Skip to contentSkip to conversation

fix note error

PATCH Applied Feifei Wang 2 participants 6 messages Jul 23, 2021 – Aug 10, 2021 Network drivers

lore
Share & tools
lore
#0 lore raw

cover fix note error

Fix drivers/net note error and do some optimization for i40e NEON path.

Feifei Wang (4):
drivers/net: remove redundant phrases drivers/net: fix note error for Rx vector net/i40e: reorder Rx NEON code for better readability net/i40e: change code order to reduce L1 cache misses

 drivers/net/fm10k/fm10k_rxtx_vec.c       |   6 +-
 drivers/net/i40e/i40e_rxtx_vec_altivec.c |  10 +--
 drivers/net/i40e/i40e_rxtx_vec_neon.c    | 101 ++++++++++-------------
 drivers/net/i40e/i40e_rxtx_vec_sse.c     |   6 +-
 drivers/net/iavf/iavf_rxtx_vec_sse.c     |  12 +--
 drivers/net/ice/ice_rxtx_vec_sse.c       |   6 +-
 drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c   |   6 +-
 7 files changed, 68 insertions(+), 79 deletions(-)
Signature · 1 line
2.25.1
#1 lore raw

PATCH 1/4 drivers/net: remove redundant phrases

For the note of Rx vec path,when extract and record EOP bit, the code note should be "as the count of dd bits doesn't care", remove the redundant "count".

fm10k:

Fixes 7092be8437bd ("fm10k: add vector Rx")

i40e-altive:

Fixes c3def6a8724c ("net/i40e: implement vector PMD for altivec")

i40e-neon:

Fixes ae0eb310f253 ("net/i40e: implement vector PMD for ARM")

i40e-sse:

Fixes 9ed94e5bb04e ("i40e: add vector Rx")

iavf:

Fixes 319c421f3890 ("net/avf: enable SSE Rx Tx")
Fixes 1162f5a0ef31 ("net/iavf: support flexible Rx descriptor in SSE path")

ice:

Fixes c68a52b8b38c ("net/ice: support vector SSE in Rx")

ixgbe:

Fixes cf4b4708a88a ("ixgbe: improve slow-path perf with vector scattered Rx")
---
 drivers/net/fm10k/fm10k_rxtx_vec.c       | 2 +-
 drivers/net/i40e/i40e_rxtx_vec_altivec.c | 2 +-
 drivers/net/i40e/i40e_rxtx_vec_neon.c    | 2 +-
 drivers/net/i40e/i40e_rxtx_vec_sse.c     | 2 +-
 drivers/net/iavf/iavf_rxtx_vec_sse.c     | 4 ++--
 drivers/net/ice/ice_rxtx_vec_sse.c       | 2 +-
 drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c   | 2 +-
 7 files changed, 8 insertions(+), 8 deletions(-)
drivers/net/fm10k/fm10k_rxtx_vec.c +1 -1
index 39e3cdac1f..cae5322d48 100644
@@ -544,7 +544,7 @@ fm10k_recv_raw_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
 			/* and with mask to extract bits, flipping 1-0 */
 			__m128i eop_bits = _mm_andnot_si128(staterr, eop_check);
 			/* the staterr values are not in order, as the count
-			 * count of dd bits doesn't care. However, for end of
+			 * of dd bits doesn't care. However, for end of
 			 * packet tracking, we do care, so shuffle. This also
 			 * compresses the 32-bit values to 8-bit
 			 */
drivers/net/i40e/i40e_rxtx_vec_altivec.c +1 -1
index 1ad74646d6..edaa462ac8 100644
@@ -398,7 +398,7 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 				(vector unsigned char)vec_nor(staterr, staterr),
 				(vector unsigned char)eop_check);
 			/* the staterr values are not in order, as the count
-			 * count of dd bits doesn't care. However, for end of
+			 * of dd bits doesn't care. However, for end of
 			 * packet tracking, we do care, so shuffle. This also
 			 * compresses the 32-bit values to 8-bit
 			 */
drivers/net/i40e/i40e_rxtx_vec_neon.c +1 -1
index 1f5539bda8..32336fdb80 100644
@@ -387,7 +387,7 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *__rte_restrict rxq,
 			eop_bits = vmvnq_u8(vreinterpretq_u8_u16(staterr));
 			eop_bits = vandq_u8(eop_bits, eop_check);
 			/* the staterr values are not in order, as the count
-			 * count of dd bits doesn't care. However, for end of
+			 * of dd bits doesn't care. However, for end of
 			 * packet tracking, we do care, so shuffle. This also
 			 * compresses the 32-bit values to 8-bit
 			 */
drivers/net/i40e/i40e_rxtx_vec_sse.c +1 -1
index bfa5aff48d..03a0320353 100644
@@ -557,7 +557,7 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 			/* and with mask to extract bits, flipping 1-0 */
 			__m128i eop_bits = _mm_andnot_si128(staterr, eop_check);
 			/* the staterr values are not in order, as the count
-			 * count of dd bits doesn't care. However, for end of
+			 * of dd bits doesn't care. However, for end of
 			 * packet tracking, we do care, so shuffle. This also
 			 * compresses the 32-bit values to 8-bit
 			 */
drivers/net/iavf/iavf_rxtx_vec_sse.c +2 -2
index bf87696fa4..b813d96ef4 100644
@@ -590,7 +590,7 @@ _recv_raw_pkts_vec(struct iavf_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 			/* and with mask to extract bits, flipping 1-0 */
 			__m128i eop_bits = _mm_andnot_si128(staterr, eop_check);
 			/* the staterr values are not in order, as the count
-			 * count of dd bits doesn't care. However, for end of
+			 * of dd bits doesn't care. However, for end of
 			 * packet tracking, we do care, so shuffle. This also
 			 * compresses the 32-bit values to 8-bit
 			 */
@@ -884,7 +884,7 @@ _recv_raw_pkts_vec_flex_rxd(struct iavf_rx_queue *rxq,
 			/* and with mask to extract bits, flipping 1-0 */
 			__m128i eop_bits = _mm_andnot_si128(staterr, eop_check);
 			/* the staterr values are not in order, as the count
-			 * count of dd bits doesn't care. However, for end of
+			 * of dd bits doesn't care. However, for end of
 			 * packet tracking, we do care, so shuffle. This also
 			 * compresses the 32-bit values to 8-bit
 			 */
drivers/net/ice/ice_rxtx_vec_sse.c +1 -1
index 673e44a243..5f7e13ee39 100644
@@ -545,7 +545,7 @@ _ice_recv_raw_pkts_vec(struct ice_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 			/* and with mask to extract bits, flipping 1-0 */
 			__m128i eop_bits = _mm_andnot_si128(staterr, eop_check);
 			/* the staterr values are not in order, as the count
-			 * count of dd bits doesn't care. However, for end of
+			 * of dd bits doesn't care. However, for end of
 			 * packet tracking, we do care, so shuffle. This also
 			 * compresses the 32-bit values to 8-bit
 			 */
drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c +1 -1
index 7610fd93db..3a3ef51172 100644
@@ -540,7 +540,7 @@ _recv_raw_pkts_vec(struct ixgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 			/* and with mask to extract bits, flipping 1-0 */
 			__m128i eop_bits = _mm_andnot_si128(staterr, eop_check);
 			/* the staterr values are not in order, as the count
-			 * count of dd bits doesn't care. However, for end of
+			 * of dd bits doesn't care. However, for end of
 			 * packet tracking, we do care, so shuffle. This also
 			 * compresses the 32-bit values to 8-bit
 			 */
Signature · 1 line
2.25.1
#2 lore raw

PATCH 2/4 drivers/net: fix note error for Rx vector

In reply to Feifei Wang

For the loop to process packets in Rx vector path, some notes for the code are wrong, fix these errors.

fm10k:

Fixes 7092be8437bd ("fm10k: add vector Rx")

i40e-altive:

Fixes c3def6a8724c ("net/i40e: implement vector PMD for altivec")

i40e-neon:

Fixes ae0eb310f253 ("net/i40e: implement vector PMD for ARM")

i40e-sse:

Fixes 9ed94e5bb04e ("i40e: add vector Rx")

iavf:

Fixes 319c421f3890 ("net/avf: enable SSE Rx Tx")
Fixes 1162f5a0ef31 ("net/iavf: support flexible Rx descriptor in SSE path")

ice:

Fixes c68a52b8b38c ("net/ice: support vector SSE in Rx")

ixgbe:

Fixes cf4b4708a88a ("ixgbe: improve slow-path perf with vector scattered Rx")
---
 drivers/net/fm10k/fm10k_rxtx_vec.c       | 4 ++--
 drivers/net/i40e/i40e_rxtx_vec_altivec.c | 8 ++++----
 drivers/net/i40e/i40e_rxtx_vec_neon.c    | 8 ++++----
 drivers/net/i40e/i40e_rxtx_vec_sse.c     | 4 ++--
 drivers/net/iavf/iavf_rxtx_vec_sse.c     | 8 ++++----
 drivers/net/ice/ice_rxtx_vec_sse.c       | 4 ++--
 drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c   | 4 ++--
 7 files changed, 20 insertions(+), 20 deletions(-)
drivers/net/fm10k/fm10k_rxtx_vec.c +2 -2
index cae5322d48..83af01dc2d 100644
@@ -472,7 +472,7 @@ fm10k_recv_raw_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
 		mbp1 = _mm_loadu_si128((__m128i *)&mbufp[pos]);
 
 		/* Read desc statuses backwards to avoid race condition */
-		/* A.1 load 4 pkts desc */
+		/* A.1 load desc[3] */
 		descs0[3] = _mm_loadu_si128((__m128i *)(rxdp + 3));
 		rte_compiler_barrier();
 
@@ -484,9 +484,9 @@ fm10k_recv_raw_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
 		mbp2 = _mm_loadu_si128((__m128i *)&mbufp[pos+2]);
 #endif
 
+		/* A.1 load desc[2-0] */
 		descs0[2] = _mm_loadu_si128((__m128i *)(rxdp + 2));
 		rte_compiler_barrier();
-		/* B.1 load 2 mbuf point */
 		descs0[1] = _mm_loadu_si128((__m128i *)(rxdp + 1));
 		rte_compiler_barrier();
 		descs0[0] = _mm_loadu_si128((__m128i *)(rxdp));
drivers/net/i40e/i40e_rxtx_vec_altivec.c +4 -4
index edaa462ac8..b99323992f 100644
@@ -281,22 +281,22 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 						  * in one XMM reg.
 						  */
 
-		/* B.1 load 1 mbuf point */
+		/* B.1 load 2 mbuf point */
 		mbp1 = *(vector unsigned long *)&sw_ring[pos];
 		/* Read desc statuses backwards to avoid race condition */
-		/* A.1 load 4 pkts desc */
+		/* A.1 load desc[3] */
 		descs[3] = *(vector unsigned long *)(rxdp + 3);
 		rte_compiler_barrier();
 
 		/* B.2 copy 2 mbuf point into rx_pkts  */
 		*(vector unsigned long *)&rx_pkts[pos] = mbp1;
 
-		/* B.1 load 1 mbuf point */
+		/* B.1 load 2 mbuf point */
 		mbp2 = *(vector unsigned long *)&sw_ring[pos + 2];
 
+		/* A.1 load desc[2-0] */
 		descs[2] = *(vector unsigned long *)(rxdp + 2);
 		rte_compiler_barrier();
-		/* B.1 load 2 mbuf point */
 		descs[1] = *(vector unsigned long *)(rxdp + 1);
 		rte_compiler_barrier();
 		descs[0] = *(vector unsigned long *)(rxdp);
drivers/net/i40e/i40e_rxtx_vec_neon.c +4 -4
index 32336fdb80..fb624a4882 100644
@@ -280,20 +280,20 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *__rte_restrict rxq,
 
 		int32x4_t len_shl = {0, 0, 0, PKTLEN_SHIFT};
 
-		/* B.1 load 1 mbuf point */
+		/* B.1 load 2 mbuf point */
 		mbp1 = vld1q_u64((uint64_t *)&sw_ring[pos]);
 		/* Read desc statuses backwards to avoid race condition */
-		/* A.1 load 4 pkts desc */
+		/* A.1 load desc[3] */
 		descs[3] =  vld1q_u64((uint64_t *)(rxdp + 3));
 
 		/* B.2 copy 2 mbuf point into rx_pkts  */
 		vst1q_u64((uint64_t *)&rx_pkts[pos], mbp1);
 
-		/* B.1 load 1 mbuf point */
+		/* B.1 load 2 mbuf point */
 		mbp2 = vld1q_u64((uint64_t *)&sw_ring[pos + 2]);
 
+		/* A.1 load desc[2-0] */
 		descs[2] =  vld1q_u64((uint64_t *)(rxdp + 2));
-		/* B.1 load 2 mbuf point */
 		descs[1] =  vld1q_u64((uint64_t *)(rxdp + 1));
 		descs[0] =  vld1q_u64((uint64_t *)(rxdp));
 
drivers/net/i40e/i40e_rxtx_vec_sse.c +2 -2
index 03a0320353..b235502db5 100644
@@ -462,7 +462,7 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 		/* B.1 load 2 (64 bit) or 4 (32 bit) mbuf points */
 		mbp1 = _mm_loadu_si128((__m128i *)&sw_ring[pos]);
 		/* Read desc statuses backwards to avoid race condition */
-		/* A.1 load 4 pkts desc */
+		/* A.1 load desc[3] */
 		descs[3] = _mm_loadu_si128((__m128i *)(rxdp + 3));
 		rte_compiler_barrier();
 
@@ -474,9 +474,9 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 		mbp2 = _mm_loadu_si128((__m128i *)&sw_ring[pos+2]);
 #endif
 
+		/* A.1 load desc[2-0] */
 		descs[2] = _mm_loadu_si128((__m128i *)(rxdp + 2));
 		rte_compiler_barrier();
-		/* B.1 load 2 mbuf point */
 		descs[1] = _mm_loadu_si128((__m128i *)(rxdp + 1));
 		rte_compiler_barrier();
 		descs[0] = _mm_loadu_si128((__m128i *)(rxdp));
drivers/net/iavf/iavf_rxtx_vec_sse.c +4 -4
index b813d96ef4..ee1e905525 100644
@@ -494,7 +494,7 @@ _recv_raw_pkts_vec(struct iavf_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 		/* B.1 load 2 (64 bit) or 4 (32 bit) mbuf points */
 		mbp1 = _mm_loadu_si128((__m128i *)&sw_ring[pos]);
 		/* Read desc statuses backwards to avoid race condition */
-		/* A.1 load 4 pkts desc */
+		/* A.1 load desc[3] */
 		descs[3] = _mm_loadu_si128((__m128i *)(rxdp + 3));
 		rte_compiler_barrier();
 
@@ -506,9 +506,9 @@ _recv_raw_pkts_vec(struct iavf_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 		mbp2 = _mm_loadu_si128((__m128i *)&sw_ring[pos + 2]);
 #endif
 
+		/* A.1 load desc[2-0] */
 		descs[2] = _mm_loadu_si128((__m128i *)(rxdp + 2));
 		rte_compiler_barrier();
-		/* B.1 load 2 mbuf point */
 		descs[1] = _mm_loadu_si128((__m128i *)(rxdp + 1));
 		rte_compiler_barrier();
 		descs[0] = _mm_loadu_si128((__m128i *)(rxdp));
@@ -755,7 +755,7 @@ _recv_raw_pkts_vec_flex_rxd(struct iavf_rx_queue *rxq,
 		/* B.1 load 2 (64 bit) or 4 (32 bit) mbuf points */
 		mbp1 = _mm_loadu_si128((__m128i *)&sw_ring[pos]);
 		/* Read desc statuses backwards to avoid race condition */
-		/* A.1 load 4 pkts desc */
+		/* A.1 load desc[3] */
 		descs[3] = _mm_loadu_si128((__m128i *)(rxdp + 3));
 		rte_compiler_barrier();
 
@@ -767,9 +767,9 @@ _recv_raw_pkts_vec_flex_rxd(struct iavf_rx_queue *rxq,
 		mbp2 = _mm_loadu_si128((__m128i *)&sw_ring[pos + 2]);
 #endif
 
+		/* A.1 load desc[2-0] */
 		descs[2] = _mm_loadu_si128((__m128i *)(rxdp + 2));
 		rte_compiler_barrier();
-		/* B.1 load 2 mbuf point */
 		descs[1] = _mm_loadu_si128((__m128i *)(rxdp + 1));
 		rte_compiler_barrier();
 		descs[0] = _mm_loadu_si128((__m128i *)(rxdp));
drivers/net/ice/ice_rxtx_vec_sse.c +2 -2
index 5f7e13ee39..653bd28b41 100644
@@ -416,7 +416,7 @@ _ice_recv_raw_pkts_vec(struct ice_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 		/* B.1 load 2 (64 bit) or 4 (32 bit) mbuf points */
 		mbp1 = _mm_loadu_si128((__m128i *)&sw_ring[pos]);
 		/* Read desc statuses backwards to avoid race condition */
-		/* A.1 load 4 pkts desc */
+		/* A.1 load desc[3] */
 		descs[3] = _mm_loadu_si128((__m128i *)(rxdp + 3));
 		rte_compiler_barrier();
 
@@ -428,9 +428,9 @@ _ice_recv_raw_pkts_vec(struct ice_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 		mbp2 = _mm_loadu_si128((__m128i *)&sw_ring[pos + 2]);
 #endif
 
+		/* A.1 load desc[2-0] */
 		descs[2] = _mm_loadu_si128((__m128i *)(rxdp + 2));
 		rte_compiler_barrier();
-		/* B.1 load 2 mbuf point */
 		descs[1] = _mm_loadu_si128((__m128i *)(rxdp + 1));
 		rte_compiler_barrier();
 		descs[0] = _mm_loadu_si128((__m128i *)(rxdp));
drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c +2 -2
index 3a3ef51172..1dea95e73b 100644
@@ -454,7 +454,7 @@ _recv_raw_pkts_vec(struct ixgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 		mbp1 = _mm_loadu_si128((__m128i *)&sw_ring[pos]);
 
 		/* Read desc statuses backwards to avoid race condition */
-		/* A.1 load 4 pkts desc */
+		/* A.1 load desc[3] */
 		descs[3] = _mm_loadu_si128((__m128i *)(rxdp + 3));
 		rte_compiler_barrier();
 
@@ -466,9 +466,9 @@ _recv_raw_pkts_vec(struct ixgbe_rx_queue *rxq, struct rte_mbuf **rx_pkts,
 		mbp2 = _mm_loadu_si128((__m128i *)&sw_ring[pos+2]);
 #endif
 
+		/* A.1 load desc[2-0] */
 		descs[2] = _mm_loadu_si128((__m128i *)(rxdp + 2));
 		rte_compiler_barrier();
-		/* B.1 load 2 mbuf point */
 		descs[1] = _mm_loadu_si128((__m128i *)(rxdp + 1));
 		rte_compiler_barrier();
 		descs[0] = _mm_loadu_si128((__m128i *)(rxdp));
Signature · 1 line
2.25.1
#3 lore raw

PATCH 3/4 net/i40e: reorder Rx NEON code for better readability

In reply to Feifei Wang

Rearrange the code in logical order for better readability and maintenance convenience in Rx NEON path.

No performance change with this patch in arm platform.

---
 drivers/net/i40e/i40e_rxtx_vec_neon.c | 99 ++++++++++++---------------
 1 file changed, 44 insertions(+), 55 deletions(-)
drivers/net/i40e/i40e_rxtx_vec_neon.c +44 -55
index fb624a4882..8f3188e910 100644
@@ -280,24 +280,18 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *__rte_restrict rxq,
 
 		int32x4_t len_shl = {0, 0, 0, PKTLEN_SHIFT};
 
-		/* B.1 load 2 mbuf point */
-		mbp1 = vld1q_u64((uint64_t *)&sw_ring[pos]);
-		/* Read desc statuses backwards to avoid race condition */
-		/* A.1 load desc[3] */
+		/* A.1 load desc[3-0] */
 		descs[3] =  vld1q_u64((uint64_t *)(rxdp + 3));
-
-		/* B.2 copy 2 mbuf point into rx_pkts  */
-		vst1q_u64((uint64_t *)&rx_pkts[pos], mbp1);
-
-		/* B.1 load 2 mbuf point */
-		mbp2 = vld1q_u64((uint64_t *)&sw_ring[pos + 2]);
-
-		/* A.1 load desc[2-0] */
 		descs[2] =  vld1q_u64((uint64_t *)(rxdp + 2));
 		descs[1] =  vld1q_u64((uint64_t *)(rxdp + 1));
 		descs[0] =  vld1q_u64((uint64_t *)(rxdp));
 
-		/* B.2 copy 2 mbuf point into rx_pkts  */
+		/* B.1 load 4 mbuf point */
+		mbp1 = vld1q_u64((uint64_t *)&sw_ring[pos]);
+		mbp2 = vld1q_u64((uint64_t *)&sw_ring[pos + 2]);
+
+		/* B.2 copy 4 mbuf point into rx_pkts  */
+		vst1q_u64((uint64_t *)&rx_pkts[pos], mbp1);
 		vst1q_u64((uint64_t *)&rx_pkts[pos + 2], mbp2);
 
 		if (split_packet) {
@@ -307,28 +301,9 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *__rte_restrict rxq,
 			rte_mbuf_prefetch_part2(rx_pkts[pos + 3]);
 		}
 
-		/* pkt 3,4 shift the pktlen field to be 16-bit aligned*/
-		uint32x4_t len3 = vshlq_u32(vreinterpretq_u32_u64(descs[3]),
-					    len_shl);
-		descs[3] = vreinterpretq_u64_u16(vsetq_lane_u16
-				(vgetq_lane_u16(vreinterpretq_u16_u32(len3), 7),
-				 vreinterpretq_u16_u64(descs[3]),
-				 7));
-		uint32x4_t len2 = vshlq_u32(vreinterpretq_u32_u64(descs[2]),
-					    len_shl);
-		descs[2] = vreinterpretq_u64_u16(vsetq_lane_u16
-				(vgetq_lane_u16(vreinterpretq_u16_u32(len2), 7),
-				 vreinterpretq_u16_u64(descs[2]),
-				 7));
-
-		/* D.1 pkt 3,4 convert format from desc to pktmbuf */
-		pkt_mb4 = vqtbl1q_u8(vreinterpretq_u8_u64(descs[3]), shuf_msk);
-		pkt_mb3 = vqtbl1q_u8(vreinterpretq_u8_u64(descs[2]), shuf_msk);
-
 		/* C.1 4=>2 filter staterr info only */
 		sterr_tmp2 = vzipq_u16(vreinterpretq_u16_u64(descs[1]),
 				       vreinterpretq_u16_u64(descs[3]));
-		/* C.1 4=>2 filter staterr info only */
 		sterr_tmp1 = vzipq_u16(vreinterpretq_u16_u64(descs[0]),
 				       vreinterpretq_u16_u64(descs[2]));
 
@@ -338,13 +313,19 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *__rte_restrict rxq,
 
 		desc_to_olflags_v(rxq, descs, &rx_pkts[pos]);
 
-		/* D.2 pkt 3,4 set in_port/nb_seg and remove crc */
-		tmp = vsubq_u16(vreinterpretq_u16_u8(pkt_mb4), crc_adjust);
-		pkt_mb4 = vreinterpretq_u8_u16(tmp);
-		tmp = vsubq_u16(vreinterpretq_u16_u8(pkt_mb3), crc_adjust);
-		pkt_mb3 = vreinterpretq_u8_u16(tmp);
-
-		/* pkt 1,2 shift the pktlen field to be 16-bit aligned*/
+		/* pkts shift the pktlen field to be 16-bit aligned*/
+		uint32x4_t len3 = vshlq_u32(vreinterpretq_u32_u64(descs[3]),
+					    len_shl);
+		descs[3] = vreinterpretq_u64_u16(vsetq_lane_u16
+				(vgetq_lane_u16(vreinterpretq_u16_u32(len3), 7),
+				 vreinterpretq_u16_u64(descs[3]),
+				 7));
+		uint32x4_t len2 = vshlq_u32(vreinterpretq_u32_u64(descs[2]),
+					    len_shl);
+		descs[2] = vreinterpretq_u64_u16(vsetq_lane_u16
+				(vgetq_lane_u16(vreinterpretq_u16_u32(len2), 7),
+				 vreinterpretq_u16_u64(descs[2]),
+				 7));
 		uint32x4_t len1 = vshlq_u32(vreinterpretq_u32_u64(descs[1]),
 					    len_shl);
 		descs[1] = vreinterpretq_u64_u16(vsetq_lane_u16
@@ -358,22 +339,38 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *__rte_restrict rxq,
 				 vreinterpretq_u16_u64(descs[0]),
 				 7));
 
-		/* D.1 pkt 1,2 convert format from desc to pktmbuf */
+		/* D.1 pkts convert format from desc to pktmbuf */
+		pkt_mb4 = vqtbl1q_u8(vreinterpretq_u8_u64(descs[3]), shuf_msk);
+		pkt_mb3 = vqtbl1q_u8(vreinterpretq_u8_u64(descs[2]), shuf_msk);
 		pkt_mb2 = vqtbl1q_u8(vreinterpretq_u8_u64(descs[1]), shuf_msk);
 		pkt_mb1 = vqtbl1q_u8(vreinterpretq_u8_u64(descs[0]), shuf_msk);
 
-		/* D.3 copy final 3,4 data to rx_pkts */
-		vst1q_u8((void *)&rx_pkts[pos + 3]->rx_descriptor_fields1,
-				 pkt_mb4);
-		vst1q_u8((void *)&rx_pkts[pos + 2]->rx_descriptor_fields1,
-				 pkt_mb3);
-
-		/* D.2 pkt 1,2 set in_port/nb_seg and remove crc */
+		/* D.2 pkts set in_port/nb_seg and remove crc */
+		tmp = vsubq_u16(vreinterpretq_u16_u8(pkt_mb4), crc_adjust);
+		pkt_mb4 = vreinterpretq_u8_u16(tmp);
+		tmp = vsubq_u16(vreinterpretq_u16_u8(pkt_mb3), crc_adjust);
+		pkt_mb3 = vreinterpretq_u8_u16(tmp);
 		tmp = vsubq_u16(vreinterpretq_u16_u8(pkt_mb2), crc_adjust);
 		pkt_mb2 = vreinterpretq_u8_u16(tmp);
 		tmp = vsubq_u16(vreinterpretq_u16_u8(pkt_mb1), crc_adjust);
 		pkt_mb1 = vreinterpretq_u8_u16(tmp);
 
+		/* D.3 copy final data to rx_pkts */
+		vst1q_u8((void *)&rx_pkts[pos + 3]->rx_descriptor_fields1,
+				pkt_mb4);
+		vst1q_u8((void *)&rx_pkts[pos + 2]->rx_descriptor_fields1,
+				pkt_mb3);
+		vst1q_u8((void *)&rx_pkts[pos + 1]->rx_descriptor_fields1,
+				pkt_mb2);
+		vst1q_u8((void *)&rx_pkts[pos]->rx_descriptor_fields1,
+				pkt_mb1);
+
+		desc_to_ptype_v(descs, &rx_pkts[pos], ptype_tbl);
+
+		if (likely(pos + RTE_I40E_DESCS_PER_LOOP < nb_pkts)) {
+			rte_prefetch_non_temporal(rxdp + RTE_I40E_DESCS_PER_LOOP);
+		}
+
 		/* C* extract and record EOP bit */
 		if (split_packet) {
 			uint8x16_t eop_shuf_mask = {
@@ -411,14 +408,6 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *__rte_restrict rxq,
 					    I40E_UINT16_BIT - 1));
 		stat = ~vgetq_lane_u64(vreinterpretq_u64_u16(staterr), 0);
 
-		rte_prefetch_non_temporal(rxdp + RTE_I40E_DESCS_PER_LOOP);
-
-		/* D.3 copy final 1,2 data to rx_pkts */
-		vst1q_u8((void *)&rx_pkts[pos + 1]->rx_descriptor_fields1,
-			 pkt_mb2);
-		vst1q_u8((void *)&rx_pkts[pos]->rx_descriptor_fields1,
-			 pkt_mb1);
-		desc_to_ptype_v(descs, &rx_pkts[pos], ptype_tbl);
 		/* C.4 calc avaialbe number of desc */
 		if (unlikely(stat == 0)) {
 			nb_pkts_recd += RTE_I40E_DESCS_PER_LOOP;
Signature · 1 line
2.25.1
#4 lore raw

PATCH 4/4 net/i40e: change code order to reduce L1 cache misses

In reply to Feifei Wang

For N1 platform, packet mbuf load and descs load are hot spots to limit the performance for "desc_to_ptype_v" and "desc_to_olflags_v" functions in i40e rx NEON path. This is because packet mbuf and descs are evicted from l1d-cache to l2d-cache.

To reduce l1d-cache-misses and improve the performance, change the code order and move "desc_to_ptype_v" and "desc_to_olflags_v" functions forward to the location, where packet mbuf and descs are just loaded.

Test Result:
dpdk:21.08-rc1
gcc-9
For n1sdp, the patch improves the performance by 1.8%. For thunderx2, no performance changes.

---
 drivers/net/i40e/i40e_rxtx_vec_neon.c | 24 ++++++++++++------------
 1 file changed, 12 insertions(+), 12 deletions(-)
drivers/net/i40e/i40e_rxtx_vec_neon.c +12 -12
index 8f3188e910..b2683fda60 100644
@@ -301,18 +301,6 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *__rte_restrict rxq,
 			rte_mbuf_prefetch_part2(rx_pkts[pos + 3]);
 		}
 
-		/* C.1 4=>2 filter staterr info only */
-		sterr_tmp2 = vzipq_u16(vreinterpretq_u16_u64(descs[1]),
-				       vreinterpretq_u16_u64(descs[3]));
-		sterr_tmp1 = vzipq_u16(vreinterpretq_u16_u64(descs[0]),
-				       vreinterpretq_u16_u64(descs[2]));
-
-		/* C.2 get 4 pkts staterr value  */
-		staterr = vzipq_u16(sterr_tmp1.val[1],
-				    sterr_tmp2.val[1]).val[0];
-
-		desc_to_olflags_v(rxq, descs, &rx_pkts[pos]);
-
 		/* pkts shift the pktlen field to be 16-bit aligned*/
 		uint32x4_t len3 = vshlq_u32(vreinterpretq_u32_u64(descs[3]),
 					    len_shl);
@@ -367,10 +355,22 @@ _recv_raw_pkts_vec(struct i40e_rx_queue *__rte_restrict rxq,
 
 		desc_to_ptype_v(descs, &rx_pkts[pos], ptype_tbl);
 
+		desc_to_olflags_v(rxq, descs, &rx_pkts[pos]);
+
 		if (likely(pos + RTE_I40E_DESCS_PER_LOOP < nb_pkts)) {
 			rte_prefetch_non_temporal(rxdp + RTE_I40E_DESCS_PER_LOOP);
 		}
 
+		/* C.1 4=>2 filter staterr info only */
+		sterr_tmp2 = vzipq_u16(vreinterpretq_u16_u64(descs[1]),
+				       vreinterpretq_u16_u64(descs[3]));
+		sterr_tmp1 = vzipq_u16(vreinterpretq_u16_u64(descs[0]),
+				       vreinterpretq_u16_u64(descs[2]));
+
+		/* C.2 get 4 pkts staterr value  */
+		staterr = vzipq_u16(sterr_tmp1.val[1],
+				    sterr_tmp2.val[1]).val[0];
+
 		/* C* extract and record EOP bit */
 		if (split_packet) {
 			uint8x16_t eop_shuf_mask = {
Signature · 1 line
2.25.1
Applied
#5 lore raw
In reply to Feifei Wang
Quoted 25 lines

-----Original Message-----
From: dev <[email protected]> On Behalf Of Feifei Wang
Sent: Friday, July 23, 2021 11:11 AM
Cc: [email protected]; [email protected]; Feifei Wang <[email protected]>
Subject: [dpdk-dev] [PATCH v1 0/4] fix note error

Fix drivers/net note error and do some optimization for i40e NEON path.

Feifei Wang (4):
drivers/net: remove redundant phrases drivers/net: fix note error for Rx vector net/i40e: reorder Rx NEON code for better readability net/i40e: change code order to reduce L1 cache misses

 drivers/net/fm10k/fm10k_rxtx_vec.c       |   6 +-
 drivers/net/i40e/i40e_rxtx_vec_altivec.c |  10 +--
 drivers/net/i40e/i40e_rxtx_vec_neon.c    | 101 ++++++++++-------------
 drivers/net/i40e/i40e_rxtx_vec_sse.c     |   6 +-
 drivers/net/iavf/iavf_rxtx_vec_sse.c     |  12 +--
 drivers/net/ice/ice_rxtx_vec_sse.c       |   6 +-
 drivers/net/ixgbe/ixgbe_rxtx_vec_sse.c   |   6 +-
 7 files changed, 68 insertions(+), 79 deletions(-)
Quoted 2 lines

--
2.25.1

Applied to dpdk-next-net-intel.

Thanks
Qi