From: <pbhagavatula@marvell.com>
To: <jerinj@marvell.com>, David Christensen <drc@linux.vnet.ibm.com>,
"Ruifeng Wang" <ruifeng.wang@arm.com>,
Bruce Richardson <bruce.richardson@intel.com>,
Konstantin Ananyev <konstantin.v.ananyev@yandex.ru>
Cc: <dev@dpdk.org>, Pavan Nikhilesh <pbhagavatula@marvell.com>
Subject: [PATCH v2 3/5] examples/l3fwd: use lpm vector path for event vector
Date: Fri, 2 Sep 2022 14:48:31 +0530 [thread overview]
Message-ID: <20220902091833.9074-3-pbhagavatula@marvell.com> (raw)
In-Reply-To: <20220902091833.9074-1-pbhagavatula@marvell.com>
From: Pavan Nikhilesh <pbhagavatula@marvell.com>
Use lpm vector path to process event vector.
Signed-off-by: Pavan Nikhilesh <pbhagavatula@marvell.com>
---
examples/l3fwd/l3fwd_altivec.h | 29 ++++++++++++++
examples/l3fwd/l3fwd_event.h | 71 ++++++++++++++++++++++++++++++++++
examples/l3fwd/l3fwd_lpm.c | 38 ++++++++++--------
examples/l3fwd/l3fwd_neon.h | 45 +++++++++++++++++++++
examples/l3fwd/l3fwd_sse.h | 44 +++++++++++++++++++++
5 files changed, 211 insertions(+), 16 deletions(-)
diff --git a/examples/l3fwd/l3fwd_altivec.h b/examples/l3fwd/l3fwd_altivec.h
index 87018f5dbe..e45e138e59 100644
--- a/examples/l3fwd/l3fwd_altivec.h
+++ b/examples/l3fwd/l3fwd_altivec.h
@@ -222,4 +222,33 @@ send_packets_multi(struct lcore_conf *qconf, struct rte_mbuf **pkts_burst,
}
}
+static __rte_always_inline uint16_t
+process_dst_port(uint16_t *dst_ports, uint16_t nb_elem)
+{
+ uint16_t i = 0, res;
+
+ while (nb_elem > 7) {
+ __vector unsigned short dp1;
+ __vector unsigned short dp;
+
+ dp = (__vector unsigned short)vec_splats((short)dst_ports[0]);
+ dp1 = *((__vector unsigned short *)&dst_ports[i]);
+ res = vec_all_eq(dp1, dp);
+ if (!res)
+ return BAD_PORT;
+
+ nb_elem -= 8;
+ i += 8;
+ }
+
+ while (nb_elem) {
+ if (dst_ports[i] != dst_ports[0])
+ return BAD_PORT;
+ nb_elem--;
+ i++;
+ }
+
+ return dst_ports[0];
+}
+
#endif /* _L3FWD_ALTIVEC_H_ */
diff --git a/examples/l3fwd/l3fwd_event.h b/examples/l3fwd/l3fwd_event.h
index b93841a16f..3fe38aada0 100644
--- a/examples/l3fwd/l3fwd_event.h
+++ b/examples/l3fwd/l3fwd_event.h
@@ -82,6 +82,27 @@ struct l3fwd_event_resources {
uint64_t vector_tmo_ns;
};
+#if defined(RTE_ARCH_X86)
+#include "l3fwd_sse.h"
+#elif defined __ARM_NEON
+#include "l3fwd_neon.h"
+#elif defined(RTE_ARCH_PPC_64)
+#include "l3fwd_altivec.h"
+#else
+static inline uint16_t
+process_dst_port(uint16_t *dst_ports, uint16_t nb_elem)
+{
+ int i;
+
+ for (i = 0; i < nb_elem; i++) {
+ if (dst_ports[i] != dst_ports[0])
+ return BAD_PORT;
+ }
+
+ return dst_ports[0];
+}
+#endif
+
static inline void
event_vector_attr_validate(struct rte_event_vector *vec, struct rte_mbuf *mbuf)
{
@@ -103,7 +124,57 @@ event_vector_txq_set(struct rte_event_vector *vec, uint16_t txq)
}
}
+static inline uint16_t
+filter_bad_packets(struct rte_mbuf **mbufs, uint16_t *dst_port,
+ uint16_t nb_pkts)
+{
+ uint16_t *des_pos, free = 0;
+ struct rte_mbuf **pos;
+ int i;
+
+ /* Filter out and free bad packets */
+ for (i = 0; i < nb_pkts; i++) {
+ if (dst_port[i] == BAD_PORT) {
+ rte_pktmbuf_free(mbufs[i]);
+ if (!free) {
+ pos = &mbufs[i];
+ des_pos = &dst_port[i];
+ }
+ free++;
+ continue;
+ }
+
+ if (free) {
+ *pos = mbufs[i];
+ pos++;
+ *des_pos = dst_port[i];
+ des_pos++;
+ }
+ }
+ return nb_pkts - free;
+}
+
+static inline void
+process_event_vector(struct rte_event_vector *vec, uint16_t *dst_port)
+{
+ uint16_t port, i;
+
+ vec->nb_elem = filter_bad_packets(vec->mbufs, dst_port, vec->nb_elem);
+ /* Verify destination array */
+ port = process_dst_port(dst_port, vec->nb_elem);
+ if (port == BAD_PORT) {
+ vec->attr_valid = 0;
+ for (i = 0; i < vec->nb_elem; i++) {
+ vec->mbufs[i]->port = dst_port[i];
+ rte_event_eth_tx_adapter_txq_set(vec->mbufs[i], 0);
+ }
+ } else {
+ vec->attr_valid = 1;
+ vec->port = port;
+ vec->queue = 0;
+ }
+}
struct l3fwd_event_resources *l3fwd_get_eventdev_rsrc(void);
void l3fwd_event_resource_setup(struct rte_eth_conf *port_conf);
diff --git a/examples/l3fwd/l3fwd_lpm.c b/examples/l3fwd/l3fwd_lpm.c
index d1b850dd5b..1652b7c470 100644
--- a/examples/l3fwd/l3fwd_lpm.c
+++ b/examples/l3fwd/l3fwd_lpm.c
@@ -425,24 +425,27 @@ lpm_event_main_loop_tx_q_burst(__rte_unused void *dummy)
}
static __rte_always_inline void
-lpm_process_event_vector(struct rte_event_vector *vec, struct lcore_conf *lconf)
+lpm_process_event_vector(struct rte_event_vector *vec, struct lcore_conf *lconf,
+ uint16_t *dst_port)
{
struct rte_mbuf **mbufs = vec->mbufs;
int i;
- /* Process first packet to init vector attributes */
- lpm_process_event_pkt(lconf, mbufs[0]);
+#if defined RTE_ARCH_X86 || defined __ARM_NEON || defined RTE_ARCH_PPC_64
if (vec->attr_valid) {
- if (mbufs[0]->port != BAD_PORT)
- vec->port = mbufs[0]->port;
- else
- vec->attr_valid = 0;
+ l3fwd_lpm_process_packets(vec->nb_elem, mbufs, vec->port,
+ dst_port, lconf, 1);
+ } else {
+ for (i = 0; i < vec->nb_elem; i++)
+ l3fwd_lpm_process_packets(1, &mbufs[i], mbufs[i]->port,
+ &dst_port[i], lconf, 1);
}
+#else
+ for (i = 0; i < vec->nb_elem; i++)
+ dst_port[i] = lpm_process_event_pkt(lconf, mbufs[i]);
+#endif
- for (i = 1; i < vec->nb_elem; i++) {
- lpm_process_event_pkt(lconf, mbufs[i]);
- event_vector_attr_validate(vec, mbufs[i]);
- }
+ process_event_vector(vec, dst_port);
}
/* Same eventdev loop for single and burst of vector */
@@ -458,6 +461,7 @@ lpm_event_loop_vector(struct l3fwd_event_resources *evt_rsrc,
struct rte_event events[MAX_PKT_BURST];
int i, nb_enq = 0, nb_deq = 0;
struct lcore_conf *lconf;
+ uint16_t *dst_port_list;
unsigned int lcore_id;
if (event_p_id < 0)
@@ -465,7 +469,11 @@ lpm_event_loop_vector(struct l3fwd_event_resources *evt_rsrc,
lcore_id = rte_lcore_id();
lconf = &lcore_conf[lcore_id];
-
+ dst_port_list =
+ rte_zmalloc("", sizeof(uint16_t) * evt_rsrc->vector_size,
+ RTE_CACHE_LINE_SIZE);
+ if (dst_port_list == NULL)
+ return;
RTE_LOG(INFO, L3FWD, "entering %s on lcore %u\n", __func__, lcore_id);
while (!force_quit) {
@@ -483,10 +491,8 @@ lpm_event_loop_vector(struct l3fwd_event_resources *evt_rsrc,
events[i].op = RTE_EVENT_OP_FORWARD;
}
- lpm_process_event_vector(events[i].vec, lconf);
-
- if (flags & L3FWD_EVENT_TX_DIRECT)
- event_vector_txq_set(events[i].vec, 0);
+ lpm_process_event_vector(events[i].vec, lconf,
+ dst_port_list);
}
if (flags & L3FWD_EVENT_TX_ENQ) {
diff --git a/examples/l3fwd/l3fwd_neon.h b/examples/l3fwd/l3fwd_neon.h
index ce515e0bc4..4d98288707 100644
--- a/examples/l3fwd/l3fwd_neon.h
+++ b/examples/l3fwd/l3fwd_neon.h
@@ -194,4 +194,49 @@ send_packets_multi(struct lcore_conf *qconf, struct rte_mbuf **pkts_burst,
}
}
+static __rte_always_inline uint16_t
+process_dst_port(uint16_t *dst_ports, uint16_t nb_elem)
+{
+ uint16_t i = 0, res;
+
+#if defined(RTE_ARCH_ARM64)
+ while (nb_elem > 7) {
+ uint16x8_t dp = vdupq_n_u16(dst_ports[0]);
+ uint16x8_t dp1;
+
+ dp1 = vld1q_u16(&dst_ports[i]);
+ dp1 = vceqq_u16(dp1, dp);
+ res = vminvq_u16(dp1);
+ if (!res)
+ return BAD_PORT;
+
+ nb_elem -= 8;
+ i += 8;
+ }
+
+ while (nb_elem > 3) {
+ uint16x4_t dp = vdup_n_u16(dst_ports[0]);
+ uint16x4_t dp1;
+
+ dp1 = vld1_u16(&dst_ports[i]);
+ dp1 = vceq_u16(dp1, dp);
+ res = vminv_u16(dp1);
+ if (!res)
+ return BAD_PORT;
+
+ nb_elem -= 4;
+ i += 4;
+ }
+#endif
+
+ while (nb_elem) {
+ if (dst_ports[i] != dst_ports[0])
+ return BAD_PORT;
+ nb_elem--;
+ i++;
+ }
+
+ return dst_ports[0];
+}
+
#endif /* _L3FWD_NEON_H_ */
diff --git a/examples/l3fwd/l3fwd_sse.h b/examples/l3fwd/l3fwd_sse.h
index 0f0d0323a2..083729cdef 100644
--- a/examples/l3fwd/l3fwd_sse.h
+++ b/examples/l3fwd/l3fwd_sse.h
@@ -194,4 +194,48 @@ send_packets_multi(struct lcore_conf *qconf, struct rte_mbuf **pkts_burst,
}
}
+static __rte_always_inline uint16_t
+process_dst_port(uint16_t *dst_ports, uint16_t nb_elem)
+{
+ uint16_t i = 0, res;
+
+ while (nb_elem > 7) {
+ __m128i dp = _mm_set1_epi16(dst_ports[0]);
+ __m128i dp1;
+
+ dp1 = _mm_loadu_si128((__m128i *)&dst_ports[i]);
+ dp1 = _mm_cmpeq_epi16(dp1, dp);
+ res = _mm_movemask_epi8(dp1);
+ if (res != 0xFFFF)
+ return BAD_PORT;
+
+ nb_elem -= 8;
+ i += 8;
+ }
+
+ while (nb_elem > 3) {
+ __m128i dp = _mm_set1_epi16(dst_ports[0]);
+ __m128i dp1;
+
+ dp1 = _mm_loadu_si128((__m128i *)&dst_ports[i]);
+ dp1 = _mm_cmpeq_epi16(dp1, dp);
+ dp1 = _mm_unpacklo_epi16(dp1, dp1);
+ res = _mm_movemask_ps((__m128)dp1);
+ if (res != 0xF)
+ return BAD_PORT;
+
+ nb_elem -= 4;
+ i += 4;
+ }
+
+ while (nb_elem) {
+ if (dst_ports[i] != dst_ports[0])
+ return BAD_PORT;
+ nb_elem--;
+ i++;
+ }
+
+ return dst_ports[0];
+}
+
#endif /* _L3FWD_SSE_H_ */
--
2.25.1
next prev parent reply other threads:[~2022-09-02 9:19 UTC|newest]
Thread overview: 41+ messages / expand[flat|nested] mbox.gz Atom feed top
2022-08-29 9:44 [PATCH 1/5] examples/l3fwd: fix port group mask generation pbhagavatula
2022-08-29 9:44 ` [PATCH 2/5] examples/l3fwd: split processing and send stages pbhagavatula
2022-08-29 9:44 ` [PATCH 3/5] examples/l3fwd: use lpm vector path for event vector pbhagavatula
2022-08-29 9:44 ` [PATCH 4/5] examples/l3fwd: use em " pbhagavatula
2022-08-29 9:44 ` [PATCH 5/5] examples/l3fwd: fix event vector processing in fib pbhagavatula
2022-09-02 9:18 ` [PATCH v2 1/5] examples/l3fwd: fix port group mask generation pbhagavatula
2022-09-02 9:18 ` [PATCH v2 2/5] examples/l3fwd: split processing and send stages pbhagavatula
2022-09-02 9:18 ` pbhagavatula [this message]
2022-09-02 9:18 ` [PATCH v2 4/5] examples/l3fwd: fix event vector processing in fib pbhagavatula
2022-09-02 9:18 ` [PATCH v2 5/5] examples/l3fwd: use em vector path for event vector pbhagavatula
2022-09-08 18:33 ` [PATCH v2 1/5] examples/l3fwd: fix port group mask generation David Christensen
2022-09-09 5:56 ` [EXT] " Pavan Nikhilesh Bhagavatula
2022-09-11 18:12 ` [PATCH v3 " pbhagavatula
2022-09-11 18:12 ` [PATCH v3 2/5] examples/l3fwd: split processing and send stages pbhagavatula
2022-09-11 18:12 ` [PATCH v3 3/5] examples/l3fwd: use lpm vector path for event vector pbhagavatula
2022-09-11 18:12 ` [PATCH v3 4/5] examples/l3fwd: fix event vector processing in fib pbhagavatula
2022-10-07 20:03 ` [EXT] " Shijith Thotton
2022-09-11 18:12 ` [PATCH v3 5/5] examples/l3fwd: use em vector path for event vector pbhagavatula
2022-10-07 20:01 ` [EXT] " Shijith Thotton
2022-10-11 9:08 ` [PATCH v4 1/5] examples/l3fwd: fix port group mask generation pbhagavatula
2022-10-11 9:08 ` [PATCH v4 2/5] examples/l3fwd: split processing and send stages pbhagavatula
2022-10-11 9:08 ` [PATCH v4 3/5] examples/l3fwd: use lpm vector path for event vector pbhagavatula
2022-10-11 9:08 ` [PATCH v4 4/5] examples/l3fwd: fix event vector processing in fib pbhagavatula
2022-10-11 9:08 ` [PATCH v4 5/5] examples/l3fwd: use em vector path for event vector pbhagavatula
2022-10-11 10:12 ` [PATCH v5 1/5] examples/l3fwd: fix port group mask generation pbhagavatula
2022-10-11 10:12 ` [PATCH v5 2/5] examples/l3fwd: split processing and send stages pbhagavatula
2022-10-17 12:06 ` [EXT] " Shijith Thotton
2022-10-11 10:12 ` [PATCH v5 3/5] examples/l3fwd: use lpm vector path for event vector pbhagavatula
2022-10-17 12:06 ` [EXT] " Shijith Thotton
2022-10-11 10:12 ` [PATCH v5 4/5] examples/l3fwd: fix event vector processing in fib pbhagavatula
2022-10-17 12:06 ` [EXT] " Shijith Thotton
2022-10-11 10:12 ` [PATCH v5 5/5] examples/l3fwd: use em vector path for event vector pbhagavatula
2022-10-12 8:57 ` [EXT] " Shijith Thotton
2022-10-17 12:05 ` [EXT] [PATCH v5 1/5] examples/l3fwd: fix port group mask generation Shijith Thotton
2022-10-20 16:15 ` Pavan Nikhilesh Bhagavatula
2022-10-25 16:05 ` [PATCH v6 " pbhagavatula
2022-10-25 16:05 ` [PATCH v6 2/5] examples/l3fwd: split processing and send stages pbhagavatula
2022-10-25 16:05 ` [PATCH v6 3/5] examples/l3fwd: use lpm vector path for event vector pbhagavatula
2022-10-25 16:05 ` [PATCH v6 4/5] examples/l3fwd: fix event vector processing in fib pbhagavatula
2022-10-25 16:05 ` [PATCH v6 5/5] examples/l3fwd: use em vector path for event vector pbhagavatula
2022-10-31 14:52 ` [PATCH v6 1/5] examples/l3fwd: fix port group mask generation Thomas Monjalon
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=20220902091833.9074-3-pbhagavatula@marvell.com \
--to=pbhagavatula@marvell.com \
--cc=bruce.richardson@intel.com \
--cc=dev@dpdk.org \
--cc=drc@linux.vnet.ibm.com \
--cc=jerinj@marvell.com \
--cc=konstantin.v.ananyev@yandex.ru \
--cc=ruifeng.wang@arm.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).