx_pkts[0];
+		mbuf1 = (uint64_t *)tx_pkts[1];
+		mbuf2 = (uint64_t *)tx_pkts[2];
+		mbuf3 = (uint64_t *)tx_pkts[3];
 
 		mbuf0 = (uint64_t *)((uintptr_t)mbuf0 +
 				     offsetof(struct rte_mbuf, buf_iova));
@@ -927,6 +921,7 @@  nix_xmit_pkts_vector(void *tx_queue, struct rte_mbuf **tx_pkts,
 			lmt_status = otx2_lmt_submit(io_addr);
 
 		} while (lmt_status == 0);
+		tx_pkts = tx_pkts + NIX_DESCS_PER_LOOP;
 	}
 
 	return pkts;