net/memif: replace memcpy() with rte_memcpy() to improve perf
PATCH 1/1 net/memif: replace memcpy() with rte_memcpy() to improve perf
Replacing memcpy() with rte_memcpy() improved 64 byte packet performance by 33% on a POWER9 system and by 10% on an x86_64 system.
--- drivers/net/memif/rte_eth_memif.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)
drivers/net/memif/rte_eth_memif.c +4 -3
@@ -384,8 +384,8 @@ eth_memif_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)rte_pktmbuf_pkt_len (mbuf_head) += cp_len; rte_memcpy(rte_pktmbuf_mtod_offset(mbuf, void *, dst_off), - (uint8_t *)memif_get_buffer(proc_private, d0) + src_off, - cp_len); + (uint8_t *)memif_get_buffer(proc_private, d0) + + src_off, cp_len); src_off += cp_len; dst_off += cp_len;
@@ -644,7 +644,8 @@ eth_memif_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) } cp_len =RTE_MIN (dst_len, src_len); -rte_memcpy ((uint8_t *)memif_get_buffer (proc_private, d0) + dst_off, +rte_memcpy ((uint8_t *) +memif_get_buffer (proc_private, d0) + dst_off, rte_pktmbuf_mtod_offset(mbuf, void *, src_off), cp_len);
Signature · 1 line
2.27.0
Quoted 3 lines · On Wed, 23 Jun 2021 09:49:35 -0700 David Christensen <[email protected]>…
Replacing
memcpy()withrte_memcpy()improved 64 byte packet performance by 33% on a POWER9 system and by 10% on anx86_64system.
I see rte_memcpy was already used in the patch diff
Quoted 10 lines
Signed-off-by David Christensen <[email protected]>--- drivers/net/memif/rte_eth_memif.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)drivers/net/memif/rte_eth_memif.c +0
-0
drivers/net/memif/rte_eth_memif.c eth_memif_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) · line 388
Quoted 7 lines
drivers/net/memif/rte_eth_memif.c +2
-2@@ -384,8 +384,8 @@ eth_memif_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)rte_pktmbuf_pkt_len (mbuf_head) += cp_len; rte_memcpy(rte_pktmbuf_mtod_offset(mbuf, void *, dst_off), - (uint8_t *)memif_get_buffer(proc_private, d0) + src_off, - cp_len); + (uint8_t *)memif_get_buffer(proc_private, d0) + + src_off, cp_len);
This just changes line break for no good reason.
drivers/net/memif/rte_eth_memif.c eth_memif_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) · line 651
Quoted 13 lines
drivers/net/memif/rte_eth_memif.c +2
-1src_off += cp_len; dst_off += cp_len;@@ -644,7 +644,8 @@ eth_memif_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) } cp_len =RTE_MIN (dst_len, src_len); -rte_memcpy ((uint8_t *)memif_get_buffer (proc_private, d0) + dst_off, +rte_memcpy ((uint8_t *) +memif_get_buffer (proc_private, d0) + dst_off, rte_pktmbuf_mtod_offset(mbuf, void *, src_off), cp_len);
ditto.
Look like this patch is confused, the description does not match the code.
Quoted 44 lines · Stephen Hemminger
Quoted 3 lines · On Wed, 23 Jun 2021 09:49:35 -0700 David Christensen <[email protected]>…
Replacing
memcpy()withrte_memcpy()improved 64 byte packet performance by 33% on a POWER9 system and by 10% on anx86_64system.I see
rte_memcpywas already used in the patch diffQuoted 10 lines
Signed-off-by David Christensen <[email protected]>--- drivers/net/memif/rte_eth_memif.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-)drivers/net/memif/rte_eth_memif.c +0
-0drivers/net/memif/rte_eth_memif.c eth_memif_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) · line 388
Quoted 7 lines
drivers/net/memif/rte_eth_memif.c +2
-2@@ -384,8 +384,8 @@ eth_memif_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)rte_pktmbuf_pkt_len (mbuf_head) += cp_len; rte_memcpy(rte_pktmbuf_mtod_offset(mbuf, void *, dst_off), - (uint8_t *)memif_get_buffer(proc_private, d0) + src_off, - cp_len); + (uint8_t *)memif_get_buffer(proc_private, d0) + + src_off, cp_len);This just changes line break for no good reason.
drivers/net/memif/rte_eth_memif.c eth_memif_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) · line 651
Quoted 13 lines
drivers/net/memif/rte_eth_memif.c +2
-1src_off += cp_len; dst_off += cp_len;@@ -644,7 +644,8 @@ eth_memif_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) } cp_len =RTE_MIN (dst_len, src_len); -rte_memcpy ((uint8_t *)memif_get_buffer (proc_private, d0) + dst_off, +rte_memcpy ((uint8_t *) +memif_get_buffer (proc_private, d0) + dst_off, rte_pktmbuf_mtod_offset(mbuf, void *, src_off), cp_len);ditto.
Quoted 2 lines
Look like this patch is confused, the description does not match the code.
I guess it was a patch on top of [1]. Marking this one as rejected since I've applied style fixes before applying it.
[1]
https://patches.dpdk.org/project/dpdk/patch/[email protected]/
PATCH 1/1 net/memif: replace memcpy() with rte_memcpy() to improve perf
In reply to David ChristensenReplacing memcpy() with rte_memcpy() improved 64 byte packet performance by 33% on a POWER9 system and by 10% on an x86_64 system.
---
v2: Fixed last patch which was bungled due to a bad local git squash
drivers/net/memif/rte_eth_memif.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/net/memif/rte_eth_memif.c +2 -2
@@ -383,7 +383,7 @@ eth_memif_rx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts)if (mbuf != mbuf_head)rte_pktmbuf_pkt_len (mbuf_head) += cp_len; - memcpy(rte_pktmbuf_mtod_offset(mbuf, void *, dst_off), + rte_memcpy(rte_pktmbuf_mtod_offset(mbuf, void *, dst_off), (uint8_t *)memif_get_buffer(proc_private, d0) + src_off, cp_len);
@@ -644,7 +644,7 @@ eth_memif_tx(void *queue, struct rte_mbuf **bufs, uint16_t nb_pkts) } cp_len =RTE_MIN (dst_len, src_len); -memcpy ((uint8_t *)memif_get_buffer (proc_private, d0) + dst_off, +rte_memcpy ((uint8_t *)memif_get_buffer (proc_private, d0) + dst_off, rte_pktmbuf_mtod_offset(mbuf, void *, src_off), cp_len);
--
2.27.0
Quoted 6 lines · David Christensen
Replacing
memcpy()withrte_memcpy()improved 64 byte packet performance by 33% on a POWER9 system and by 10% on anx86_64system.Signed-off-by David Christensen <[email protected]>---
Andrew Rybchenko
AppliedQuoted 10 lines · Tyler Retzlaff
Quoted 6 lines · David Christensen
Replacing
memcpy()withrte_memcpy()improved 64 byte packet performance by 33% on a POWER9 system and by 10% on anx86_64system.Signed-off-by David Christensen <[email protected]>---
Acked-by Tyler Retzlaff <[email protected]>
Fixed long lines and changed summary to:
net/memif: use rte_memcpy() to improve performance
Applied, thanks.