The execution time of rx/tx burst routine depends on the burst size.
It would be meaningful to research this dependency, the patch
provides an extra profiling data per rx/tx burst size.
Signed-off-by: Viacheslav Ovsiienko <viachesl...@mellanox.com>
---
app/test-pmd/csumonly.c | 11 +++++++----
app/test-pmd/flowgen.c | 11 +++++++----
app/test-pmd/icmpecho.c | 12 ++++++++----
app/test-pmd/iofwd.c | 11 +++++++----
app/test-pmd/macfwd.c | 11 +++++++----
app/test-pmd/macswap.c | 11 +++++++----
app/test-pmd/rxonly.c | 2 +-
app/test-pmd/softnicfwd.c | 11 +++++++----
app/test-pmd/testpmd.c | 31 ++++++++++++++++++++++++-------
app/test-pmd/testpmd.h | 26 ++++++++++++++++++++++----
app/test-pmd/txonly.c | 9 ++++++---
11 files changed, 103 insertions(+), 43 deletions(-)
diff --git a/app/test-pmd/csumonly.c b/app/test-pmd/csumonly.c
index 4104737..c966892 100644
--- a/app/test-pmd/csumonly.c
+++ b/app/test-pmd/csumonly.c
@@ -797,7 +797,7 @@ struct simple_gre_hdr {
TEST_PMD_CORE_CYC_RX_START(start_rx_tsc);
nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
nb_pkt_per_burst);
- TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc);
+ TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc, nb_rx);
if (unlikely(nb_rx == 0))
return;
#ifdef RTE_TEST_PMD_RECORD_BURST_STATS
@@ -1067,7 +1067,7 @@ struct simple_gre_hdr {
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, tx_pkts_burst,
nb_prep);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_tx);
/*
* Retry if necessary
@@ -1075,11 +1075,14 @@ struct simple_gre_hdr {
if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
retry = 0;
while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
+ uint16_t nb_rt;
+
rte_delay_us(burst_tx_delay_time);
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
- nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
+ nb_rt = rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
&tx_pkts_burst[nb_tx], nb_rx - nb_tx);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ nb_tx += nb_rt;
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_rt);
}
}
fs->tx_packets += nb_tx;
diff --git a/app/test-pmd/flowgen.c b/app/test-pmd/flowgen.c
index 51e87b0..9189e7b 100644
--- a/app/test-pmd/flowgen.c
+++ b/app/test-pmd/flowgen.c
@@ -107,7 +107,7 @@
TEST_PMD_CORE_CYC_RX_START(start_rx_tsc);
nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
nb_pkt_per_burst);
- TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc);
+ TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc, nb_rx);
fs->rx_packets += nb_rx;
for (i = 0; i < nb_rx; i++)
@@ -179,18 +179,21 @@
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_pkt);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_tx);
/*
* Retry if necessary
*/
if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
retry = 0;
while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
+ uint16_t nb_rt;
+
rte_delay_us(burst_tx_delay_time);
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
- nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
+ nb_rt = rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
&pkts_burst[nb_tx], nb_rx - nb_tx);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ nb_tx += nb_rt;
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_rt);
}
}
fs->tx_packets += nb_tx;
diff --git a/app/test-pmd/icmpecho.c b/app/test-pmd/icmpecho.c
index 8843183..0c8a8af 100644
--- a/app/test-pmd/icmpecho.c
+++ b/app/test-pmd/icmpecho.c
@@ -304,7 +304,7 @@
TEST_PMD_CORE_CYC_RX_START(start_rx_tsc);
nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
nb_pkt_per_burst);
- TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc);
+ TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc, nb_rx);
if (unlikely(nb_rx == 0))
return;
@@ -492,7 +492,7 @@
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst,
nb_replies);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_tx);
/*
* Retry if necessary
*/
@@ -500,13 +500,17 @@
retry = 0;
while (nb_tx < nb_replies &&
retry++ < burst_tx_retry_num) {
+ uint16_t nb_rt;
+
rte_delay_us(burst_tx_delay_time);
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
- nb_tx += rte_eth_tx_burst(fs->tx_port,
+ nb_rt = rte_eth_tx_burst(fs->tx_port,
fs->tx_queue,
&pkts_burst[nb_tx],
nb_replies - nb_tx);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ nb_tx += nb_rt;
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc,
+ nb_rt);
}
}
fs->tx_packets += nb_tx;
diff --git a/app/test-pmd/iofwd.c b/app/test-pmd/iofwd.c
index 9ff6531..b05ed02 100644
--- a/app/test-pmd/iofwd.c
+++ b/app/test-pmd/iofwd.c
@@ -62,7 +62,7 @@
TEST_PMD_CORE_CYC_RX_START(start_rx_tsc);
nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue,
pkts_burst, nb_pkt_per_burst);
- TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc);
+ TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc, nb_rx);
if (unlikely(nb_rx == 0))
return;
fs->rx_packets += nb_rx;
@@ -73,18 +73,21 @@
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
pkts_burst, nb_rx);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_tx);
/*
* Retry if necessary
*/
if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
retry = 0;
while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
+ uint16_t nb_rt;
+
rte_delay_us(burst_tx_delay_time);
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
- nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
+ nb_rt = rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
&pkts_burst[nb_tx], nb_rx - nb_tx);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ nb_tx += nb_rt;
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_rt);
}
}
fs->tx_packets += nb_tx;
diff --git a/app/test-pmd/macfwd.c b/app/test-pmd/macfwd.c
index f4a213e..a4aae0c 100644
--- a/app/test-pmd/macfwd.c
+++ b/app/test-pmd/macfwd.c
@@ -67,7 +67,7 @@
TEST_PMD_CORE_CYC_RX_START(start_rx_tsc);
nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
nb_pkt_per_burst);
- TEST_PMD_CORE_CYC_RX_ADD(fs, start_tx_tsc);
+ TEST_PMD_CORE_CYC_RX_ADD(fs, start_tx_tsc, nb_rx);
if (unlikely(nb_rx == 0))
return;
@@ -102,18 +102,21 @@
}
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_rx);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_tx);
/*
* Retry if necessary
*/
if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
retry = 0;
while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
+ uint16_t nb_rt;
+
rte_delay_us(burst_tx_delay_time);
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
- nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
+ nb_rt = rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
&pkts_burst[nb_tx], nb_rx - nb_tx);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ nb_tx += nb_rt;
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_rt);
}
}
diff --git a/app/test-pmd/macswap.c b/app/test-pmd/macswap.c
index 5cb3133..57628ba 100644
--- a/app/test-pmd/macswap.c
+++ b/app/test-pmd/macswap.c
@@ -68,7 +68,7 @@
TEST_PMD_CORE_CYC_RX_START(start_rx_tsc);
nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
nb_pkt_per_burst);
- TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc);
+ TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc, nb_rx);
if (unlikely(nb_rx == 0))
return;
@@ -82,18 +82,21 @@
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_rx);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_tx);
/*
* Retry if necessary
*/
if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
retry = 0;
while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
+ uint16_t nb_rt;
+
rte_delay_us(burst_tx_delay_time);
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
- nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
+ nb_rt = rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
&pkts_burst[nb_tx], nb_rx - nb_tx);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ nb_tx += nb_rt;
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_rt);
}
}
fs->tx_packets += nb_tx;
diff --git a/app/test-pmd/rxonly.c b/app/test-pmd/rxonly.c
index 2820d7f..ee79e7b 100644
--- a/app/test-pmd/rxonly.c
+++ b/app/test-pmd/rxonly.c
@@ -60,7 +60,7 @@
TEST_PMD_CORE_CYC_RX_START(start_rx_tsc);
nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
nb_pkt_per_burst);
- TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc);
+ TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc, nb_rx);
if (unlikely(nb_rx == 0))
return;
diff --git a/app/test-pmd/softnicfwd.c b/app/test-pmd/softnicfwd.c
index b78f2ce..793677d 100644
--- a/app/test-pmd/softnicfwd.c
+++ b/app/test-pmd/softnicfwd.c
@@ -96,7 +96,7 @@ struct tm_hierarchy {
TEST_PMD_CORE_CYC_RX_START(start_rx_tsc);
nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue,
pkts_burst, nb_pkt_per_burst);
- TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc);
+ TEST_PMD_CORE_CYC_RX_ADD(fs, start_rx_tsc, nb_rx);
fs->rx_packets += nb_rx;
#ifdef RTE_TEST_PMD_RECORD_BURST_STATS
@@ -106,17 +106,20 @@ struct tm_hierarchy {
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
pkts_burst, nb_rx);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_tx);
/* Retry if necessary */
if (unlikely(nb_tx < nb_rx) && fs->retry_enabled) {
retry = 0;
while (nb_tx < nb_rx && retry++ < burst_tx_retry_num) {
+ uint16_t nb_rt;
+
rte_delay_us(burst_tx_delay_time);
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
- nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
+ nb_rt = rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
&pkts_burst[nb_tx], nb_rx - nb_tx);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ nb_tx += nb_rt;
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_rt);
}
}
fs->tx_packets += nb_tx;
diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
index b195880..1d4b55b 100644
--- a/app/test-pmd/testpmd.c
+++ b/app/test-pmd/testpmd.c
@@ -1515,7 +1515,7 @@ struct extmem_param {
#ifdef RTE_TEST_PMD_RECORD_BURST_STATS
static void
-pkt_burst_stats_display(const char *rx_tx, struct pkt_burst_stats *pbs)
+pkt_burst_stats_display(int nrx_tx, struct pkt_burst_stats *pbs)
{
unsigned int total_burst;
unsigned int nb_burst;
@@ -1549,8 +1549,8 @@ struct extmem_param {
if (total_burst == 0)
return;
burst_percent[0] = (burst_stats[0] * 100) / total_burst;
- printf(" %s-bursts : %u [%d%% of %d pkts", rx_tx, total_burst,
- burst_percent[0], (int) pktnb_stats[0]);
+ printf(" %s-bursts : %u [%d%% of %d pkts", nrx_tx ? "TX" : "RX",
+ total_burst, burst_percent[0], (int) pktnb_stats[0]);
if (burst_stats[0] == total_burst) {
printf("]\n");
return;
@@ -1568,6 +1568,23 @@ struct extmem_param {
}
printf(" + %d%% of %d pkts + %d%% of others]\n",
burst_percent[1], (int) pktnb_stats[1], burst_percent[2]);
+#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
+ if (!(fwdprof_flags & (nrx_tx ? RECORD_CORE_CYCLES_TX
+ : RECORD_CORE_CYCLES_RX)))
+ return;
+ for (nb_pkt = 0; nb_pkt < MAX_PKT_BURST; nb_pkt++) {
+ nb_burst = nrx_tx ? pbs->pkt_retry_spread[nb_pkt]
+ : pbs->pkt_burst_spread[nb_pkt];
+ if (nb_burst == 0)
+ continue;
+ printf(" CPU cycles/%u packet burst=%u (total cycles="
+ "%"PRIu64" / total %s bursts=%u)\n",
+ (unsigned int)nb_pkt,
+ (unsigned int)(pbs->pkt_ticks_spread[nb_pkt] / nb_burst),
+ pbs->pkt_ticks_spread[nb_pkt],
+ nrx_tx ? "TX" : "RX", nb_burst);
+ }
+#endif
}
#endif /* RTE_TEST_PMD_RECORD_BURST_STATS */
@@ -1601,8 +1618,8 @@ struct extmem_param {
}
#ifdef RTE_TEST_PMD_RECORD_BURST_STATS
- pkt_burst_stats_display("RX", &fs->rx_burst_stats);
- pkt_burst_stats_display("TX", &fs->tx_burst_stats);
+ pkt_burst_stats_display(false, &fs->rx_burst_stats);
+ pkt_burst_stats_display(true, &fs->tx_burst_stats);
#endif
}
@@ -1742,10 +1759,10 @@ struct extmem_param {
#ifdef RTE_TEST_PMD_RECORD_BURST_STATS
if (ports_stats[pt_id].rx_stream)
- pkt_burst_stats_display("RX",
+ pkt_burst_stats_display(false,
&ports_stats[pt_id].rx_stream->rx_burst_stats);
if (ports_stats[pt_id].tx_stream)
- pkt_burst_stats_display("TX",
+ pkt_burst_stats_display(true,
&ports_stats[pt_id].tx_stream->tx_burst_stats);
#endif
diff --git a/app/test-pmd/testpmd.h b/app/test-pmd/testpmd.h
index 6177a50..90eb0ef 100644
--- a/app/test-pmd/testpmd.h
+++ b/app/test-pmd/testpmd.h
@@ -89,6 +89,10 @@ enum {
*/
struct pkt_burst_stats {
unsigned int pkt_burst_spread[MAX_PKT_BURST];
+#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
+ unsigned int pkt_retry_spread[MAX_PKT_BURST];
+ uint64_t pkt_ticks_spread[MAX_PKT_BURST];
+#endif
};
#endif
@@ -340,21 +344,35 @@ struct queue_stats_mappings {
{if (fwdprof_flags & RECORD_CORE_CYCLES_FWD) \
{uint64_t tsc = rte_rdtsc(); tsc -= (s); fs->core_cycles += tsc; } }
-#define TEST_PMD_CORE_CYC_TX_ADD(fs, s) \
+#ifdef RTE_TEST_PMD_RECORD_BURST_STATS
+#define TEST_PMD_CORE_CYC_TX_ADD(fs, s, np) \
+{if (fwdprof_flags & RECORD_CORE_CYCLES_TX) \
+{uint64_t tsc = rte_rdtsc(); tsc -= (s); fs->core_tx_cycles += tsc; \
+fs->tx_burst_stats.pkt_ticks_spread[np] += tsc; \
+fs->tx_burst_stats.pkt_retry_spread[np]++; } }
+
+#define TEST_PMD_CORE_CYC_RX_ADD(fs, s, np) \
+{if (fwdprof_flags & RECORD_CORE_CYCLES_RX) \
+{uint64_t tsc = rte_rdtsc(); tsc -= (s); fs->core_rx_cycles += tsc; \
+fs->rx_burst_stats.pkt_ticks_spread[np] += tsc; } }
+
+#else /* RTE_TEST_PMD_RECORD_BURST_STATS */
+#define TEST_PMD_CORE_CYC_TX_ADD(fs, s, np) \
{if (fwdprof_flags & RECORD_CORE_CYCLES_TX) \
{uint64_t tsc = rte_rdtsc(); tsc -= (s); fs->core_tx_cycles += tsc; } }
-#define TEST_PMD_CORE_CYC_RX_ADD(fs, s) \
+#define TEST_PMD_CORE_CYC_RX_ADD(fs, s, np) \
{if (fwdprof_flags & RECORD_CORE_CYCLES_RX) \
{uint64_t tsc = rte_rdtsc(); tsc -= (s); fs->core_rx_cycles += tsc; } }
+#endif /* RTE_TEST_PMD_RECORD_BURST_STATS */
#else
/* No profiling statistics is configured. */
#define TEST_PMD_CORE_CYC_TX_START(a)
#define TEST_PMD_CORE_CYC_RX_START(a)
#define TEST_PMD_CORE_CYC_FWD_ADD(fs, s)
-#define TEST_PMD_CORE_CYC_TX_ADD(fs, s)
-#define TEST_PMD_CORE_CYC_RX_ADD(fs, s)
+#define TEST_PMD_CORE_CYC_TX_ADD(fs, s, np)
+#define TEST_PMD_CORE_CYC_RX_ADD(fs, s, np)
#endif /* RTE_TEST_PMD_RECORD_CORE_CYCLES */
/* globals used for configuration */
diff --git a/app/test-pmd/txonly.c b/app/test-pmd/txonly.c
index 8ff7410..593044e 100644
--- a/app/test-pmd/txonly.c
+++ b/app/test-pmd/txonly.c
@@ -299,18 +299,21 @@
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
nb_tx = rte_eth_tx_burst(fs->tx_port, fs->tx_queue, pkts_burst, nb_pkt);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_tx);
/*
* Retry if necessary
*/
if (unlikely(nb_tx < nb_pkt) && fs->retry_enabled) {
retry = 0;
while (nb_tx < nb_pkt && retry++ < burst_tx_retry_num) {
+ uint16_t nb_rt;
+
rte_delay_us(burst_tx_delay_time);
TEST_PMD_CORE_CYC_TX_START(start_tx_tsc);
- nb_tx += rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
+ nb_rt = rte_eth_tx_burst(fs->tx_port, fs->tx_queue,
&pkts_burst[nb_tx], nb_pkt - nb_tx);
- TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc);
+ nb_tx += nb_rt;
+ TEST_PMD_CORE_CYC_TX_ADD(fs, start_tx_tsc, nb_rt);
}
}
fs->tx_packets += nb_tx;
--
1.8.3.1
