DPDK patches and discussions
 help / color / mirror / Atom feed
From: Nithin Dabilpuram <ndabilpuram@marvell.com>
To: Jerin Jacob <jerinj@marvell.com>,
	Nithin Dabilpuram <ndabilpuram@marvell.com>,
	Kiran Kumar K <kirankumark@marvell.com>
Cc: <dev@dpdk.org>, Andrew Pinski <apinski@marvell.com>
Subject: [dpdk-dev] [PATCH] net/octeontx2: perf improvement to rx vector func
Date: Tue, 10 Dec 2019 17:38:44 +0530	[thread overview]
Message-ID: <20191210120844.50017-1-ndabilpuram@marvell.com> (raw)

From: Jerin Jacob <jerinj@marvell.com>

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

Signed-off-by: Andrew Pinski <apinski@marvell.com>
Signed-off-by: Jerin Jacob <jerinj@marvell.com>
Signed-off-by: Nithin Dabilpuram <ndabilpuram@marvell.com>
---
 drivers/net/octeontx2/otx2_rx.c | 48 ++++++++++++++++++++---------------------
 1 file changed, 24 insertions(+), 24 deletions(-)

diff --git a/drivers/net/octeontx2/otx2_rx.c b/drivers/net/octeontx2/otx2_rx.c
index 48565db..db4a221 100644
--- a/drivers/net/octeontx2/otx2_rx.c
+++ b/drivers/net/octeontx2/otx2_rx.c
@@ -184,17 +184,21 @@ nix_recv_pkts_vector(void *rx_queue, struct rte_mbuf **rx_pkts,
 		f3 = vqtbl1q_u8(cq3_w8, shuf_msk);
 
 		/* Load CQE word0 and word 1 */
-		uint64x2_t cq0_w0 = vld1q_u64((uint64_t *)(cq0 + CQE_SZ(0)));
-		uint64x2_t cq1_w0 = vld1q_u64((uint64_t *)(cq0 + CQE_SZ(1)));
-		uint64x2_t cq2_w0 = vld1q_u64((uint64_t *)(cq0 + CQE_SZ(2)));
-		uint64x2_t cq3_w0 = vld1q_u64((uint64_t *)(cq0 + CQE_SZ(3)));
+		uint64_t cq0_w0 = ((uint64_t *)(cq0 + CQE_SZ(0)))[0];
+		uint64_t cq0_w1 = ((uint64_t *)(cq0 + CQE_SZ(0)))[1];
+		uint64_t cq1_w0 = ((uint64_t *)(cq0 + CQE_SZ(1)))[0];
+		uint64_t cq1_w1 = ((uint64_t *)(cq0 + CQE_SZ(1)))[1];
+		uint64_t cq2_w0 = ((uint64_t *)(cq0 + CQE_SZ(2)))[0];
+		uint64_t cq2_w1 = ((uint64_t *)(cq0 + CQE_SZ(2)))[1];
+		uint64_t cq3_w0 = ((uint64_t *)(cq0 + CQE_SZ(3)))[0];
+		uint64_t cq3_w1 = ((uint64_t *)(cq0 + CQE_SZ(3)))[1];
 
 		if (flags & NIX_RX_OFFLOAD_RSS_F) {
 			/* Fill rss in the rx_descriptor_fields1 */
-			f0 = vsetq_lane_u32(vgetq_lane_u32(cq0_w0, 0), f0, 3);
-			f1 = vsetq_lane_u32(vgetq_lane_u32(cq1_w0, 0), f1, 3);
-			f2 = vsetq_lane_u32(vgetq_lane_u32(cq2_w0, 0), f2, 3);
-			f3 = vsetq_lane_u32(vgetq_lane_u32(cq3_w0, 0), f3, 3);
+			f0 = vsetq_lane_u32(cq0_w0, f0, 3);
+			f1 = vsetq_lane_u32(cq1_w0, f1, 3);
+			f2 = vsetq_lane_u32(cq2_w0, f2, 3);
+			f3 = vsetq_lane_u32(cq3_w0, f3, 3);
 			ol_flags0 = PKT_RX_RSS_HASH;
 			ol_flags1 = PKT_RX_RSS_HASH;
 			ol_flags2 = PKT_RX_RSS_HASH;
@@ -206,25 +210,21 @@ nix_recv_pkts_vector(void *rx_queue, struct rte_mbuf **rx_pkts,
 
 		if (flags & NIX_RX_OFFLOAD_PTYPE_F) {
 			/* Fill packet_type in the rx_descriptor_fields1 */
-			f0 = vsetq_lane_u32(nix_ptype_get(lookup_mem,
-					    vgetq_lane_u64(cq0_w0, 1)), f0, 0);
-			f1 = vsetq_lane_u32(nix_ptype_get(lookup_mem,
-					    vgetq_lane_u64(cq1_w0, 1)), f1, 0);
-			f2 = vsetq_lane_u32(nix_ptype_get(lookup_mem,
-					    vgetq_lane_u64(cq2_w0, 1)), f2, 0);
-			f3 = vsetq_lane_u32(nix_ptype_get(lookup_mem,
-					    vgetq_lane_u64(cq3_w0, 1)), f3, 0);
+			f0 = vsetq_lane_u32(nix_ptype_get(lookup_mem, cq0_w1),
+					    f0, 0);
+			f1 = vsetq_lane_u32(nix_ptype_get(lookup_mem, cq1_w1),
+					    f1, 0);
+			f2 = vsetq_lane_u32(nix_ptype_get(lookup_mem, cq2_w1),
+					    f2, 0);
+			f3 = vsetq_lane_u32(nix_ptype_get(lookup_mem, cq3_w1),
+					    f3, 0);
 		}
 
 		if (flags & NIX_RX_OFFLOAD_CHECKSUM_F) {
-			ol_flags0 |= nix_rx_olflags_get(lookup_mem,
-						vgetq_lane_u64(cq0_w0, 1));
-			ol_flags1 |= nix_rx_olflags_get(lookup_mem,
-						vgetq_lane_u64(cq1_w0, 1));
-			ol_flags2 |= nix_rx_olflags_get(lookup_mem,
-						vgetq_lane_u64(cq2_w0, 1));
-			ol_flags3 |= nix_rx_olflags_get(lookup_mem,
-						vgetq_lane_u64(cq3_w0, 1));
+			ol_flags0 |= nix_rx_olflags_get(lookup_mem, cq0_w1);
+			ol_flags1 |= nix_rx_olflags_get(lookup_mem, cq1_w1);
+			ol_flags2 |= nix_rx_olflags_get(lookup_mem, cq2_w1);
+			ol_flags3 |= nix_rx_olflags_get(lookup_mem, cq3_w1);
 		}
 
 		if (flags & NIX_RX_OFFLOAD_VLAN_STRIP_F) {
-- 
2.8.4


             reply	other threads:[~2019-12-10 12:08 UTC|newest]

Thread overview: 2+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2019-12-10 12:08 Nithin Dabilpuram [this message]
2020-01-13  7:40 ` Jerin Jacob

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20191210120844.50017-1-ndabilpuram@marvell.com \
    --to=ndabilpuram@marvell.com \
    --cc=apinski@marvell.com \
    --cc=dev@dpdk.org \
    --cc=jerinj@marvell.com \
    --cc=kirankumark@marvell.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).