[dpdk-dev] [PATCH 3/4] virtio: move SSE based Rx implementation to separate file

[Jerin Jacob <jerin.jacob@caviumnetworks.com>]

split out SSE instruction based virtio simple rx
implementation to a separate file

Signed-off-by: Jerin Jacob <jerin.jacob@caviumnetworks.com>
---
 drivers/net/virtio/virtio_rxtx_simple.c     | 166 +-------------------
 drivers/net/virtio/virtio_rxtx_simple_sse.h | 225 ++++++++++++++++++++++++++++
 2 files changed, 226 insertions(+), 165 deletions(-)
 create mode 100644 drivers/net/virtio/virtio_rxtx_simple_sse.h

diff --git a/drivers/net/virtio/virtio_rxtx_simple.c b/drivers/net/virtio/virtio_rxtx_simple.c
index 67430da..ca87605 100644
--- a/drivers/net/virtio/virtio_rxtx_simple.c
+++ b/drivers/net/virtio/virtio_rxtx_simple.c
@@ -130,171 +130,7 @@ virtio_rxq_rearm_vec(struct virtnet_rx *rxvq)
 }
 
 #ifdef RTE_MACHINE_CPUFLAG_SSSE3
-
-#include <tmmintrin.h>
-
-/* virtio vPMD receive routine, only accept(nb_pkts >= RTE_VIRTIO_DESC_PER_LOOP)
- *
- * This routine is for non-mergeable RX, one desc for each guest buffer.
- * This routine is based on the RX ring layout optimization. Each entry in the
- * avail ring points to the desc with the same index in the desc ring and this
- * will never be changed in the driver.
- *
- * - nb_pkts < RTE_VIRTIO_DESC_PER_LOOP, just return no packet
- */
-uint16_t
-virtio_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
-	uint16_t nb_pkts)
-{
-	struct virtnet_rx *rxvq = rx_queue;
-	struct virtqueue *vq = rxvq->vq;
-	uint16_t nb_used;
-	uint16_t desc_idx;
-	struct vring_used_elem *rused;
-	struct rte_mbuf **sw_ring;
-	struct rte_mbuf **sw_ring_end;
-	uint16_t nb_pkts_received;
-	__m128i shuf_msk1, shuf_msk2, len_adjust;
-
-	shuf_msk1 = _mm_set_epi8(
-		0xFF, 0xFF, 0xFF, 0xFF,
-		0xFF, 0xFF,		/* vlan tci */
-		5, 4,			/* dat len */
-		0xFF, 0xFF, 5, 4,	/* pkt len */
-		0xFF, 0xFF, 0xFF, 0xFF	/* packet type */
-
-	);
-
-	shuf_msk2 = _mm_set_epi8(
-		0xFF, 0xFF, 0xFF, 0xFF,
-		0xFF, 0xFF,		/* vlan tci */
-		13, 12,			/* dat len */
-		0xFF, 0xFF, 13, 12,	/* pkt len */
-		0xFF, 0xFF, 0xFF, 0xFF	/* packet type */
-	);
-
-	/* Subtract the header length.
-	*  In which case do we need the header length in used->len ?
-	*/
-	len_adjust = _mm_set_epi16(
-		0, 0,
-		0,
-		(uint16_t)-vq->hw->vtnet_hdr_size,
-		0, (uint16_t)-vq->hw->vtnet_hdr_size,
-		0, 0);
-
-	if (unlikely(nb_pkts < RTE_VIRTIO_DESC_PER_LOOP))
-		return 0;
-
-	nb_used = VIRTQUEUE_NUSED(vq);
-
-	rte_compiler_barrier();
-
-	if (unlikely(nb_used == 0))
-		return 0;
-
-	nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, RTE_VIRTIO_DESC_PER_LOOP);
-	nb_used = RTE_MIN(nb_used, nb_pkts);
-
-	desc_idx = (uint16_t)(vq->vq_used_cons_idx & (vq->vq_nentries - 1));
-	rused = &vq->vq_ring.used->ring[desc_idx];
-	sw_ring  = &vq->sw_ring[desc_idx];
-	sw_ring_end = &vq->sw_ring[vq->vq_nentries];
-
-	_mm_prefetch((const void *)rused, _MM_HINT_T0);
-
-	if (vq->vq_free_cnt >= RTE_VIRTIO_VPMD_RX_REARM_THRESH) {
-		virtio_rxq_rearm_vec(rxvq);
-		if (unlikely(virtqueue_kick_prepare(vq)))
-			virtqueue_notify(vq);
-	}
-
-	for (nb_pkts_received = 0;
-		nb_pkts_received < nb_used;) {
-		__m128i desc[RTE_VIRTIO_DESC_PER_LOOP / 2];
-		__m128i mbp[RTE_VIRTIO_DESC_PER_LOOP / 2];
-		__m128i pkt_mb[RTE_VIRTIO_DESC_PER_LOOP];
-
-		mbp[0] = _mm_loadu_si128((__m128i *)(sw_ring + 0));
-		desc[0] = _mm_loadu_si128((__m128i *)(rused + 0));
-		_mm_storeu_si128((__m128i *)&rx_pkts[0], mbp[0]);
-
-		mbp[1] = _mm_loadu_si128((__m128i *)(sw_ring + 2));
-		desc[1] = _mm_loadu_si128((__m128i *)(rused + 2));
-		_mm_storeu_si128((__m128i *)&rx_pkts[2], mbp[1]);
-
-		mbp[2] = _mm_loadu_si128((__m128i *)(sw_ring + 4));
-		desc[2] = _mm_loadu_si128((__m128i *)(rused + 4));
-		_mm_storeu_si128((__m128i *)&rx_pkts[4], mbp[2]);
-
-		mbp[3] = _mm_loadu_si128((__m128i *)(sw_ring + 6));
-		desc[3] = _mm_loadu_si128((__m128i *)(rused + 6));
-		_mm_storeu_si128((__m128i *)&rx_pkts[6], mbp[3]);
-
-		pkt_mb[1] = _mm_shuffle_epi8(desc[0], shuf_msk2);
-		pkt_mb[0] = _mm_shuffle_epi8(desc[0], shuf_msk1);
-		pkt_mb[1] = _mm_add_epi16(pkt_mb[1], len_adjust);
-		pkt_mb[0] = _mm_add_epi16(pkt_mb[0], len_adjust);
-		_mm_storeu_si128((void *)&rx_pkts[1]->rx_descriptor_fields1,
-			pkt_mb[1]);
-		_mm_storeu_si128((void *)&rx_pkts[0]->rx_descriptor_fields1,
-			pkt_mb[0]);
-
-		pkt_mb[3] = _mm_shuffle_epi8(desc[1], shuf_msk2);
-		pkt_mb[2] = _mm_shuffle_epi8(desc[1], shuf_msk1);
-		pkt_mb[3] = _mm_add_epi16(pkt_mb[3], len_adjust);
-		pkt_mb[2] = _mm_add_epi16(pkt_mb[2], len_adjust);
-		_mm_storeu_si128((void *)&rx_pkts[3]->rx_descriptor_fields1,
-			pkt_mb[3]);
-		_mm_storeu_si128((void *)&rx_pkts[2]->rx_descriptor_fields1,
-			pkt_mb[2]);
-
-		pkt_mb[5] = _mm_shuffle_epi8(desc[2], shuf_msk2);
-		pkt_mb[4] = _mm_shuffle_epi8(desc[2], shuf_msk1);
-		pkt_mb[5] = _mm_add_epi16(pkt_mb[5], len_adjust);
-		pkt_mb[4] = _mm_add_epi16(pkt_mb[4], len_adjust);
-		_mm_storeu_si128((void *)&rx_pkts[5]->rx_descriptor_fields1,
-			pkt_mb[5]);
-		_mm_storeu_si128((void *)&rx_pkts[4]->rx_descriptor_fields1,
-			pkt_mb[4]);
-
-		pkt_mb[7] = _mm_shuffle_epi8(desc[3], shuf_msk2);
-		pkt_mb[6] = _mm_shuffle_epi8(desc[3], shuf_msk1);
-		pkt_mb[7] = _mm_add_epi16(pkt_mb[7], len_adjust);
-		pkt_mb[6] = _mm_add_epi16(pkt_mb[6], len_adjust);
-		_mm_storeu_si128((void *)&rx_pkts[7]->rx_descriptor_fields1,
-			pkt_mb[7]);
-		_mm_storeu_si128((void *)&rx_pkts[6]->rx_descriptor_fields1,
-			pkt_mb[6]);
-
-		if (unlikely(nb_used <= RTE_VIRTIO_DESC_PER_LOOP)) {
-			if (sw_ring + nb_used <= sw_ring_end)
-				nb_pkts_received += nb_used;
-			else
-				nb_pkts_received += sw_ring_end - sw_ring;
-			break;
-		} else {
-			if (unlikely(sw_ring + RTE_VIRTIO_DESC_PER_LOOP >=
-				sw_ring_end)) {
-				nb_pkts_received += sw_ring_end - sw_ring;
-				break;
-			} else {
-				nb_pkts_received += RTE_VIRTIO_DESC_PER_LOOP;
-
-				rx_pkts += RTE_VIRTIO_DESC_PER_LOOP;
-				sw_ring += RTE_VIRTIO_DESC_PER_LOOP;
-				rused   += RTE_VIRTIO_DESC_PER_LOOP;
-				nb_used -= RTE_VIRTIO_DESC_PER_LOOP;
-			}
-		}
-	}
-
-	vq->vq_used_cons_idx += nb_pkts_received;
-	vq->vq_free_cnt += nb_pkts_received;
-	rxvq->stats.packets += nb_pkts_received;
-	return nb_pkts_received;
-}
-
+#include "virtio_rxtx_simple_sse.h"
 #endif
 
 #define VIRTIO_TX_FREE_THRESH 32
diff --git a/drivers/net/virtio/virtio_rxtx_simple_sse.h b/drivers/net/virtio/virtio_rxtx_simple_sse.h
new file mode 100644
index 0000000..4e8728f
--- /dev/null
+++ b/drivers/net/virtio/virtio_rxtx_simple_sse.h
@@ -0,0 +1,225 @@
+/*-
+ *   BSD LICENSE
+ *
+ *   Copyright(c) 2010-2015 Intel Corporation. All rights reserved.
+ *   All rights reserved.
+ *
+ *   Redistribution and use in source and binary forms, with or without
+ *   modification, are permitted provided that the following conditions
+ *   are met:
+ *
+ *     * Redistributions of source code must retain the above copyright
+ *       notice, this list of conditions and the following disclaimer.
+ *     * Redistributions in binary form must reproduce the above copyright
+ *       notice, this list of conditions and the following disclaimer in
+ *       the documentation and/or other materials provided with the
+ *       distribution.
+ *     * Neither the name of Intel Corporation nor the names of its
+ *       contributors may be used to endorse or promote products derived
+ *       from this software without specific prior written permission.
+ *
+ *   THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
+ *   "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
+ *   LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
+ *   A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
+ *   OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
+ *   SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
+ *   LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
+ *   DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
+ *   THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
+ *   (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
+ *   OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
+ */
+
+#include <stdint.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <errno.h>
+
+#include <tmmintrin.h>
+
+#include <rte_byteorder.h>
+#include <rte_branch_prediction.h>
+#include <rte_cycles.h>
+#include <rte_ether.h>
+#include <rte_ethdev.h>
+#include <rte_errno.h>
+#include <rte_memory.h>
+#include <rte_memzone.h>
+#include <rte_mempool.h>
+#include <rte_malloc.h>
+#include <rte_mbuf.h>
+#include <rte_prefetch.h>
+#include <rte_string_fns.h>
+
+#include "virtio_logs.h"
+#include "virtio_ethdev.h"
+#include "virtqueue.h"
+#include "virtio_rxtx.h"
+
+#define RTE_VIRTIO_VPMD_RX_BURST 32
+#define RTE_VIRTIO_DESC_PER_LOOP 8
+#define RTE_VIRTIO_VPMD_RX_REARM_THRESH RTE_VIRTIO_VPMD_RX_BURST
+
+/* virtio vPMD receive routine, only accept(nb_pkts >= RTE_VIRTIO_DESC_PER_LOOP)
+ *
+ * This routine is for non-mergeable RX, one desc for each guest buffer.
+ * This routine is based on the RX ring layout optimization. Each entry in the
+ * avail ring points to the desc with the same index in the desc ring and this
+ * will never be changed in the driver.
+ *
+ * - nb_pkts < RTE_VIRTIO_DESC_PER_LOOP, just return no packet
+ */
+uint16_t
+virtio_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
+	uint16_t nb_pkts)
+{
+	struct virtnet_rx *rxvq = rx_queue;
+	struct virtqueue *vq = rxvq->vq;
+	uint16_t nb_used;
+	uint16_t desc_idx;
+	struct vring_used_elem *rused;
+	struct rte_mbuf **sw_ring;
+	struct rte_mbuf **sw_ring_end;
+	uint16_t nb_pkts_received;
+	__m128i shuf_msk1, shuf_msk2, len_adjust;
+
+	shuf_msk1 = _mm_set_epi8(
+		0xFF, 0xFF, 0xFF, 0xFF,
+		0xFF, 0xFF,		/* vlan tci */
+		5, 4,			/* dat len */
+		0xFF, 0xFF, 5, 4,	/* pkt len */
+		0xFF, 0xFF, 0xFF, 0xFF	/* packet type */
+
+	);
+
+	shuf_msk2 = _mm_set_epi8(
+		0xFF, 0xFF, 0xFF, 0xFF,
+		0xFF, 0xFF,		/* vlan tci */
+		13, 12,			/* dat len */
+		0xFF, 0xFF, 13, 12,	/* pkt len */
+		0xFF, 0xFF, 0xFF, 0xFF	/* packet type */
+	);
+
+	/* Subtract the header length.
+	*  In which case do we need the header length in used->len ?
+	*/
+	len_adjust = _mm_set_epi16(
+		0, 0,
+		0,
+		(uint16_t)-vq->hw->vtnet_hdr_size,
+		0, (uint16_t)-vq->hw->vtnet_hdr_size,
+		0, 0);
+
+	if (unlikely(nb_pkts < RTE_VIRTIO_DESC_PER_LOOP))
+		return 0;
+
+	nb_used = VIRTQUEUE_NUSED(vq);
+
+	rte_compiler_barrier();
+
+	if (unlikely(nb_used == 0))
+		return 0;
+
+	nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, RTE_VIRTIO_DESC_PER_LOOP);
+	nb_used = RTE_MIN(nb_used, nb_pkts);
+
+	desc_idx = (uint16_t)(vq->vq_used_cons_idx & (vq->vq_nentries - 1));
+	rused = &vq->vq_ring.used->ring[desc_idx];
+	sw_ring  = &vq->sw_ring[desc_idx];
+	sw_ring_end = &vq->sw_ring[vq->vq_nentries];
+
+	_mm_prefetch((const void *)rused, _MM_HINT_T0);
+
+	if (vq->vq_free_cnt >= RTE_VIRTIO_VPMD_RX_REARM_THRESH) {
+		virtio_rxq_rearm_vec(rxvq);
+		if (unlikely(virtqueue_kick_prepare(vq)))
+			virtqueue_notify(vq);
+	}
+
+	for (nb_pkts_received = 0;
+		nb_pkts_received < nb_used;) {
+		__m128i desc[RTE_VIRTIO_DESC_PER_LOOP / 2];
+		__m128i mbp[RTE_VIRTIO_DESC_PER_LOOP / 2];
+		__m128i pkt_mb[RTE_VIRTIO_DESC_PER_LOOP];
+
+		mbp[0] = _mm_loadu_si128((__m128i *)(sw_ring + 0));
+		desc[0] = _mm_loadu_si128((__m128i *)(rused + 0));
+		_mm_storeu_si128((__m128i *)&rx_pkts[0], mbp[0]);
+
+		mbp[1] = _mm_loadu_si128((__m128i *)(sw_ring + 2));
+		desc[1] = _mm_loadu_si128((__m128i *)(rused + 2));
+		_mm_storeu_si128((__m128i *)&rx_pkts[2], mbp[1]);
+
+		mbp[2] = _mm_loadu_si128((__m128i *)(sw_ring + 4));
+		desc[2] = _mm_loadu_si128((__m128i *)(rused + 4));
+		_mm_storeu_si128((__m128i *)&rx_pkts[4], mbp[2]);
+
+		mbp[3] = _mm_loadu_si128((__m128i *)(sw_ring + 6));
+		desc[3] = _mm_loadu_si128((__m128i *)(rused + 6));
+		_mm_storeu_si128((__m128i *)&rx_pkts[6], mbp[3]);
+
+		pkt_mb[1] = _mm_shuffle_epi8(desc[0], shuf_msk2);
+		pkt_mb[0] = _mm_shuffle_epi8(desc[0], shuf_msk1);
+		pkt_mb[1] = _mm_add_epi16(pkt_mb[1], len_adjust);
+		pkt_mb[0] = _mm_add_epi16(pkt_mb[0], len_adjust);
+		_mm_storeu_si128((void *)&rx_pkts[1]->rx_descriptor_fields1,
+			pkt_mb[1]);
+		_mm_storeu_si128((void *)&rx_pkts[0]->rx_descriptor_fields1,
+			pkt_mb[0]);
+
+		pkt_mb[3] = _mm_shuffle_epi8(desc[1], shuf_msk2);
+		pkt_mb[2] = _mm_shuffle_epi8(desc[1], shuf_msk1);
+		pkt_mb[3] = _mm_add_epi16(pkt_mb[3], len_adjust);
+		pkt_mb[2] = _mm_add_epi16(pkt_mb[2], len_adjust);
+		_mm_storeu_si128((void *)&rx_pkts[3]->rx_descriptor_fields1,
+			pkt_mb[3]);
+		_mm_storeu_si128((void *)&rx_pkts[2]->rx_descriptor_fields1,
+			pkt_mb[2]);
+
+		pkt_mb[5] = _mm_shuffle_epi8(desc[2], shuf_msk2);
+		pkt_mb[4] = _mm_shuffle_epi8(desc[2], shuf_msk1);
+		pkt_mb[5] = _mm_add_epi16(pkt_mb[5], len_adjust);
+		pkt_mb[4] = _mm_add_epi16(pkt_mb[4], len_adjust);
+		_mm_storeu_si128((void *)&rx_pkts[5]->rx_descriptor_fields1,
+			pkt_mb[5]);
+		_mm_storeu_si128((void *)&rx_pkts[4]->rx_descriptor_fields1,
+			pkt_mb[4]);
+
+		pkt_mb[7] = _mm_shuffle_epi8(desc[3], shuf_msk2);
+		pkt_mb[6] = _mm_shuffle_epi8(desc[3], shuf_msk1);
+		pkt_mb[7] = _mm_add_epi16(pkt_mb[7], len_adjust);
+		pkt_mb[6] = _mm_add_epi16(pkt_mb[6], len_adjust);
+		_mm_storeu_si128((void *)&rx_pkts[7]->rx_descriptor_fields1,
+			pkt_mb[7]);
+		_mm_storeu_si128((void *)&rx_pkts[6]->rx_descriptor_fields1,
+			pkt_mb[6]);
+
+		if (unlikely(nb_used <= RTE_VIRTIO_DESC_PER_LOOP)) {
+			if (sw_ring + nb_used <= sw_ring_end)
+				nb_pkts_received += nb_used;
+			else
+				nb_pkts_received += sw_ring_end - sw_ring;
+			break;
+		} else {
+			if (unlikely(sw_ring + RTE_VIRTIO_DESC_PER_LOOP >=
+				sw_ring_end)) {
+				nb_pkts_received += sw_ring_end - sw_ring;
+				break;
+			} else {
+				nb_pkts_received += RTE_VIRTIO_DESC_PER_LOOP;
+
+				rx_pkts += RTE_VIRTIO_DESC_PER_LOOP;
+				sw_ring += RTE_VIRTIO_DESC_PER_LOOP;
+				rused   += RTE_VIRTIO_DESC_PER_LOOP;
+				nb_used -= RTE_VIRTIO_DESC_PER_LOOP;
+			}
+		}
+	}
+
+	vq->vq_used_cons_idx += nb_pkts_received;
+	vq->vq_free_cnt += nb_pkts_received;
+	rxvq->stats.packets += nb_pkts_received;
+	return nb_pkts_received;
+}
-- 
2.5.5