The limitation of burst size in vector rx was removed, since it should
retrieve as much received packets as possible. And also the scattered
receive path should use a wrapper function to achieve the goal of
burst maximizing. And do some code cleaning for vector rx path.
Bugzilla ID: 516
Fixes: af0c246a3800 ("net/iavf: enable AVX2 for iavf")
Fixes: 319c421f3890 ("net/avf: enable SSE Rx Tx")
Fixes: 1162f5a0ef31 ("net/iavf: support flexible Rx descriptor in SSE path")
Fixes: 5b6e8859081d ("net/iavf: support flexible Rx descriptor in AVX path")
Signed-off-by: Jeff Guo <jia....@intel.com>
---
drivers/net/iavf/iavf_rxtx.h | 1 +
drivers/net/iavf/iavf_rxtx_vec_avx2.c | 78 ++++++++---------
drivers/net/iavf/iavf_rxtx_vec_sse.c | 119 ++++++++++++++++++--------
3 files changed, 121 insertions(+), 77 deletions(-)
diff --git a/drivers/net/iavf/iavf_rxtx.h b/drivers/net/iavf/iavf_rxtx.h
index 59625a979..f71f9fbdb 100644
--- a/drivers/net/iavf/iavf_rxtx.h
+++ b/drivers/net/iavf/iavf_rxtx.h
@@ -21,6 +21,7 @@
#define IAVF_VPMD_TX_MAX_BURST 32
#define IAVF_RXQ_REARM_THRESH 32
#define IAVF_VPMD_DESCS_PER_LOOP 4
+#define IAVF_VPMD_DESCS_PER_LOOP_AVX 8
#define IAVF_VPMD_TX_MAX_FREE_BUF 64
#define IAVF_NO_VECTOR_FLAGS ( \
diff --git a/drivers/net/iavf/iavf_rxtx_vec_avx2.c
b/drivers/net/iavf/iavf_rxtx_vec_avx2.c
index e5e0fd309..c6ca5a4a8 100644
--- a/drivers/net/iavf/iavf_rxtx_vec_avx2.c
+++ b/drivers/net/iavf/iavf_rxtx_vec_avx2.c
@@ -29,7 +29,7 @@ iavf_rxq_rearm(struct iavf_rx_queue *rxq)
__m128i dma_addr0;
dma_addr0 = _mm_setzero_si128();
- for (i = 0; i < IAVF_VPMD_DESCS_PER_LOOP; i++) {
+ for (i = 0; i < IAVF_VPMD_DESCS_PER_LOOP_AVX; i++) {
rxp[i] = &rxq->fake_mbuf;
_mm_store_si128((__m128i *)&rxdp[i].read,
dma_addr0);
@@ -134,13 +134,19 @@ iavf_rxq_rearm(struct iavf_rx_queue *rxq)
#define PKTLEN_SHIFT 10
+/**
+ * vPMD raw receive routine,
+ * only accept(nb_pkts >= IAVF_VPMD_DESCS_PER_LOOP_AVX)
+ *
+ * Notice:
+ * - nb_pkts < IAVF_VPMD_DESCS_PER_LOOP_AVX, just return no packet
+ * - floor align nb_pkts to a IAVF_VPMD_DESCS_PER_LOOP_AVX power-of-two
+ */
static inline uint16_t
_iavf_recv_raw_pkts_vec_avx2(struct iavf_rx_queue *rxq,
struct rte_mbuf **rx_pkts,
uint16_t nb_pkts, uint8_t *split_packet)
{
-#define IAVF_DESCS_PER_LOOP_AVX 8
-
/* const uint32_t *ptype_tbl = rxq->vsi->adapter->ptype_tbl; */
const uint32_t *type_table = rxq->vsi->adapter->ptype_tbl;
@@ -153,8 +159,8 @@ _iavf_recv_raw_pkts_vec_avx2(struct iavf_rx_queue *rxq,
rte_prefetch0(rxdp);
- /* nb_pkts has to be floor-aligned to IAVF_DESCS_PER_LOOP_AVX */
- nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, IAVF_DESCS_PER_LOOP_AVX);
+ /* nb_pkts has to be floor-aligned to IAVF_VPMD_DESCS_PER_LOOP_AVX */
+ nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, IAVF_VPMD_DESCS_PER_LOOP_AVX);
/* See if we need to rearm the RX queue - gives the prefetch a bit
* of time to act
@@ -297,8 +303,8 @@ _iavf_recv_raw_pkts_vec_avx2(struct iavf_rx_queue *rxq,
uint16_t i, received;
for (i = 0, received = 0; i < nb_pkts;
- i += IAVF_DESCS_PER_LOOP_AVX,
- rxdp += IAVF_DESCS_PER_LOOP_AVX) {
+ i += IAVF_VPMD_DESCS_PER_LOOP_AVX,
+ rxdp += IAVF_VPMD_DESCS_PER_LOOP_AVX) {
/* step 1, copy over 8 mbuf pointers to rx_pkts array */
_mm256_storeu_si256((void *)&rx_pkts[i],
_mm256_loadu_si256((void *)&sw_ring[i]));
@@ -368,7 +374,7 @@ _iavf_recv_raw_pkts_vec_avx2(struct iavf_rx_queue *rxq,
if (split_packet) {
int j;
- for (j = 0; j < IAVF_DESCS_PER_LOOP_AVX; j++)
+ for (j = 0; j < IAVF_VPMD_DESCS_PER_LOOP_AVX; j++)
rte_mbuf_prefetch_part2(rx_pkts[i + j]);
}
@@ -583,7 +589,7 @@ _iavf_recv_raw_pkts_vec_avx2(struct iavf_rx_queue *rxq,
split_bits = _mm_shuffle_epi8(split_bits, eop_shuffle);
*(uint64_t *)split_packet =
_mm_cvtsi128_si64(split_bits);
- split_packet += IAVF_DESCS_PER_LOOP_AVX;
+ split_packet += IAVF_VPMD_DESCS_PER_LOOP_AVX;
}
/* perform dd_check */
@@ -599,7 +605,7 @@ _iavf_recv_raw_pkts_vec_avx2(struct iavf_rx_queue *rxq,
(_mm_cvtsi128_si64
(_mm256_castsi256_si128(status0_7)));
received += burst;
- if (burst != IAVF_DESCS_PER_LOOP_AVX)
+ if (burst != IAVF_VPMD_DESCS_PER_LOOP_AVX)
break;
}
@@ -633,13 +639,19 @@ flex_rxd_to_fdir_flags_vec_avx2(const __m256i fdir_id0_7)
return fdir_flags;
}
+/**
+ * vPMD raw receive routine for flex RxD,
+ * only accept(nb_pkts >= IAVF_VPMD_DESCS_PER_LOOP_AVX)
+ *
+ * Notice:
+ * - nb_pkts < IAVF_VPMD_DESCS_PER_LOOP_AVX, just return no packet
+ * - floor align nb_pkts to a IAVF_VPMD_DESCS_PER_LOOP_AVX power-of-two
+ */
static inline uint16_t
_iavf_recv_raw_pkts_vec_avx2_flex_rxd(struct iavf_rx_queue *rxq,
struct rte_mbuf **rx_pkts,
uint16_t nb_pkts, uint8_t *split_packet)
{
-#define IAVF_DESCS_PER_LOOP_AVX 8
-
const uint32_t *type_table = rxq->vsi->adapter->ptype_tbl;
const __m256i mbuf_init = _mm256_set_epi64x(0, 0,
@@ -650,8 +662,8 @@ _iavf_recv_raw_pkts_vec_avx2_flex_rxd(struct iavf_rx_queue
*rxq,
rte_prefetch0(rxdp);
- /* nb_pkts has to be floor-aligned to IAVF_DESCS_PER_LOOP_AVX */
- nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, IAVF_DESCS_PER_LOOP_AVX);
+ /* nb_pkts has to be floor-aligned to IAVF_VPMD_DESCS_PER_LOOP_AVX */
+ nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, IAVF_VPMD_DESCS_PER_LOOP_AVX);
/* See if we need to rearm the RX queue - gives the prefetch a bit
* of time to act
@@ -794,8 +806,8 @@ _iavf_recv_raw_pkts_vec_avx2_flex_rxd(struct iavf_rx_queue
*rxq,
uint16_t i, received;
for (i = 0, received = 0; i < nb_pkts;
- i += IAVF_DESCS_PER_LOOP_AVX,
- rxdp += IAVF_DESCS_PER_LOOP_AVX) {
+ i += IAVF_VPMD_DESCS_PER_LOOP_AVX,
+ rxdp += IAVF_VPMD_DESCS_PER_LOOP_AVX) {
/* step 1, copy over 8 mbuf pointers to rx_pkts array */
_mm256_storeu_si256((void *)&rx_pkts[i],
_mm256_loadu_si256((void *)&sw_ring[i]));
@@ -851,7 +863,7 @@ _iavf_recv_raw_pkts_vec_avx2_flex_rxd(struct iavf_rx_queue
*rxq,
if (split_packet) {
int j;
- for (j = 0; j < IAVF_DESCS_PER_LOOP_AVX; j++)
+ for (j = 0; j < IAVF_VPMD_DESCS_PER_LOOP_AVX; j++)
rte_mbuf_prefetch_part2(rx_pkts[i + j]);
}
@@ -1193,7 +1205,7 @@ _iavf_recv_raw_pkts_vec_avx2_flex_rxd(struct
iavf_rx_queue *rxq,
split_bits = _mm_shuffle_epi8(split_bits, eop_shuffle);
*(uint64_t *)split_packet =
_mm_cvtsi128_si64(split_bits);
- split_packet += IAVF_DESCS_PER_LOOP_AVX;
+ split_packet += IAVF_VPMD_DESCS_PER_LOOP_AVX;
}
/* perform dd_check */
@@ -1209,7 +1221,7 @@ _iavf_recv_raw_pkts_vec_avx2_flex_rxd(struct
iavf_rx_queue *rxq,
(_mm_cvtsi128_si64
(_mm256_castsi256_si128(status0_7)));
received += burst;
- if (burst != IAVF_DESCS_PER_LOOP_AVX)
+ if (burst != IAVF_VPMD_DESCS_PER_LOOP_AVX)
break;
}
@@ -1224,10 +1236,6 @@ _iavf_recv_raw_pkts_vec_avx2_flex_rxd(struct
iavf_rx_queue *rxq,
return received;
}
-/**
- * Notice:
- * - nb_pkts < IAVF_DESCS_PER_LOOP, just return no packet
- */
uint16_t
iavf_recv_pkts_vec_avx2(void *rx_queue, struct rte_mbuf **rx_pkts,
uint16_t nb_pkts)
@@ -1235,10 +1243,6 @@ iavf_recv_pkts_vec_avx2(void *rx_queue, struct rte_mbuf
**rx_pkts,
return _iavf_recv_raw_pkts_vec_avx2(rx_queue, rx_pkts, nb_pkts, NULL);
}
-/**
- * Notice:
- * - nb_pkts < IAVF_DESCS_PER_LOOP, just return no packet
- */
uint16_t
iavf_recv_pkts_vec_avx2_flex_rxd(void *rx_queue, struct rte_mbuf **rx_pkts,
uint16_t nb_pkts)
@@ -1249,8 +1253,6 @@ iavf_recv_pkts_vec_avx2_flex_rxd(void *rx_queue, struct
rte_mbuf **rx_pkts,
/**
* vPMD receive routine that reassembles single burst of 32 scattered packets
- * Notice:
- * - nb_pkts < IAVF_DESCS_PER_LOOP, just return no packet
*/
static uint16_t
iavf_recv_scattered_burst_vec_avx2(void *rx_queue, struct rte_mbuf **rx_pkts,
@@ -1259,6 +1261,9 @@ iavf_recv_scattered_burst_vec_avx2(void *rx_queue, struct
rte_mbuf **rx_pkts,
struct iavf_rx_queue *rxq = rx_queue;
uint8_t split_flags[IAVF_VPMD_RX_MAX_BURST] = {0};
+ /* split_flags only can support max of IAVF_VPMD_RX_MAX_BURST */
+ nb_pkts = RTE_MIN(nb_pkts, IAVF_VPMD_RX_MAX_BURST);
+
/* get some new buffers */
uint16_t nb_bufs = _iavf_recv_raw_pkts_vec_avx2(rxq, rx_pkts, nb_pkts,
split_flags);
@@ -1290,9 +1295,6 @@ iavf_recv_scattered_burst_vec_avx2(void *rx_queue, struct
rte_mbuf **rx_pkts,
/**
* vPMD receive routine that reassembles scattered packets.
- * Main receive routine that can handle arbitrary burst sizes
- * Notice:
- * - nb_pkts < IAVF_DESCS_PER_LOOP, just return no packet
*/
uint16_t
iavf_recv_scattered_pkts_vec_avx2(void *rx_queue, struct rte_mbuf **rx_pkts,
@@ -1313,10 +1315,8 @@ iavf_recv_scattered_pkts_vec_avx2(void *rx_queue, struct
rte_mbuf **rx_pkts,
}
/**
- * vPMD receive routine that reassembles single burst of
- * 32 scattered packets for flex RxD
- * Notice:
- * - nb_pkts < IAVF_DESCS_PER_LOOP, just return no packet
+ * vPMD receive routine that reassembles single burst of 32 scattered packets
+ * for flex RxD
*/
static uint16_t
iavf_recv_scattered_burst_vec_avx2_flex_rxd(void *rx_queue,
@@ -1326,6 +1326,9 @@ iavf_recv_scattered_burst_vec_avx2_flex_rxd(void
*rx_queue,
struct iavf_rx_queue *rxq = rx_queue;
uint8_t split_flags[IAVF_VPMD_RX_MAX_BURST] = {0};
+ /* split_flags only can support max of IAVF_VPMD_RX_MAX_BURST */
+ nb_pkts = RTE_MIN(nb_pkts, IAVF_VPMD_RX_MAX_BURST);
+
/* get some new buffers */
uint16_t nb_bufs = _iavf_recv_raw_pkts_vec_avx2_flex_rxd(rxq,
rx_pkts, nb_pkts, split_flags);
@@ -1357,9 +1360,6 @@ iavf_recv_scattered_burst_vec_avx2_flex_rxd(void
*rx_queue,
/**
* vPMD receive routine that reassembles scattered packets for flex RxD.
- * Main receive routine that can handle arbitrary burst sizes
- * Notice:
- * - nb_pkts < IAVF_DESCS_PER_LOOP, just return no packet
*/
uint16_t
iavf_recv_scattered_pkts_vec_avx2_flex_rxd(void *rx_queue,
diff --git a/drivers/net/iavf/iavf_rxtx_vec_sse.c
b/drivers/net/iavf/iavf_rxtx_vec_sse.c
index 85c5bd4af..b5362ecf3 100644
--- a/drivers/net/iavf/iavf_rxtx_vec_sse.c
+++ b/drivers/net/iavf/iavf_rxtx_vec_sse.c
@@ -379,10 +379,12 @@ flex_desc_to_ptype_v(__m128i descs[4], struct rte_mbuf
**rx_pkts,
rx_pkts[3]->packet_type = type_table[_mm_extract_epi16(ptype_all, 7)];
}
-/* Notice:
+/**
+ * vPMD raw receive routine, only accept(nb_pkts >= IAVF_VPMD_DESCS_PER_LOOP)
+ *
+ * Notice:
* - nb_pkts < IAVF_VPMD_DESCS_PER_LOOP, just return no packet
- * - nb_pkts > IAVF_VPMD_RX_MAX_BURST, only scan IAVF_VPMD_RX_MAX_BURST
- * numbers of DD bits
+ * - floor align nb_pkts to a IAVF_VPMD_DESCS_PER_LOOP power-of-two
*/
static inline uint16_t
_recv_raw_pkts_vec(struct iavf_rx_queue *rxq, struct rte_mbuf **rx_pkts,
@@ -413,9 +415,6 @@ _recv_raw_pkts_vec(struct iavf_rx_queue *rxq, struct
rte_mbuf **rx_pkts,
offsetof(struct rte_mbuf, rx_descriptor_fields1) + 8);
__m128i dd_check, eop_check;
- /* nb_pkts shall be less equal than IAVF_VPMD_RX_MAX_BURST */
- nb_pkts = RTE_MIN(nb_pkts, IAVF_VPMD_RX_MAX_BURST);
-
/* nb_pkts has to be floor-aligned to IAVF_VPMD_DESCS_PER_LOOP */
nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, IAVF_VPMD_DESCS_PER_LOOP);
@@ -627,10 +626,13 @@ _recv_raw_pkts_vec(struct iavf_rx_queue *rxq, struct
rte_mbuf **rx_pkts,
return nb_pkts_recd;
}
-/* Notice:
+/**
+ * vPMD raw receive routine for flex RxD,
+ * only accept(nb_pkts >= IAVF_VPMD_DESCS_PER_LOOP)
+ *
+ * Notice:
* - nb_pkts < IAVF_VPMD_DESCS_PER_LOOP, just return no packet
- * - nb_pkts > IAVF_VPMD_RX_MAX_BURST, only scan IAVF_VPMD_RX_MAX_BURST
- * numbers of DD bits
+ * - floor align nb_pkts to a IAVF_VPMD_DESCS_PER_LOOP power-of-two
*/
static inline uint16_t
_recv_raw_pkts_vec_flex_rxd(struct iavf_rx_queue *rxq,
@@ -688,9 +690,6 @@ _recv_raw_pkts_vec_flex_rxd(struct iavf_rx_queue *rxq,
const __m128i eop_check = _mm_set_epi64x(0x0000000200000002LL,
0x0000000200000002LL);
- /* nb_pkts shall be less equal than IAVF_VPMD_RX_MAX_BURST */
- nb_pkts = RTE_MIN(nb_pkts, IAVF_VPMD_RX_MAX_BURST);
-
/* nb_pkts has to be floor-aligned to IAVF_VPMD_DESCS_PER_LOOP */
nb_pkts = RTE_ALIGN_FLOOR(nb_pkts, IAVF_VPMD_DESCS_PER_LOOP);
@@ -921,11 +920,6 @@ _recv_raw_pkts_vec_flex_rxd(struct iavf_rx_queue *rxq,
return nb_pkts_recd;
}
-/* Notice:
- * - nb_pkts < IAVF_DESCS_PER_LOOP, just return no packet
- * - nb_pkts > IAVF_VPMD_RX_MAX_BURST, only scan IAVF_VPMD_RX_MAX_BURST
- * numbers of DD bits
- */
uint16_t
iavf_recv_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
uint16_t nb_pkts)
@@ -933,11 +927,6 @@ iavf_recv_pkts_vec(void *rx_queue, struct rte_mbuf
**rx_pkts,
return _recv_raw_pkts_vec(rx_queue, rx_pkts, nb_pkts, NULL);
}
-/* Notice:
- * - nb_pkts < IAVF_DESCS_PER_LOOP, just return no packet
- * - nb_pkts > IAVF_VPMD_RX_MAX_BURST, only scan IAVF_VPMD_RX_MAX_BURST
- * numbers of DD bits
- */
uint16_t
iavf_recv_pkts_vec_flex_rxd(void *rx_queue, struct rte_mbuf **rx_pkts,
uint16_t nb_pkts)
@@ -945,20 +934,20 @@ iavf_recv_pkts_vec_flex_rxd(void *rx_queue, struct
rte_mbuf **rx_pkts,
return _recv_raw_pkts_vec_flex_rxd(rx_queue, rx_pkts, nb_pkts, NULL);
}
-/* vPMD receive routine that reassembles scattered packets
- * Notice:
- * - nb_pkts < IAVF_VPMD_DESCS_PER_LOOP, just return no packet
- * - nb_pkts > VPMD_RX_MAX_BURST, only scan IAVF_VPMD_RX_MAX_BURST
- * numbers of DD bits
+/**
+ * vPMD receive routine that reassembles single burst of 32 scattered packets
*/
-uint16_t
-iavf_recv_scattered_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
- uint16_t nb_pkts)
+static uint16_t
+iavf_recv_scattered_burst_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
+ uint16_t nb_pkts)
{
struct iavf_rx_queue *rxq = rx_queue;
uint8_t split_flags[IAVF_VPMD_RX_MAX_BURST] = {0};
unsigned int i = 0;
+ /* split_flags only can support max of IAVF_VPMD_RX_MAX_BURST */
+ nb_pkts = RTE_MIN(nb_pkts, IAVF_VPMD_RX_MAX_BURST);
+
/* get some new buffers */
uint16_t nb_bufs = _recv_raw_pkts_vec(rxq, rx_pkts, nb_pkts,
split_flags);
@@ -986,21 +975,48 @@ iavf_recv_scattered_pkts_vec(void *rx_queue, struct
rte_mbuf **rx_pkts,
&split_flags[i]);
}
-/* vPMD receive routine that reassembles scattered packets for flex RxD
- * Notice:
- * - nb_pkts < IAVF_VPMD_DESCS_PER_LOOP, just return no packet
- * - nb_pkts > VPMD_RX_MAX_BURST, only scan IAVF_VPMD_RX_MAX_BURST
- * numbers of DD bits
+/**
+ * vPMD receive routine that reassembles scattered packets.
*/
uint16_t
-iavf_recv_scattered_pkts_vec_flex_rxd(void *rx_queue,
- struct rte_mbuf **rx_pkts,
- uint16_t nb_pkts)
+iavf_recv_scattered_pkts_vec(void *rx_queue, struct rte_mbuf **rx_pkts,
+ uint16_t nb_pkts)
+{
+ uint16_t retval = 0;
+
+ while (nb_pkts > IAVF_VPMD_RX_MAX_BURST) {
+ uint16_t burst;
+
+ burst = iavf_recv_scattered_burst_vec(rx_queue,
+ rx_pkts + retval,
+ IAVF_VPMD_RX_MAX_BURST);
+ retval += burst;
+ nb_pkts -= burst;
+ if (burst < IAVF_VPMD_RX_MAX_BURST)
+ return retval;
+ }
+
+ return retval + iavf_recv_scattered_burst_vec(rx_queue,
+ rx_pkts + retval,
+ nb_pkts);
+}
+
+/**
+ * vPMD receive routine that reassembles single burst of 32 scattered packets
+ * for flex RxD
+ */
+static uint16_t
+iavf_recv_scattered_burst_vec_flex_rxd(void *rx_queue,
+ struct rte_mbuf **rx_pkts,
+ uint16_t nb_pkts)
{
struct iavf_rx_queue *rxq = rx_queue;
uint8_t split_flags[IAVF_VPMD_RX_MAX_BURST] = {0};
unsigned int i = 0;
+ /* split_flags only can support max of IAVF_VPMD_RX_MAX_BURST */
+ nb_pkts = RTE_MIN(nb_pkts, IAVF_VPMD_RX_MAX_BURST);
+
/* get some new buffers */
uint16_t nb_bufs = _recv_raw_pkts_vec_flex_rxd(rxq, rx_pkts, nb_pkts,
split_flags);
@@ -1028,6 +1044,33 @@ iavf_recv_scattered_pkts_vec_flex_rxd(void *rx_queue,
&split_flags[i]);
}
+/**
+ * vPMD receive routine that reassembles scattered packets for flex RxD
+ */
+uint16_t
+iavf_recv_scattered_pkts_vec_flex_rxd(void *rx_queue,
+ struct rte_mbuf **rx_pkts,
+ uint16_t nb_pkts)
+{
+ uint16_t retval = 0;
+
+ while (nb_pkts > IAVF_VPMD_RX_MAX_BURST) {
+ uint16_t burst;
+
+ burst = iavf_recv_scattered_burst_vec_flex_rxd(rx_queue,
+ rx_pkts + retval,
+ IAVF_VPMD_RX_MAX_BURST);
+ retval += burst;
+ nb_pkts -= burst;
+ if (burst < IAVF_VPMD_RX_MAX_BURST)
+ return retval;
+ }
+
+ return retval + iavf_recv_scattered_burst_vec_flex_rxd(rx_queue,
+ rx_pkts + retval,
+ nb_pkts);
+}
+
static inline void
vtx1(volatile struct iavf_tx_desc *txdp, struct rte_mbuf *pkt, uint64_t flags)
{
--
2.20.1
