[EXT] [PATCH v5 2/5] examples/l3fwd: split processing and send stages

[Shijith Thotton]

>
>Split packet processing from packet send stage, as send stage
>is not common for poll and event mode.
>
>Signed-off-by: Pavan Nikhilesh <pbhagavatula at marvell.com>

Acked-by: Shijith Thotton <sthotton at marvell.com>

>---
> examples/l3fwd/l3fwd_em_hlm.h      | 39 +++++++++++++++++++-----------
> examples/l3fwd/l3fwd_lpm_altivec.h | 25 ++++++++++++++++---
> examples/l3fwd/l3fwd_lpm_neon.h    | 35 ++++++++++++++++++++-------
> examples/l3fwd/l3fwd_lpm_sse.h     | 25 ++++++++++++++++---
> 4 files changed, 95 insertions(+), 29 deletions(-)
>
>diff --git a/examples/l3fwd/l3fwd_em_hlm.h b/examples/l3fwd/l3fwd_em_hlm.h
>index e76f2760b0..12b997e477 100644
>--- a/examples/l3fwd/l3fwd_em_hlm.h
>+++ b/examples/l3fwd/l3fwd_em_hlm.h
>@@ -177,16 +177,12 @@ em_get_dst_port(const struct lcore_conf *qconf, struct
>rte_mbuf *pkt,
> 	return portid;
> }
>
>-/*
>- * Buffer optimized handling of packets, invoked
>- * from main_loop.
>- */
> static inline void
>-l3fwd_em_send_packets(int nb_rx, struct rte_mbuf **pkts_burst,
>-		uint16_t portid, struct lcore_conf *qconf)
>+l3fwd_em_process_packets(int nb_rx, struct rte_mbuf **pkts_burst,
>+			 uint16_t *dst_port, uint16_t portid,
>+			 struct lcore_conf *qconf, const uint8_t do_step3)
> {
> 	int32_t i, j, pos;
>-	uint16_t dst_port[MAX_PKT_BURST];
>
> 	/*
> 	 * Send nb_rx - nb_rx % EM_HASH_LOOKUP_COUNT packets
>@@ -233,13 +229,30 @@ l3fwd_em_send_packets(int nb_rx, struct rte_mbuf
>**pkts_burst,
> 				dst_port[j + i] = em_get_dst_port(qconf,
> 						pkts_burst[j + i], portid);
> 		}
>+
>+		for (i = 0; i < EM_HASH_LOOKUP_COUNT && do_step3; i +=
>FWDSTEP)
>+			processx4_step3(&pkts_burst[j + i], &dst_port[j + i]);
> 	}
>
>-	for (; j < nb_rx; j++)
>+	for (; j < nb_rx; j++) {
> 		dst_port[j] = em_get_dst_port(qconf, pkts_burst[j], portid);
>+		if (do_step3)
>+			process_packet(pkts_burst[j], &pkts_burst[j]->port);
>+	}
>+}
>
>-	send_packets_multi(qconf, pkts_burst, dst_port, nb_rx);
>+/*
>+ * Buffer optimized handling of packets, invoked
>+ * from main_loop.
>+ */
>+static inline void
>+l3fwd_em_send_packets(int nb_rx, struct rte_mbuf **pkts_burst, uint16_t
>portid,
>+		      struct lcore_conf *qconf)
>+{
>+	uint16_t dst_port[MAX_PKT_BURST];
>
>+	l3fwd_em_process_packets(nb_rx, pkts_burst, dst_port, portid, qconf,
>0);
>+	send_packets_multi(qconf, pkts_burst, dst_port, nb_rx);
> }
>
> /*
>@@ -260,11 +273,8 @@ l3fwd_em_process_events(int nb_rx, struct rte_event
>**ev,
> 	 */
> 	int32_t n = RTE_ALIGN_FLOOR(nb_rx, EM_HASH_LOOKUP_COUNT);
>
>-	for (j = 0; j < EM_HASH_LOOKUP_COUNT && j < nb_rx; j++) {
>+	for (j = 0; j < nb_rx; j++)
> 		pkts_burst[j] = ev[j]->mbuf;
>-		rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[j],
>-					       struct rte_ether_hdr *) + 1);
>-	}
>
> 	for (j = 0; j < n; j += EM_HASH_LOOKUP_COUNT) {
>
>@@ -305,7 +315,8 @@ l3fwd_em_process_events(int nb_rx, struct rte_event
>**ev,
> 			}
> 			continue;
> 		}
>-		processx4_step3(&pkts_burst[j], &dst_port[j]);
>+		for (i = 0; i < EM_HASH_LOOKUP_COUNT; i += FWDSTEP)
>+			processx4_step3(&pkts_burst[j + i], &dst_port[j + i]);
>
> 		for (i = 0; i < EM_HASH_LOOKUP_COUNT; i++)
> 			pkts_burst[j + i]->port = dst_port[j + i];
>diff --git a/examples/l3fwd/l3fwd_lpm_altivec.h
>b/examples/l3fwd/l3fwd_lpm_altivec.h
>index 0c6852a7bb..adb82f1478 100644
>--- a/examples/l3fwd/l3fwd_lpm_altivec.h
>+++ b/examples/l3fwd/l3fwd_lpm_altivec.h
>@@ -96,11 +96,11 @@ processx4_step2(const struct lcore_conf *qconf,
>  * from main_loop.
>  */
> static inline void
>-l3fwd_lpm_send_packets(int nb_rx, struct rte_mbuf **pkts_burst,
>-			uint8_t portid, struct lcore_conf *qconf)
>+l3fwd_lpm_process_packets(int nb_rx, struct rte_mbuf **pkts_burst,
>+			  uint8_t portid, uint16_t *dst_port,
>+			  struct lcore_conf *qconf, const uint8_t do_step3)
> {
> 	int32_t j;
>-	uint16_t dst_port[MAX_PKT_BURST];
> 	__vector unsigned int dip[MAX_PKT_BURST / FWDSTEP];
> 	uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
> 	const int32_t k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
>@@ -114,22 +114,41 @@ l3fwd_lpm_send_packets(int nb_rx, struct rte_mbuf
>**pkts_burst,
> 				ipv4_flag[j / FWDSTEP],
> 				portid, &pkts_burst[j], &dst_port[j]);
>
>+	if (do_step3)
>+		for (j = 0; j != k; j += FWDSTEP)
>+			processx4_step3(&pkts_burst[j], &dst_port[j]);
>+
> 	/* Classify last up to 3 packets one by one */
> 	switch (nb_rx % FWDSTEP) {
> 	case 3:
> 		dst_port[j] = lpm_get_dst_port(qconf, pkts_burst[j], portid);
>+		if (do_step3)
>+			process_packet(pkts_burst[j], &dst_port[j]);
> 		j++;
> 		/* fall-through */
> 	case 2:
> 		dst_port[j] = lpm_get_dst_port(qconf, pkts_burst[j], portid);
>+		if (do_step3)
>+			process_packet(pkts_burst[j], &dst_port[j]);
> 		j++;
> 		/* fall-through */
> 	case 1:
> 		dst_port[j] = lpm_get_dst_port(qconf, pkts_burst[j], portid);
>+		if (do_step3)
>+			process_packet(pkts_burst[j], &dst_port[j]);
> 		j++;
> 		/* fall-through */
> 	}
>+}
>+
>+static inline void
>+l3fwd_lpm_send_packets(int nb_rx, struct rte_mbuf **pkts_burst, uint8_t
>portid,
>+		       struct lcore_conf *qconf)
>+{
>+	uint16_t dst_port[MAX_PKT_BURST];
>
>+	l3fwd_lpm_process_packets(nb_rx, pkts_burst, portid, dst_port, qconf,
>+				  0);
> 	send_packets_multi(qconf, pkts_burst, dst_port, nb_rx);
> }
>
>diff --git a/examples/l3fwd/l3fwd_lpm_neon.h
>b/examples/l3fwd/l3fwd_lpm_neon.h
>index 78ee83b76c..2a68c4c15e 100644
>--- a/examples/l3fwd/l3fwd_lpm_neon.h
>+++ b/examples/l3fwd/l3fwd_lpm_neon.h
>@@ -80,16 +80,12 @@ processx4_step2(const struct lcore_conf *qconf,
> 	}
> }
>
>-/*
>- * Buffer optimized handling of packets, invoked
>- * from main_loop.
>- */
> static inline void
>-l3fwd_lpm_send_packets(int nb_rx, struct rte_mbuf **pkts_burst,
>-			uint16_t portid, struct lcore_conf *qconf)
>+l3fwd_lpm_process_packets(int nb_rx, struct rte_mbuf **pkts_burst,
>+			  uint16_t portid, uint16_t *dst_port,
>+			  struct lcore_conf *qconf, const uint8_t do_step3)
> {
> 	int32_t i = 0, j = 0;
>-	uint16_t dst_port[MAX_PKT_BURST];
> 	int32x4_t dip;
> 	uint32_t ipv4_flag;
> 	const int32_t k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
>@@ -100,7 +96,6 @@ l3fwd_lpm_send_packets(int nb_rx, struct rte_mbuf
>**pkts_burst,
> 			rte_prefetch0(rte_pktmbuf_mtod(pkts_burst[i],
> 							void *));
> 		}
>-
> 		for (j = 0; j != k - FWDSTEP; j += FWDSTEP) {
> 			for (i = 0; i < FWDSTEP; i++) {
> 				rte_prefetch0(rte_pktmbuf_mtod(
>@@ -111,11 +106,15 @@ l3fwd_lpm_send_packets(int nb_rx, struct rte_mbuf
>**pkts_burst,
> 			processx4_step1(&pkts_burst[j], &dip, &ipv4_flag);
> 			processx4_step2(qconf, dip, ipv4_flag, portid,
> 					&pkts_burst[j], &dst_port[j]);
>+			if (do_step3)
>+				processx4_step3(&pkts_burst[j], &dst_port[j]);
> 		}
>
> 		processx4_step1(&pkts_burst[j], &dip, &ipv4_flag);
> 		processx4_step2(qconf, dip, ipv4_flag, portid, &pkts_burst[j],
> 				&dst_port[j]);
>+		if (do_step3)
>+			processx4_step3(&pkts_burst[j], &dst_port[j]);
>
> 		j += FWDSTEP;
> 	}
>@@ -138,26 +137,44 @@ l3fwd_lpm_send_packets(int nb_rx, struct rte_mbuf
>**pkts_burst,
> 							void *));
> 			j++;
> 		}
>-
> 		j -= m;
> 		/* Classify last up to 3 packets one by one */
> 		switch (m) {
> 		case 3:
> 			dst_port[j] = lpm_get_dst_port(qconf, pkts_burst[j],
> 						       portid);
>+			if (do_step3)
>+				process_packet(pkts_burst[j], &dst_port[j]);
> 			j++;
> 			/* fallthrough */
> 		case 2:
> 			dst_port[j] = lpm_get_dst_port(qconf, pkts_burst[j],
> 						       portid);
>+			if (do_step3)
>+				process_packet(pkts_burst[j], &dst_port[j]);
> 			j++;
> 			/* fallthrough */
> 		case 1:
> 			dst_port[j] = lpm_get_dst_port(qconf, pkts_burst[j],
> 						       portid);
>+			if (do_step3)
>+				process_packet(pkts_burst[j], &dst_port[j]);
> 		}
> 	}
>+}
>+
>+/*
>+ * Buffer optimized handling of packets, invoked
>+ * from main_loop.
>+ */
>+static inline void
>+l3fwd_lpm_send_packets(int nb_rx, struct rte_mbuf **pkts_burst, uint16_t
>portid,
>+		       struct lcore_conf *qconf)
>+{
>+	uint16_t dst_port[MAX_PKT_BURST];
>
>+	l3fwd_lpm_process_packets(nb_rx, pkts_burst, portid, dst_port, qconf,
>+				  0);
> 	send_packets_multi(qconf, pkts_burst, dst_port, nb_rx);
> }
>
>diff --git a/examples/l3fwd/l3fwd_lpm_sse.h b/examples/l3fwd/l3fwd_lpm_sse.h
>index 3f637a23d1..db15030320 100644
>--- a/examples/l3fwd/l3fwd_lpm_sse.h
>+++ b/examples/l3fwd/l3fwd_lpm_sse.h
>@@ -82,11 +82,11 @@ processx4_step2(const struct lcore_conf *qconf,
>  * from main_loop.
>  */
> static inline void
>-l3fwd_lpm_send_packets(int nb_rx, struct rte_mbuf **pkts_burst,
>-			uint16_t portid, struct lcore_conf *qconf)
>+l3fwd_lpm_process_packets(int nb_rx, struct rte_mbuf **pkts_burst,
>+			  uint16_t portid, uint16_t *dst_port,
>+			  struct lcore_conf *qconf, const uint8_t do_step3)
> {
> 	int32_t j;
>-	uint16_t dst_port[MAX_PKT_BURST];
> 	__m128i dip[MAX_PKT_BURST / FWDSTEP];
> 	uint32_t ipv4_flag[MAX_PKT_BURST / FWDSTEP];
> 	const int32_t k = RTE_ALIGN_FLOOR(nb_rx, FWDSTEP);
>@@ -99,21 +99,40 @@ l3fwd_lpm_send_packets(int nb_rx, struct rte_mbuf
>**pkts_burst,
> 		processx4_step2(qconf, dip[j / FWDSTEP],
> 				ipv4_flag[j / FWDSTEP], portid, &pkts_burst[j],
>&dst_port[j]);
>
>+	if (do_step3)
>+		for (j = 0; j != k; j += FWDSTEP)
>+			processx4_step3(&pkts_burst[j], &dst_port[j]);
>+
> 	/* Classify last up to 3 packets one by one */
> 	switch (nb_rx % FWDSTEP) {
> 	case 3:
> 		dst_port[j] = lpm_get_dst_port(qconf, pkts_burst[j], portid);
>+		if (do_step3)
>+			process_packet(pkts_burst[j], &dst_port[j]);
> 		j++;
> 		/* fall-through */
> 	case 2:
> 		dst_port[j] = lpm_get_dst_port(qconf, pkts_burst[j], portid);
>+		if (do_step3)
>+			process_packet(pkts_burst[j], &dst_port[j]);
> 		j++;
> 		/* fall-through */
> 	case 1:
> 		dst_port[j] = lpm_get_dst_port(qconf, pkts_burst[j], portid);
>+		if (do_step3)
>+			process_packet(pkts_burst[j], &dst_port[j]);
> 		j++;
> 	}
>+}
>+
>+static inline void
>+l3fwd_lpm_send_packets(int nb_rx, struct rte_mbuf **pkts_burst, uint16_t
>portid,
>+		       struct lcore_conf *qconf)
>+{
>+	uint16_t dst_port[MAX_PKT_BURST];
>
>+	l3fwd_lpm_process_packets(nb_rx, pkts_burst, portid, dst_port, qconf,
>+				  0);
> 	send_packets_multi(qconf, pkts_burst, dst_port, nb_rx);
> }
>
>--
>2.25.1