app/testpmd: print clock with CPU cycles per pkt
PATCH 1/5 app/testpmd: print clock with CPU cycles per pkt
On aarch64 platforms, the cycles are counted using either a low-resolution generic counter or a high-resolution PMU cycle counter. Print the clock frequency along with CPU cycles/packet to identify which cycle counter is being used.
--- app/test-pmd/testpmd.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)
app/test-pmd/testpmd.c +3 -2
@@ -1953,11 +1953,12 @@ fwd_stats_display(void)"%s\n" , acc_stats_border, acc_stats_border);#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES +#define CYC_PER_MHZ 1E6if (total_recv >0 )printf ("\n CPU cycles/packet=%u (total cycles=" -"%" PRIu64" / total RX packets=%" PRIu64")\n" , +"%" PRIu64" / total RX packets=%" PRIu64") at %lu MHz Clock\n" , (unsigned int )(fwd_cycles / total_recv), - fwd_cycles, total_recv); + fwd_cycles, total_recv, (uint64_t )(rte_get_tsc_hz () /CYC_PER_MHZ ));#endif }
Signature · 1 line
2.20.1
PATCH 2/5 app/testpmd: print fractional part in CPU cycles
Change printing of CPU cycles/packet to include fractional part for accurateness.
Example:
Without patch:
CPU cycles/packet=14
(total cycles=4899533541 / total RX packets=343031966)
With patch:
CPU cycles/packet=14.28
(total cycles=4899533541 / total RX packets=343031966)
--- app/test-pmd/testpmd.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
app/test-pmd/testpmd.c +2 -2
@@ -1955,9 +1955,9 @@ fwd_stats_display(void)#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES #define CYC_PER_MHZ 1E6if (total_recv >0 ) -printf ("\n CPU cycles/packet=%u (total cycles=" +printf ("\n CPU cycles/packet=%.2f (total cycles=" "%" PRIu64" / total RX packets=%" PRIu64") at %lu MHz Clock\n" , - (unsigned int )(fwd_cycles / total_recv), + (double )(fwd_cycles / (double )total_recv), fwd_cycles, total_recv, (uint64_t )(rte_get_tsc_hz () /CYC_PER_MHZ ));#endif }
Signature · 1 line
2.20.1
Re: [dpdk-dev] [PATCH 2/5] app/testpmd: print fractional part in CPU cycles
Quoted 34 lines
-----Original Message-----
From: dev <[email protected]> On Behalf Of Dharmik Thakkar
Sent: Wednesday, May 6, 2020 10:59 PM
To: Lu, Wenzhuo <[email protected]>; Wu, Jingjing <[email protected]>; Iremonger, Bernard <[email protected]>
Cc: [email protected]; [email protected]; Dharmik Thakkar <[email protected]>
Subject: [dpdk-dev] [PATCH 2/5] app/testpmd: print fractional part in CPU cyclesChange printing of CPU cycles/packet to include fractional part for accurateness.
Example:
Without patch:
CPU cycles/packet=14
(total cycles=4899533541 / total RX packets=343031966)With patch:
CPU cycles/packet=14.28
(total cycles=4899533541 / total RX packets=343031966)Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>--- app/test-pmd/testpmd.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)app/test-pmd/testpmd.c +0
-0
app/test-pmd/testpmd.c fwd_stats_display(void) · line 1960
Quoted 8 lines
app/test-pmd/testpmd.c +2
-2@@ -1955,9 +1955,9 @@ fwd_stats_display(void)#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES #define CYC_PER_MHZ 1E6if (total_recv >0 ) - printf("\n CPU cycles/packet=%u (total cycles=" + printf("\n CPU cycles/packet=%.2f (total cycles=" "%"PRIu64" / total RX packets=%"PRIu64") at %lu MHz Clock\n", - (unsigned int)(fwd_cycles / total_recv), + (double)(fwd_cycles / (double)total_recv),
Probably safer long double - to avoid overflow.
Quoted 5 lines
app/test-pmd/testpmd.c +0
-1fwd_cycles, total_recv, (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));#endif } --2.20.1
Re: [dpdk-dev] [PATCH 2/5] app/testpmd: print fractional part in CPU cycles
Hi Konstantin,
Quoted 31 lines · Konstantin
Quoted 30 lines
-----Original Message-----
From: dev <[email protected]> On Behalf Of Dharmik Thakkar
Sent: Wednesday, May 6, 2020 10:59 PM
To: Lu, Wenzhuo <[email protected]>; Wu, Jingjing <[email protected]>; Iremonger, Bernard <[email protected]>
Cc: [email protected]; [email protected]; Dharmik Thakkar <[email protected]>
Subject: [dpdk-dev] [PATCH 2/5] app/testpmd: print fractional part in CPU cyclesChange printing of CPU cycles/packet to include fractional part for accurateness.
Example:
Without patch:
CPU cycles/packet=14
(total cycles=4899533541 / total RX packets=343031966)With patch:
CPU cycles/packet=14.28
(total cycles=4899533541 / total RX packets=343031966)Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
app/test-pmd/testpmd.c| 4 ++--1 file changed, 2 insertions(+), 2 deletions(-)
Quoted 15 lines
app/test-pmd/testpmd.c fwd_stats_display(void) · line 1960
Quoted 13 lines
app/test-pmd/testpmd.c +2
-2@@ -1955,9 +1955,9 @@ fwd_stats_display(void)#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES #define CYC_PER_MHZ 1E6if (total_recv >0 ) - printf("\n CPU cycles/packet=%u (total cycles=" + printf("\n CPU cycles/packet=%.2f (total cycles=" "%"PRIu64" / total RX packets=%"PRIu64") at %lu MHz Clock\n", - (unsigned int)(fwd_cycles / total_recv), + (double)(fwd_cycles / (double)total_recv),Probably safer long double - to avoid overflow.
Is it possible for a ‘double' to be less than 8 bytes?
Quoted 6 lines
Quoted 6 lines
app/test-pmd/testpmd.c +0
-1fwd_cycles, total_recv, (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));#endif } --2.20.1
Re: [dpdk-dev] [PATCH 2/5] app/testpmd: print fractional part in CPU cycles
Quoted 49 lines
Hi Konstantin,
Quoted 31 lines · Konstantin
Quoted 30 lines
-----Original Message-----
From: dev <[email protected]> On Behalf Of Dharmik Thakkar
Sent: Wednesday, May 6, 2020 10:59 PM
To: Lu, Wenzhuo <[email protected]>; Wu, Jingjing <[email protected]>; Iremonger, Bernard <[email protected]>
Cc: [email protected]; [email protected]; Dharmik Thakkar <[email protected]>
Subject: [dpdk-dev] [PATCH 2/5] app/testpmd: print fractional part in CPU cyclesChange printing of CPU cycles/packet to include fractional part for accurateness.
Example:
Without patch:
CPU cycles/packet=14
(total cycles=4899533541 / total RX packets=343031966)With patch:
CPU cycles/packet=14.28
(total cycles=4899533541 / total RX packets=343031966)Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
app/test-pmd/testpmd.c| 4 ++--1 file changed, 2 insertions(+), 2 deletions(-)Quoted 15 lines
app/test-pmd/testpmd.c fwd_stats_display(void) · line 1960
Quoted 13 lines
app/test-pmd/testpmd.c +2
-2@@ -1955,9 +1955,9 @@ fwd_stats_display(void)#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES #define CYC_PER_MHZ 1E6if (total_recv >0 ) - printf("\n CPU cycles/packet=%u (total cycles=" + printf("\n CPU cycles/packet=%.2f (total cycles=" "%"PRIu64" / total RX packets=%"PRIu64") at %lu MHz Clock\n", - (unsigned int)(fwd_cycles / total_recv), + (double)(fwd_cycles / (double)total_recv),Probably safer long double - to avoid overflow.
Quoted 1 line
Is it possible for a ‘double' to be less than 8 bytes?
That was my initial thought - that on some 32 bit systems it could be 4B. Though it seems I was wrong, so feel free to ignore.
BTW, what for double conversion, why not just:
double)(fwd_cycles /total_recv
?
Quoted 7 lines
Quoted 7 lines
Quoted 7 lines
app/test-pmd/testpmd.c +0
-1fwd_cycles, total_recv, (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));#endif } --2.20.1
Re: [dpdk-dev] [PATCH 2/5] app/testpmd: print fractional part in CPU cycles
Quoted 53 lines · Konstantin
Quoted 50 lines
Hi Konstantin,
Quoted 46 lines · Konstantin
Quoted 35 lines
-----Original Message-----
From: dev <[email protected]> On Behalf Of Dharmik Thakkar
Sent: Wednesday, May 6, 2020 10:59 PM
To: Lu, Wenzhuo <[email protected]>; Wu, Jingjing <[email protected]>; Iremonger, Bernard <[email protected]>
Cc: [email protected]; [email protected]; Dharmik Thakkar <[email protected]>
Subject: [dpdk-dev] [PATCH 2/5] app/testpmd: print fractional part in CPU cyclesChange printing of CPU cycles/packet to include fractional part for accurateness.
Example:
Without patch:
CPU cycles/packet=14
(total cycles=4899533541 / total RX packets=343031966)With patch:
CPU cycles/packet=14.28
(total cycles=4899533541 / total RX packets=343031966)Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
app/test-pmd/testpmd.c| 4 ++--1 file changed, 2 insertions(+), 2 deletions(-)app/test-pmd/testpmd.c +0
-0app/test-pmd/testpmd.c fwd_stats_display(void) · line 1960
Quoted 8 lines
app/test-pmd/testpmd.c +2
-2@@ -1955,9 +1955,9 @@ fwd_stats_display(void)#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES #define CYC_PER_MHZ 1E6if (total_recv >0 ) - printf("\n CPU cycles/packet=%u (total cycles=" + printf("\n CPU cycles/packet=%.2f (total cycles=" "%"PRIu64" / total RX packets=%"PRIu64") at %lu MHz Clock\n", - (unsigned int)(fwd_cycles / total_recv), + (double)(fwd_cycles / (double)total_recv),Probably safer long double - to avoid overflow.
Quoted 7 lines
Quoted 1 line
Is it possible for a ‘double' to be less than 8 bytes?
That was my initial thought - that on some 32 bit systems it could be 4B. Though it seems I was wrong, so feel free to ignore.
BTW, what for double conversion, why not just:
double)(fwd_cycles /total_recv
?
Without (double) total recv, I will always get the fractional part as .00. For the above example, with (double)(fwd_cycles / total_recv), I see 14.00 instead of 14.28
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
Quoted 10 lines
app/test-pmd/testpmd.c +0
-1fwd_cycles, total_recv, (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));#endif } --2.20.1
Re: [dpdk-dev] [PATCH 2/5] app/testpmd: print fractional part in CPU cycles
Quoted 66 lines
-----Original Message-----
From: Dharmik Thakkar <[email protected]>
Sent: Friday, May 8, 2020 6:37 PM
To: Ananyev, Konstantin <[email protected]>
Cc: Lu, Wenzhuo <[email protected]>; Wu, Jingjing <[email protected]>; Iremonger, Bernard <[email protected]>; [email protected]; nd <[email protected]>
Subject: Re: [dpdk-dev] [PATCH 2/5] app/testpmd: print fractional part in CPU cyclesQuoted 12 lines · Konstantin
Quoted 11 lines
Hi Konstantin,
Quoted 8 lines · Konstantin
Quoted 7 lines
-----Original Message-----
From: dev <[email protected]> On Behalf Of Dharmik Thakkar
Sent: Wednesday, May 6, 2020 10:59 PM
To: Lu, Wenzhuo <[email protected]>; Wu, Jingjing <[email protected]>; Iremonger, BernardQuoted 41 lines
Quoted 40 lines
Quoted 23 lines
Quoted 23 lines
Cc: [email protected]; [email protected]; Dharmik Thakkar <[email protected]>
Subject: [dpdk-dev] [PATCH 2/5] app/testpmd: print fractional part in CPU cyclesChange printing of CPU cycles/packet to include fractional part for accurateness.
Example:
Without patch:
CPU cycles/packet=14
(total cycles=4899533541 / total RX packets=343031966)With patch:
CPU cycles/packet=14.28
(total cycles=4899533541 / total RX packets=343031966)Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
app/test-pmd/testpmd.c| 4 ++--1 file changed, 2 insertions(+), 2 deletions(-)Quoted 15 lines
app/test-pmd/testpmd.c fwd_stats_display(void) · line 1960
Quoted 13 lines
app/test-pmd/testpmd.c +2
-2@@ -1955,9 +1955,9 @@ fwd_stats_display(void)#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES #define CYC_PER_MHZ 1E6if (total_recv >0 ) - printf("\n CPU cycles/packet=%u (total cycles=" + printf("\n CPU cycles/packet=%.2f (total cycles=" "%"PRIu64" / total RX packets=%"PRIu64") at %lu MHz Clock\n", - (unsigned int)(fwd_cycles / total_recv), + (double)(fwd_cycles / (double)total_recv),Probably safer long double - to avoid overflow.
Is it possible for a ‘double' to be less than 8 bytes?
Quoted 8 lines
Quoted 5 lines
That was my initial thought - that on some 32 bit systems it could be 4B. Though it seems I was wrong, so feel free to ignore.
BTW, what for double conversion, why not just:
double)(fwd_cycles /total_recv
?Without (double) total recv, I will always get the fractional part as .00. For the above example, with (double)(fwd_cycles /
total_recv), I see 14.00 instead of 14.28
AFAIK, that's because that way you told compiler to do integer division and then convert result to double. While (double)fwd_cycles /total_recv, will tell compiler to explicitly convert fwd_cycles to double and compiler will implicitly do the same for total_recv too, and result will be double too.
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
Quoted 11 lines
app/test-pmd/testpmd.c +0
-1fwd_cycles, total_recv, (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));#endif } --2.20.1
PATCH 3/5 doc: add cycles per packet in testpmd user guide
In reply to Dharmik ThakkarUpdate documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
--- doc/guides/testpmd_app_ug/testpmd_funcs.rst | 51 +++++++++++++-------- 1 file changed, 31 insertions(+), 20 deletions(-)
doc/guides/testpmd_app_ug/testpmd_funcs.rst +31 -20
@@ -351,26 +351,37 @@ Example for the io forwarding engine, with some packet drops on the tx side:: testpmd> show fwd stats all - ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- - RX-packets: 274293770 TX-packets: 274293642 TX-dropped: 128 - - ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- - RX-packets: 274301850 TX-packets: 274301850 TX-dropped: 0 - - ---------------------- Forward statistics for port 0 ---------------------- - RX-packets: 274293802 RX-dropped: 0 RX-total: 274293802 - TX-packets: 274301862 TX-dropped: 0 TX-total: 274301862 - ---------------------------------------------------------------------------- - - ---------------------- Forward statistics for port 1 ---------------------- - RX-packets: 274301894 RX-dropped: 0 RX-total: 274301894 - TX-packets: 274293706 TX-dropped: 128 TX-total: 274293834 - ---------------------------------------------------------------------------- - - +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ - RX-packets: 548595696 RX-dropped: 0 RX-total: 548595696 - TX-packets: 548595568 TX-dropped: 128 TX-total: 548595696 - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- + RX-packets: 43536504 TX-packets: 43536488 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- + RX-packets: 149738504 TX-packets: 149738504 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 1 -> TX Port= 0/Queue= 1 ------- + RX-packets: 149753052 TX-packets: 149753052 TX-dropped: 0 + + ---------------------- Forward statistics for port 0 ---------------------- + RX-packets: 43538881 RX-dropped: 72 RX-total: 43538953 + TX-packets: 299491753 TX-dropped: 0 TX-total: 299491753 + ---------------------------------------------------------------------------- + + ---------------------- Forward statistics for port 1 ---------------------- + RX-packets: 299493085 RX-dropped: 8357 RX-total: 299501442 + TX-packets: 43539683 TX-dropped: 0 TX-total: 43539683 + ---------------------------------------------------------------------------- + + +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ + RX-packets: 343031966 RX-dropped: 8429 RX-total: 343040395 + TX-packets: 343031436 TX-dropped: 0 TX-total: 343031436 + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + + CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz + +.. note:: + + Measuring CPU cycles/packet requires enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLES configuration option. + On aarch64 platforms, by default, the cycles are counted using generic counter which runs at a lower frequency than the CPU clock. + To get the cycles/packet at CPU clock frequency, please scale the cycles/packet to CPU clock frequency. Alternatively, use the PMU based cycle counter. clear fwd ~~~~~~~~~
Signature · 1 line
2.20.1
Re: [dpdk-dev] [PATCH 3/5] doc: add cycles per packet in testpmd user guide
Quoted 58 lines · Dharmik Thakkar
Update documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>--- doc/guides/testpmd_app_ug/testpmd_funcs.rst | 51 +++++++++++++-------- 1 file changed, 31 insertions(+), 20 deletions(-)doc/guides/testpmd_app_ug/testpmd_funcs.rst +19
-20@@ -351,26 +351,37 @@ Example for the io forwarding engine, with some packet drops on the tx side:: testpmd> show fwd stats all - ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- - RX-packets: 274293770 TX-packets: 274293642 TX-dropped: 128 - - ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- - RX-packets: 274301850 TX-packets: 274301850 TX-dropped: 0 - - ---------------------- Forward statistics for port 0 ---------------------- - RX-packets: 274293802 RX-dropped: 0 RX-total: 274293802 - TX-packets: 274301862 TX-dropped: 0 TX-total: 274301862 - ---------------------------------------------------------------------------- - - ---------------------- Forward statistics for port 1 ---------------------- - RX-packets: 274301894 RX-dropped: 0 RX-total: 274301894 - TX-packets: 274293706 TX-dropped: 128 TX-total: 274293834 - ---------------------------------------------------------------------------- - - +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ - RX-packets: 548595696 RX-dropped: 0 RX-total: 548595696 - TX-packets: 548595568 TX-dropped: 128 TX-total: 548595696 - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- + RX-packets: 43536504 TX-packets: 43536488 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- + RX-packets: 149738504 TX-packets: 149738504 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 1 -> TX Port= 0/Queue= 1 ------- + RX-packets: 149753052 TX-packets: 149753052 TX-dropped: 0 + + ---------------------- Forward statistics for port 0 ---------------------- + RX-packets: 43538881 RX-dropped: 72 RX-total: 43538953 + TX-packets: 299491753 TX-dropped: 0 TX-total: 299491753 + ---------------------------------------------------------------------------- + + ---------------------- Forward statistics for port 1 ---------------------- + RX-packets: 299493085 RX-dropped: 8357 RX-total: 299501442 + TX-packets: 43539683 TX-dropped: 0 TX-total: 43539683 + ---------------------------------------------------------------------------- +
doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 382
Quoted 10 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +10
-0+ +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ + RX-packets: 343031966 RX-dropped: 8429 RX-total: 343040395 + TX-packets: 343031436 TX-dropped: 0 TX-total: 343031436 + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + + CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz + +.. note:: + + Measuring CPU cycles/packet requires enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLES configuration option.
IMO, it a very useful feature for CI/CD and development as we don't need to see any Traffic generator to see any regression.
I think, enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLES configuration is overhead to use in all the environments.
How about adding a "new forward" engine with and remove CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLES.
The new forward engine will have CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLES code enabled. So the application can use the forward engine dynamically. aka the same build works for both cases.
Quoted 8 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +2
-1+ On aarch64 platforms, by default, the cycles are counted using generic counter which runs at a lower frequency than the CPU clock. + To get the cycles/packet at CPU clock frequency, please scale the cycles/packet to CPU clock frequency. Alternatively, use the PMU based cycle counter. clear fwd ~~~~~~~~~ --2.20.1
Re: [dpdk-dev] [PATCH 3/5] doc: add cycles per packet in testpmd user guide
Hi Jerin,
Quoted 78 lines · Jerin Jacob
Quoted 58 lines · Dharmik Thakkar
Update documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
doc/guides/testpmd_app_ug/testpmd_funcs.rst| 51 +++++++++++++--------1 file changed, 31 insertions(+), 20 deletions(-)doc/guides/testpmd_app_ug/testpmd_funcs.rst +19
-20@@ -351,26 +351,37 @@ Example for the io forwarding engine, with some packet drops on the tx side:: testpmd> show fwd stats all - ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- - RX-packets: 274293770 TX-packets: 274293642 TX-dropped: 128 - - ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- - RX-packets: 274301850 TX-packets: 274301850 TX-dropped: 0 - - ---------------------- Forward statistics for port 0 ---------------------- - RX-packets: 274293802 RX-dropped: 0 RX-total: 274293802 - TX-packets: 274301862 TX-dropped: 0 TX-total: 274301862 - ---------------------------------------------------------------------------- - - ---------------------- Forward statistics for port 1 ---------------------- - RX-packets: 274301894 RX-dropped: 0 RX-total: 274301894 - TX-packets: 274293706 TX-dropped: 128 TX-total: 274293834 - ---------------------------------------------------------------------------- - - +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ - RX-packets: 548595696 RX-dropped: 0 RX-total: 548595696 - TX-packets: 548595568 TX-dropped: 128 TX-total: 548595696 - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- + RX-packets: 43536504 TX-packets: 43536488 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- + RX-packets: 149738504 TX-packets: 149738504 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 1 -> TX Port= 0/Queue= 1 ------- + RX-packets: 149753052 TX-packets: 149753052 TX-dropped: 0 + + ---------------------- Forward statistics for port 0 ---------------------- + RX-packets: 43538881 RX-dropped: 72 RX-total: 43538953 + TX-packets: 299491753 TX-dropped: 0 TX-total: 299491753 + ---------------------------------------------------------------------------- + + ---------------------- Forward statistics for port 1 ---------------------- + RX-packets: 299493085 RX-dropped: 8357 RX-total: 299501442 + TX-packets: 43539683 TX-dropped: 0 TX-total: 43539683 + ---------------------------------------------------------------------------- +doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 382
Quoted 10 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +10
-0+ +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ + RX-packets: 343031966 RX-dropped: 8429 RX-total: 343040395 + TX-packets: 343031436 TX-dropped: 0 TX-total: 343031436 + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + + CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz + +.. note:: + + Measuring CPU cycles/packet requires enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLES configuration option.IMO, it a very useful feature for CI/CD and development as we don't need to see any Traffic generator to see any regression.
I think, enabling
CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLESconfiguration is overhead to use in all the environments.
Quoted 6 lines
How about adding a "new forward" engine with and remove
CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLES.
The new forward engine will haveCONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLEScode enabled. So the application can use the forward engine dynamically. aka the same build works for both cases.
Yes, this can be done as a separate patch.
The idea of this patch is to update the documentation with the missing CPU cycles/packet information.
Quoted 9 lines
Quoted 9 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +2
-0+ On aarch64 platforms, by default, the cycles are counted using generic counter which runs at a lower frequency than the CPU clock. + To get the cycles/packet at CPU clock frequency, please scale the cycles/packet to CPU clock frequency. Alternatively, use the PMU based cycle counter.clear fwd
~~~~~~~~~
--
2.20.1
Re: [dpdk-dev] [PATCH 3/5] doc: add cycles per packet in testpmd user guide
In reply to Dharmik ThakkarHi Dharmik,
Quoted 103 lines
-----Original Message-----
From: Dharmik Thakkar <[email protected]>
Sent: Wednesday, May 6, 2020 10:59 PM
To: Lu, Wenzhuo <[email protected]>; Wu, Jingjing <[email protected]>; Iremonger, Bernard <[email protected]>; Mcnamara, John <[email protected]>; Kovacevic, Marko <[email protected]>
Cc: [email protected]; [email protected]; Dharmik Thakkar <[email protected]>
Subject: [PATCH 3/5] doc: add cycles per packet in testpmd user guideUpdate documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>--- doc/guides/testpmd_app_ug/testpmd_funcs.rst | 51 +++++++++++++------ -- 1 file changed, 31 insertions(+), 20 deletions(-)doc/guides/testpmd_app_ug/testpmd_funcs.rst +0
-0@@ -351,26 +351,37 @@ Example for the io forwarding engine, with somepacket drops on the tx side::
testpmd> show fwd stats all
doc/guides/testpmd_app_ug/testpmd_funcs.rst +0
-19- ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ----- -- - RX-packets: 274293770 TX-packets: 274293642 TX-dropped: 128 - - ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ----- -- - RX-packets: 274301850 TX-packets: 274301850 TX-dropped: 0 - - ---------------------- Forward statistics for port 0 ---------------------- - RX-packets: 274293802 RX-dropped: 0 RX-total: 274293802 - TX-packets: 274301862 TX-dropped: 0 TX-total: 274301862 - ---------------------------------------------------------------------------- - - ---------------------- Forward statistics for port 1 ---------------------- - RX-packets: 274301894 RX-dropped: 0 RX-total: 274301894 - TX-packets: 274293706 TX-dropped: 128 TX-total: 274293834 - ---------------------------------------------------------------------------- - - +++++++++++++++ Accumulated forward statistics for allports+++++++++++++++
doc/guides/testpmd_app_ug/testpmd_funcs.rst +29
-6- RX-packets: 548595696 RX-dropped: 0 RX-total: 548595696 - TX-packets: 548595568 TX-dropped: 128 TX-total: 548595696 - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ++++++++++++++++++ + ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ---- --- + RX-packets: 43536504 TX-packets: 43536488 TX-dropped: + 0 + + ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ---- --- + RX-packets: 149738504 TX-packets: 149738504 TX-dropped: + 0 + + ------- Forward Stats for RX Port= 1/Queue= 1 -> TX Port= 0/Queue= 1 ---- --- + RX-packets: 149753052 TX-packets: 149753052 TX-dropped: + 0 + + ---------------------- Forward statistics for port 0 ---------------------- + RX-packets: 43538881 RX-dropped: 72 RX-total: 43538953 + TX-packets: 299491753 TX-dropped: 0 TX-total: 299491753 + + ---------------------------------------------------------------------- + ------ + + ---------------------- Forward statistics for port 1 ---------------------- + RX-packets: 299493085 RX-dropped: 8357 RX-total: 299501442 + TX-packets: 43539683 TX-dropped: 0 TX-total: 43539683 + + ---------------------------------------------------------------------- + ------ + + +++++++++++++++ Accumulated forward statistics for allports+++++++++++++++
doc/guides/testpmd_app_ug/testpmd_funcs.rst +14
-0+ RX-packets: 343031966 RX-dropped: 8429 RX-total: 343040395 + TX-packets: 343031436 TX-dropped: 0 TX-total: 343031436 + + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ ++++++++++++ + ++++++ + + CPU cycles/packet=14.28 (total cycles=4899533541 / total RX + packets=343031966) at 200 MHz + +.. note:: + + Measuring CPU cycles/packet requires enabling
Quoted 6 lines
CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLESconfiguration option.
+ On aarch64 platforms, by default, the cycles are counted using generic
counter which runs at a lower frequency than the CPU clock.
+ To get the cycles/packet at CPU clock frequency, please scale the
cycles/packet to CPU clock frequency. Alternatively, use the PMU based cycle counter.
It is preferred to use shorter lines in .rst file, preferably <= 80, up to 100 is acceptable.
Quoted 5 lines
clear fwd
~~~~~~~~~
--
2.20.1
Regards,
Bernard.
PATCH 4/5 doc: include config options in testpmd user guide
In reply to Dharmik ThakkarUpdate testpmd documentation to include RECORD configuration options, CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES and CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.
--- doc/guides/testpmd_app_ug/build_app.rst | 12 ++++++++++++ 1 file changed, 12 insertions(+)
doc/guides/testpmd_app_ug/build_app.rst +12 -0
@@ -21,6 +21,18 @@ The basic compilation steps are: export RTE_TARGET=x86_64-native-linux-gcc +##. If required, enable configuration options. For example: + + .. code-block:: console + + cd to the top-level DPDK directory + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES\)=n,\1=y,' config/common_base + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS\)=n,\1=y,' config/common_base + + Enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES enables measurement of CPU cycles. + + Enabling CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS enables display of RX and TX bursts. + #. Build the application: .. code-block:: console
Signature · 1 line
2.20.1
Re: [dpdk-dev] [PATCH 4/5] doc: include config options in testpmd user guide
Hi Dharmik,
Quoted 29 lines
-----Original Message-----
From: Dharmik Thakkar <[email protected]>
Sent: Wednesday, May 6, 2020 10:59 PM
To: Lu, Wenzhuo <[email protected]>; Wu, Jingjing <[email protected]>; Iremonger, Bernard <[email protected]>; Mcnamara, John <[email protected]>; Kovacevic, Marko <[email protected]>
Cc: [email protected]; [email protected]; Dharmik Thakkar <[email protected]>
Subject: [PATCH 4/5] doc: include config options in testpmd user guideUpdate testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>--- doc/guides/testpmd_app_ug/build_app.rst | 12 ++++++++++++ 1 file changed, 12 insertions(+)doc/guides/testpmd_app_ug/build_app.rst +0
-0
doc/guides/testpmd_app_ug/build_app.rst The basic compilation steps are: · line 24
Quoted 4 lines
doc/guides/testpmd_app_ug/build_app.rst +1
-0@@ -21,6 +21,18 @@ The basic compilation steps are: export RTE_TARGET=x86_64-native-linux-gcc +##. If required, enable configuration options. For example:
##. Should be #.
Quoted 17 lines
doc/guides/testpmd_app_ug/build_app.rst +5
-0+ + .. code-block:: console + + cd to the top-level DPDK directory + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES\)=n,\1=y,'config/common_base
doc/guides/testpmd_app_ug/build_app.rst +4
-0+ sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS\)=n,\1=y,' + config/common_base + + Enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES enablesmeasurement of CPU cycles.
doc/guides/testpmd_app_ug/build_app.rst +2
-0+ + Enabling CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS enablesdisplay of RX and TX bursts.
+
#. Build the application:
Quoted 3 lines
.. code-block:: console
--
2.20.1
Regards,
Bernard.
PATCH 5/5 doc: add aarch64 generic counter section
In reply to Dharmik ThakkarAdd a separate section for low-resolution generic counter for ARM64 profiling methods.
--- doc/guides/prog_guide/profile_app.rst | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-)
doc/guides/prog_guide/profile_app.rst +8 -3
@@ -57,11 +57,16 @@ For more derails refer to the `ARM64 specific PMU events enumeration <http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.100095_0002_04_en/way1382543438508.html>`_. -High-resolution cycle counter -~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +Low-resolution generic counter +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The default ``cntvct_el0`` based ``rte_rdtsc()`` provides a portable means to -get a wall clock counter in user space. Typically it runs at <= 100MHz. +get a wall clock counter in user space. Typically it runs at a lower clock frequency than the CPU clock frequency. +Cycles counted using this method should be scaled to CPU clock frequency. + + +High-resolution cycle counter +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The alternative method to enable ``rte_rdtsc()`` for a high resolution wall clock counter is through the ARMv8 PMU subsystem. The PMU cycle counter runs
Signature · 1 line
2.20.1
Re: [dpdk-dev] [PATCH 5/5] doc: add aarch64 generic counter section
Quoted 7 lines · Dharmik Thakkar
Add a separate section for low-resolution generic counter for ARM64 profiling methods.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>
Quoted 31 lines
--- doc/guides/prog_guide/profile_app.rst | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-)doc/guides/prog_guide/profile_app.rst +8
-4@@ -57,11 +57,16 @@ For more derails refer to the `ARM64 specific PMU events enumeration <http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.100095_0002_04_en/way1382543438508.html>`_. -High-resolution cycle counter -~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +Low-resolution generic counter +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The default ``cntvct_el0`` based ``rte_rdtsc()`` provides a portable means to -get a wall clock counter in user space. Typically it runs at <= 100MHz. +get a wall clock counter in user space. Typically it runs at a lower clock frequency than the CPU clock frequency. +Cycles counted using this method should be scaled to CPU clock frequency. + + +High-resolution cycle counter +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The alternative method to enable ``rte_rdtsc()`` for a high resolution wall clock counter is through the ARMv8 PMU subsystem. The PMU cycle counter runs --2.20.1
W dniu 06.05.2020 o 23:58, Dharmik Thakkar pisze:
Quoted 17 lines
On aarch64 platforms, the cycles are counted using either a low-resolution generic counter or a high-resolution PMU cycle counter. Print the clock frequency along with CPU cycles/packet to identify which cycle counter is being used.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>--- app/test-pmd/testpmd.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)app/test-pmd/testpmd.c +0
-0
app/test-pmd/testpmd.c fwd_stats_display(void) · line 1959
Quoted 8 lines
app/test-pmd/testpmd.c +2
-1@@ -1953,11 +1953,12 @@ fwd_stats_display(void)"%s\n" , acc_stats_border, acc_stats_border);#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES +#define CYC_PER_MHZ 1E6if (total_recv >0 ) printf("\n CPU cycles/packet=%u (total cycles=" - "%"PRIu64" / total RX packets=%"PRIu64")\n", + "%"PRIu64" / total RX packets=%"PRIu64") at %lu MHz Clock\n",
Please use PRIu64 instead of %lu to keep printing uint64_t portable for different hardware, OS and compilers.
Quoted 6 lines
app/test-pmd/testpmd.c +1
-1(unsigned int )(fwd_cycles / total_recv), - fwd_cycles, total_recv); + fwd_cycles, total_recv, (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));#endif }
Signature · 7 lines
Lukasz Wojciechowski Principal Software Engineer Samsung R&D Institute Poland Samsung Electronics Office +48 22 377 88 25 [email protected]
Hi Dharmik,
Quoted 35 lines
-----Original Message-----
From: Dharmik Thakkar <[email protected]>
Sent: Wednesday, May 6, 2020 10:59 PM
To: Lu, Wenzhuo <[email protected]>; Wu, Jingjing <[email protected]>; Iremonger, Bernard <[email protected]>
Cc: [email protected]; [email protected]; Dharmik Thakkar <[email protected]>
Subject: [PATCH 1/5] app/testpmd: print clock with CPU cycles per pktOn aarch64 platforms, the cycles are counted using either a low-resolution generic counter or a high-resolution PMU cycle counter.
Print the clock frequency along with CPU cycles/packet to identify which cycle counter is being used.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>--- app/test-pmd/testpmd.c | 5 +++-- 1 file changed, 3 insertions(+), 2 deletions(-)app/test-pmd/testpmd.c index renamed from app/test-pmd/testpmd.c +0
-0
99bacddbfdca..9a8cbbd6fc7c100644app/test-pmd/testpmd.c +0
-0@@ -1953,11 +1953,12 @@ fwd_stats_display(void)"%s\n" , acc_stats_border, acc_stats_border); #ifdef
RTE_TEST_PMD_RECORD_CORE_CYCLESapp/test-pmd/testpmd.c +2
-1+#define CYC_PER_MHZ 1E6if (total_recv >0 )printf ("\n CPU cycles/packet=%u (total cycles=" - "%"PRIu64" / total RX packets=%"PRIu64")\n", + "%"PRIu64" / total RX packets=%"PRIu64") at %lu MHz
Quoted 1 line
Clock\n",
Better to use =%"PRIu64" instead of %lu in previous line
Quoted 9 lines
app/test-pmd/testpmd.c +2
-2(unsigned int )(fwd_cycles / total_recv), - fwd_cycles, total_recv); + fwd_cycles, total_recv, (uint64_t )(rte_get_tsc_hz () / +CYC_PER_MHZ));#endif } --2.20.1
Better to fix checkpatch the following checkpatch warning:
WARNING:LONG_LINE: line over 80 characters
#77: FILE: app/test-pmd/testpmd.c:1961:
+ fwd_cycles, total_recv, (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));
Regards,
Bernard.
PATCH v2 1/5 app/testpmd: print clock with CPU cycles per pkt
In reply to Dharmik ThakkarOn aarch64 platforms, the cycles are counted using either a low-resolution generic counter or a high-resolution PMU cycle counter. Print the clock frequency along with CPU cycles/packet to identify which cycle counter is being used.
---
v2:
- Use %"PRIu64" instead of %lu
- Fix checkpatch warning
--- app/test-pmd/testpmd.c | 7 +++++-- 1 file changed, 5 insertions(+), 2 deletions(-)
app/test-pmd/testpmd.c +5 -2
@@ -1953,11 +1953,14 @@ fwd_stats_display(void)"%s\n" , acc_stats_border, acc_stats_border);#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES +#define CYC_PER_MHZ 1E6if (total_recv >0 )printf ("\n CPU cycles/packet=%u (total cycles=" -"%" PRIu64" / total RX packets=%" PRIu64")\n" , +"%" PRIu64" / total RX packets=%" PRIu64") at %" PRIu64 +" MHz Clock\n" , (unsigned int )(fwd_cycles / total_recv), - fwd_cycles, total_recv); + fwd_cycles, total_recv, + (uint64_t )(rte_get_tsc_hz () /CYC_PER_MHZ ));#endif }
Signature · 1 line
2.20.1
PATCH v2 2/5 app/testpmd: print fractional part in CPU cycles
Change printing of CPU cycles/packet to include fractional part for accurateness.
Example:
Without patch:
CPU cycles/packet=14
(total cycles=4899533541 / total RX packets=343031966)
With patch:
CPU cycles/packet=14.28
(total cycles=4899533541 / total RX packets=343031966)
---
v2:
- Change typecast to (double) fwd_cycles / total_recv
--- app/test-pmd/testpmd.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
app/test-pmd/testpmd.c +2 -2
@@ -1955,10 +1955,10 @@ fwd_stats_display(void)#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES #define CYC_PER_MHZ 1E6if (total_recv >0 ) -printf ("\n CPU cycles/packet=%u (total cycles=" +printf ("\n CPU cycles/packet=%.2F (total cycles=" "%" PRIu64" / total RX packets=%" PRIu64") at %" PRIu64" MHz Clock\n" , - (unsigned int )(fwd_cycles / total_recv), + (double ) fwd_cycles / total_recv, fwd_cycles, total_recv, (uint64_t )(rte_get_tsc_hz () /CYC_PER_MHZ ));#endif
Signature · 1 line
2.20.1
Re: [dpdk-dev] [PATCH v2 2/5] app/testpmd: print fractional part in CPU cycles
Quoted 19 lines
-----Original Message-----
From: Dharmik Thakkar <[email protected]>
Sent: Friday, May 8, 2020 11:38 PM
To: Lu, Wenzhuo <[email protected]>; Wu, Jingjing <[email protected]>; Iremonger, Bernard <[email protected]>
Cc: [email protected]; [email protected]; Dharmik Thakkar <[email protected]>
Subject: [PATCH v2 2/5] app/testpmd: print fractional part in CPU cyclesChange printing of CPU cycles/packet to include fractional part for accurateness.
Example:
Without patch:
CPU cycles/packet=14
(total cycles=4899533541 / total RX packets=343031966)
Quoted 7 lines
With patch:
CPU cycles/packet=14.28
(total cycles=4899533541 / total RX packets=343031966)Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>
PATCH v2 3/5 doc: add cycles per packet in testpmd user guide
In reply to Dharmik ThakkarUpdate documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
---
v2:
- Use shorter lines, up to 100.
--- doc/guides/testpmd_app_ug/testpmd_funcs.rst | 53 +++++++++++++-------- 1 file changed, 33 insertions(+), 20 deletions(-)
doc/guides/testpmd_app_ug/testpmd_funcs.rst +33 -20
@@ -351,26 +351,39 @@ Example for the io forwarding engine, with some packet drops on the tx side:: testpmd> show fwd stats all - ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- - RX-packets: 274293770 TX-packets: 274293642 TX-dropped: 128 - - ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- - RX-packets: 274301850 TX-packets: 274301850 TX-dropped: 0 - - ---------------------- Forward statistics for port 0 ---------------------- - RX-packets: 274293802 RX-dropped: 0 RX-total: 274293802 - TX-packets: 274301862 TX-dropped: 0 TX-total: 274301862 - ---------------------------------------------------------------------------- - - ---------------------- Forward statistics for port 1 ---------------------- - RX-packets: 274301894 RX-dropped: 0 RX-total: 274301894 - TX-packets: 274293706 TX-dropped: 128 TX-total: 274293834 - ---------------------------------------------------------------------------- - - +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ - RX-packets: 548595696 RX-dropped: 0 RX-total: 548595696 - TX-packets: 548595568 TX-dropped: 128 TX-total: 548595696 - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- + RX-packets: 43536504 TX-packets: 43536488 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- + RX-packets: 149738504 TX-packets: 149738504 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 1 -> TX Port= 0/Queue= 1 ------- + RX-packets: 149753052 TX-packets: 149753052 TX-dropped: 0 + + ---------------------- Forward statistics for port 0 ---------------------- + RX-packets: 43538881 RX-dropped: 72 RX-total: 43538953 + TX-packets: 299491753 TX-dropped: 0 TX-total: 299491753 + ---------------------------------------------------------------------------- + + ---------------------- Forward statistics for port 1 ---------------------- + RX-packets: 299493085 RX-dropped: 8357 RX-total: 299501442 + TX-packets: 43539683 TX-dropped: 0 TX-total: 43539683 + ---------------------------------------------------------------------------- + + +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ + RX-packets: 343031966 RX-dropped: 8429 RX-total: 343040395 + TX-packets: 343031436 TX-dropped: 0 TX-total: 343031436 + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + + CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz Clock + +.. note:: + + Measuring CPU cycles/packet requires enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLES + configuration option. On aarch64 platforms, by default, the cycles are counted using + generic counter which runs at a lower frequency than the CPU clock. To get the cycles/packet + at CPU clock frequency, please scale the cycles/packet to CPU clock frequency. + Alternatively, use the PMU based cycle counter. clear fwd ~~~~~~~~~
Signature · 1 line
2.20.1
Re: [dpdk-dev] [PATCH v2 3/5] doc: add cycles per packet in testpmd user guide
Quoted 12 lines
-----Original Message-----
From: Dharmik Thakkar <[email protected]>
Sent: Friday, May 8, 2020 11:38 PM
To: Lu, Wenzhuo <[email protected]>; Wu, Jingjing <[email protected]>; Iremonger, Bernard <[email protected]>; Mcnamara, John <[email protected]>; Kovacevic, Marko <[email protected]>
Cc: [email protected]; [email protected]; Dharmik Thakkar <[email protected]>
Subject: [PATCH v2 3/5] doc: add cycles per packet in testpmd user guide
Quoted 6 lines
Update documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>
Re: [dpdk-dev] [PATCH v2 3/5] doc: add cycles per packet in testpmd user guide
In reply to Dharmik ThakkarQuoted 56 lines · Dharmik Thakkar
Update documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
v2:
- Use shorter lines, up to 100.--- doc/guides/testpmd_app_ug/testpmd_funcs.rst | 53 +++++++++++++-------- 1 file changed, 33 insertions(+), 20 deletions(-)doc/guides/testpmd_app_ug/testpmd_funcs.rst +15
-20@@ -351,26 +351,39 @@ Example for the io forwarding engine, with some packet drops on the tx side:: testpmd> show fwd stats all - ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- - RX-packets: 274293770 TX-packets: 274293642 TX-dropped: 128 - - ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- - RX-packets: 274301850 TX-packets: 274301850 TX-dropped: 0 - - ---------------------- Forward statistics for port 0 ---------------------- - RX-packets: 274293802 RX-dropped: 0 RX-total: 274293802 - TX-packets: 274301862 TX-dropped: 0 TX-total: 274301862 - ---------------------------------------------------------------------------- - - ---------------------- Forward statistics for port 1 ---------------------- - RX-packets: 274301894 RX-dropped: 0 RX-total: 274301894 - TX-packets: 274293706 TX-dropped: 128 TX-total: 274293834 - ---------------------------------------------------------------------------- - - +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ - RX-packets: 548595696 RX-dropped: 0 RX-total: 548595696 - TX-packets: 548595568 TX-dropped: 128 TX-total: 548595696 - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- + RX-packets: 43536504 TX-packets: 43536488 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- + RX-packets: 149738504 TX-packets: 149738504 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 1 -> TX Port= 0/Queue= 1 ------- + RX-packets: 149753052 TX-packets: 149753052 TX-dropped: 0 + + ---------------------- Forward statistics for port 0 ---------------------- + RX-packets: 43538881 RX-dropped: 72 RX-total: 43538953 + TX-packets: 299491753 TX-dropped: 0 TX-total: 299491753 + ---------------------------------------------------------------------------- + + ---------------------- Forward statistics for port 1 ----------------------
doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 378
Quoted 10 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +10
-0+ RX-packets: 299493085 RX-dropped: 8357 RX-total: 299501442 + TX-packets: 43539683 TX-dropped: 0 TX-total: 43539683 + ---------------------------------------------------------------------------- + + +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ + RX-packets: 343031966 RX-dropped: 8429 RX-total: 343040395 + TX-packets: 343031436 TX-dropped: 0 TX-total: 343031436 + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + + CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz Clock
Hi Dharmik,
Overall it is OK to show "CPU cycles/packet", but my concern is numbers can be misleading, because for example the numbers I am getting is like below:
CPU cycles/packet=4.38 (total cycles=32871036274 / total RX packets=7511734336) at 2100 MHz Clock
doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 382
Quoted 4 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +4
-0+ +.. note:: + + Measuring CPU cycles/packet requires enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLES
s/TX_CYCLES/CYCLES
doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 386
Quoted 4 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +4
-0+ configuration option. On aarch64 platforms, by default, the cycles are counted using + generic counter which runs at a lower frequency than the CPU clock. To get the cycles/packet + at CPU clock frequency, please scale the cycles/packet to CPU clock frequency. + Alternatively, use the PMU based cycle counter.
Not sure the "aarch64 platforms" note suits to here, where documents the command. And I can see the platform documentation already updated for this.
Combining above two comments, what do you think instead of replacing existing stats, add a note saying enabling 'CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES' appends "CPU cycles/packet" and give above two samples, like:
"
Enabling RTE_TEST_PMD_RECORD_CORE_CYCLES appends "CPU cycles/packet" stats, like:
CPU cycles/packet=4.38 (total cycles=32871036274 / total RX packets=7511734336) at 2100 MHz Clock
OR
CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz Clock
"
Re: [dpdk-dev] [PATCH v2 3/5] doc: add cycles per packet in testpmd user guide
Hi Ferruh,
Quoted 72 lines · Ferruh Yigit
Quoted 56 lines · Dharmik Thakkar
Update documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
v2:
- Use shorter lines, up to 100.
---doc/guides/testpmd_app_ug/testpmd_funcs.rst| 53 +++++++++++++--------1 file changed, 33 insertions(+), 20 deletions(-)doc/guides/testpmd_app_ug/testpmd_funcs.rst +15
-20@@ -351,26 +351,39 @@ Example for the io forwarding engine, with some packet drops on the tx side:: testpmd> show fwd stats all - ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- - RX-packets: 274293770 TX-packets: 274293642 TX-dropped: 128 - - ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- - RX-packets: 274301850 TX-packets: 274301850 TX-dropped: 0 - - ---------------------- Forward statistics for port 0 ---------------------- - RX-packets: 274293802 RX-dropped: 0 RX-total: 274293802 - TX-packets: 274301862 TX-dropped: 0 TX-total: 274301862 - ---------------------------------------------------------------------------- - - ---------------------- Forward statistics for port 1 ---------------------- - RX-packets: 274301894 RX-dropped: 0 RX-total: 274301894 - TX-packets: 274293706 TX-dropped: 128 TX-total: 274293834 - ---------------------------------------------------------------------------- - - +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ - RX-packets: 548595696 RX-dropped: 0 RX-total: 548595696 - TX-packets: 548595568 TX-dropped: 128 TX-total: 548595696 - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- + RX-packets: 43536504 TX-packets: 43536488 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- + RX-packets: 149738504 TX-packets: 149738504 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 1 -> TX Port= 0/Queue= 1 ------- + RX-packets: 149753052 TX-packets: 149753052 TX-dropped: 0 + + ---------------------- Forward statistics for port 0 ---------------------- + RX-packets: 43538881 RX-dropped: 72 RX-total: 43538953 + TX-packets: 299491753 TX-dropped: 0 TX-total: 299491753 + ---------------------------------------------------------------------------- + + ---------------------- Forward statistics for port 1 ----------------------doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 378
Quoted 10 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +10
-0+ RX-packets: 299493085 RX-dropped: 8357 RX-total: 299501442 + TX-packets: 43539683 TX-dropped: 0 TX-total: 43539683 + ---------------------------------------------------------------------------- + + +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ + RX-packets: 343031966 RX-dropped: 8429 RX-total: 343040395 + TX-packets: 343031436 TX-dropped: 0 TX-total: 343031436 + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + + CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz ClockHi Dharmik,
Quoted 5 lines
Overall it is OK to show "CPU cycles/packet", but my concern is numbers can be misleading, because for example the numbers I am getting is like below:
CPU cycles/packet=4.38 (total cycles=32871036274 / total RX packets=7511734336) at 2100 MHz Clock
CPU cycles/packet shown above is a part of the example. The numbers are as per the statistics shown in the example.
Quoted 7 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 382
Quoted 5 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +4
-0+ +.. note:: + + Measuring CPU cycles/packet requires enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLESs/TX_CYCLES/CYCLES
Sorry, didn’t understand this comment.
Quoted 8 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 386
Quoted 5 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +4
-0+ configuration option. On aarch64 platforms, by default, the cycles are counted using + generic counter which runs at a lower frequency than the CPU clock. To get the cycles/packet + at CPU clock frequency, please scale the cycles/packet to CPU clock frequency. + Alternatively, use the PMU based cycle counter.Not sure the "aarch64 platforms" note suits to here, where documents the command. And I can see the platform documentation already updated for this.
This additional comment about aarch64 platforms is to provide complete information to the users with regards to the config option and offer a better understanding of the CPU cycles/packet numbers. IMO, without this note, CPU cycles/packet numbers can be misleading.
Quoted 8 lines
Combining above two comments, what do you think instead of replacing existing stats, add a note saying enabling '
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES' appends "CPU cycles/packet" and give above two samples, like:"
EnablingRTE_TEST_PMD_RECORD_CORE_CYCLESappends "CPU cycles/packet" stats, like:
Quoted 6 lines
CPU cycles/packet=4.38 (total cycles=32871036274 / total RX packets=7511734336) at 2100 MHz Clock
OR
CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz Clock
"
Yes, this is also an option. IMO, numbers shown within CPU cycles/packet example should be consistent with the forward statistics numbers.
Re: [dpdk-dev] [PATCH v2 3/5] doc: add cycles per packet in testpmd user guide
Quoted 74 lines · Dharmik Thakkar
Hi Ferruh,
Quoted 72 lines · Ferruh Yigit
Quoted 56 lines · Dharmik Thakkar
Update documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
v2:
- Use shorter lines, up to 100.
---doc/guides/testpmd_app_ug/testpmd_funcs.rst| 53 +++++++++++++--------1 file changed, 33 insertions(+), 20 deletions(-)doc/guides/testpmd_app_ug/testpmd_funcs.rst +15
-20@@ -351,26 +351,39 @@ Example for the io forwarding engine, with some packet drops on the tx side:: testpmd> show fwd stats all - ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- - RX-packets: 274293770 TX-packets: 274293642 TX-dropped: 128 - - ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- - RX-packets: 274301850 TX-packets: 274301850 TX-dropped: 0 - - ---------------------- Forward statistics for port 0 ---------------------- - RX-packets: 274293802 RX-dropped: 0 RX-total: 274293802 - TX-packets: 274301862 TX-dropped: 0 TX-total: 274301862 - ---------------------------------------------------------------------------- - - ---------------------- Forward statistics for port 1 ---------------------- - RX-packets: 274301894 RX-dropped: 0 RX-total: 274301894 - TX-packets: 274293706 TX-dropped: 128 TX-total: 274293834 - ---------------------------------------------------------------------------- - - +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ - RX-packets: 548595696 RX-dropped: 0 RX-total: 548595696 - TX-packets: 548595568 TX-dropped: 128 TX-total: 548595696 - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- + RX-packets: 43536504 TX-packets: 43536488 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- + RX-packets: 149738504 TX-packets: 149738504 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 1 -> TX Port= 0/Queue= 1 ------- + RX-packets: 149753052 TX-packets: 149753052 TX-dropped: 0 + + ---------------------- Forward statistics for port 0 ---------------------- + RX-packets: 43538881 RX-dropped: 72 RX-total: 43538953 + TX-packets: 299491753 TX-dropped: 0 TX-total: 299491753 + ---------------------------------------------------------------------------- + + ---------------------- Forward statistics for port 1 ----------------------doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 378
Quoted 10 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +10
-0+ RX-packets: 299493085 RX-dropped: 8357 RX-total: 299501442 + TX-packets: 43539683 TX-dropped: 0 TX-total: 43539683 + ---------------------------------------------------------------------------- + + +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ + RX-packets: 343031966 RX-dropped: 8429 RX-total: 343040395 + TX-packets: 343031436 TX-dropped: 0 TX-total: 343031436 + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + + CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz ClockHi Dharmik,
Quoted 7 lines
Quoted 5 lines
Overall it is OK to show "CPU cycles/packet", but my concern is numbers can be misleading, because for example the numbers I am getting is like below:
CPU cycles/packet=4.38 (total cycles=32871036274 / total RX packets=7511734336) at 2100 MHz Clock
CPU cycles/packet shown above is a part of the example. The numbers are as per the statistics shown in the example.
I know but now you are showing a performance number and it is specific to a platform, my concern it may be misleading and create false expectations for others.
Quoted 10 lines
Quoted 8 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 382
Quoted 6 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +4
-0+ +.. note:: + + Measuring CPU cycles/packet requires enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLESs/TX_CYCLES/CYCLES
Sorry, didn’t understand this comment.
There is no 'CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLES' config option.
Quoted 7 lines
Quoted 7 lines
Quoted 6 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +4
-0+ configuration option. On aarch64 platforms, by default, the cycles are counted using + generic counter which runs at a lower frequency than the CPU clock. To get the cycles/packet + at CPU clock frequency, please scale the cycles/packet to CPU clock frequency. + Alternatively, use the PMU based cycle counter.
Quoted 5 lines
Quoted 2 lines
Not sure the "aarch64 platforms" note suits to here, where documents the command. And I can see the platform documentation already updated for this.
This additional comment about aarch64 platforms is to provide complete information to the users with regards to the config option and offer a better understanding of the CPU cycles/packet numbers. IMO, without this note, CPU cycles/packet numbers can be misleading.
This is testpmd documentation to describe "show fwd stats" command. We are adding a note to say with a config option it may also show performance data, which is OK. But starting to describe the platform differences for this config option looks to much details for the scope of the document. For testpmd I would prefer command usage independent from platforms, otherwise the documentation may become too confusing/complex.
But agree to provide the information for the specific platform, which you are already doing in other patch in the platform documentation.
Quoted 9 lines
Quoted 9 lines
Combining above two comments, what do you think instead of replacing existing stats, add a note saying enabling '
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES' appends "CPU cycles/packet" and give above two samples, like:"
EnablingRTE_TEST_PMD_RECORD_CORE_CYCLESappends "CPU cycles/packet" stats, like:
Quoted 9 lines
Quoted 6 lines
CPU cycles/packet=4.38 (total cycles=32871036274 / total RX packets=7511734336) at 2100 MHz Clock
OR
CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz Clock
"Yes, this is also an option. IMO, numbers shown within CPU cycles/packet example should be consistent with the forward statistics numbers.
This part is just the sample of the "CPU cycles/packet" output, I think it provides enough context to be understandable without matching Rx packet count in above part of the stats.
Re: [dpdk-dev] [PATCH v2 3/5] doc: add cycles per packet in testpmd user guide
Quoted 82 lines · Ferruh Yigit
Quoted 77 lines · Dharmik Thakkar
Hi Ferruh,
Quoted 75 lines · Ferruh Yigit
Quoted 56 lines · Dharmik Thakkar
Update documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
v2:
- Use shorter lines, up to 100.
---doc/guides/testpmd_app_ug/testpmd_funcs.rst| 53 +++++++++++++--------1 file changed, 33 insertions(+), 20 deletions(-)doc/guides/testpmd_app_ug/testpmd_funcs.rst +15
-20@@ -351,26 +351,39 @@ Example for the io forwarding engine, with some packet drops on the tx side:: testpmd> show fwd stats all - ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- - RX-packets: 274293770 TX-packets: 274293642 TX-dropped: 128 - - ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- - RX-packets: 274301850 TX-packets: 274301850 TX-dropped: 0 - - ---------------------- Forward statistics for port 0 ---------------------- - RX-packets: 274293802 RX-dropped: 0 RX-total: 274293802 - TX-packets: 274301862 TX-dropped: 0 TX-total: 274301862 - ---------------------------------------------------------------------------- - - ---------------------- Forward statistics for port 1 ---------------------- - RX-packets: 274301894 RX-dropped: 0 RX-total: 274301894 - TX-packets: 274293706 TX-dropped: 128 TX-total: 274293834 - ---------------------------------------------------------------------------- - - +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ - RX-packets: 548595696 RX-dropped: 0 RX-total: 548595696 - TX-packets: 548595568 TX-dropped: 128 TX-total: 548595696 - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- + RX-packets: 43536504 TX-packets: 43536488 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- + RX-packets: 149738504 TX-packets: 149738504 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 1 -> TX Port= 0/Queue= 1 ------- + RX-packets: 149753052 TX-packets: 149753052 TX-dropped: 0 + + ---------------------- Forward statistics for port 0 ---------------------- + RX-packets: 43538881 RX-dropped: 72 RX-total: 43538953 + TX-packets: 299491753 TX-dropped: 0 TX-total: 299491753 + ---------------------------------------------------------------------------- + + ---------------------- Forward statistics for port 1 ----------------------doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 378
Quoted 10 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +10
-0+ RX-packets: 299493085 RX-dropped: 8357 RX-total: 299501442 + TX-packets: 43539683 TX-dropped: 0 TX-total: 43539683 + ---------------------------------------------------------------------------- + + +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ + RX-packets: 343031966 RX-dropped: 8429 RX-total: 343040395 + TX-packets: 343031436 TX-dropped: 0 TX-total: 343031436 + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + + CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz ClockHi Dharmik,
Overall it is OK to show "CPU cycles/packet", but my concern is numbers can be misleading, because for example the numbers I am getting is like below:
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
CPU cycles/packet=4.38 (total cycles=32871036274 / total RX packets=7511734336) at 2100 MHz Clock
CPU cycles/packet shown above is a part of the example. The numbers are as per the statistics shown in the example.
I know but now you are showing a performance number and it is specific to a platform, my concern it may be misleading and create false expectations for others.
Ok, makes sense.
Quoted 10 lines
Quoted 10 lines
Quoted 9 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 382
Quoted 7 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +4
-0+ +.. note:: + + Measuring CPU cycles/packet requires enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLESs/TX_CYCLES/CYCLES
Quoted 3 lines
Quoted 1 line
Sorry, didn’t understand this comment.
There is no '
CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLES' config option.
Got it, will update.
Quoted 14 lines
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +4
-0+ configuration option. On aarch64 platforms, by default, the cycles are counted using + generic counter which runs at a lower frequency than the CPU clock. To get the cycles/packet + at CPU clock frequency, please scale the cycles/packet to CPU clock frequency. + Alternatively, use the PMU based cycle counter.Quoted 5 lines
Quoted 2 lines
Not sure the "aarch64 platforms" note suits to here, where documents the command. And I can see the platform documentation already updated for this.
This additional comment about aarch64 platforms is to provide complete information to the users with regards to the config option and offer a better understanding of the CPU cycles/packet numbers. IMO, without this note, CPU cycles/packet numbers can be misleading.
Quoted 8 lines
This is testpmd documentation to describe "show fwd stats" command. We are adding a note to say with a config option it may also show performance data, which is OK. But starting to describe the platform differences for this config option looks to much details for the scope of the document. For testpmd I would prefer command usage independent from platforms, otherwise the documentation may become too confusing/complex.
But agree to provide the information for the specific platform, which you are already doing in other patch in the platform documentation.
Alright, I will remove platform in the next version.
Quoted 17 lines
Quoted 17 lines
Quoted 16 lines
Combining above two comments, what do you think instead of replacing existing stats, add a note saying enabling '
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES' appends "CPU cycles/packet" and give above two samples, like:"
EnablingRTE_TEST_PMD_RECORD_CORE_CYCLESappends "CPU cycles/packet" stats, like:CPU cycles/packet=4.38 (total cycles=32871036274 / total RX packets=7511734336) at 2100 MHz Clock
OR
CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz Clock
"
Quoted 6 lines
Quoted 2 lines
Yes, this is also an option. IMO, numbers shown within CPU cycles/packet example should be consistent with the forward statistics numbers.
This part is just the sample of the "CPU cycles/packet" output, I think it provides enough context to be understandable without matching Rx packet count in above part of the stats.
Instead of adding 2 samples, do you think using xxx is best? For example:
CPU cycles/packet=xx.dd (total cycles=xxxx / total RX packets=xxxx) at xxxx MHz clock
Re: [dpdk-dev] [PATCH v2 3/5] doc: add cycles per packet in testpmd user guide
Quoted 153 lines · Dharmik Thakkar
Quoted 82 lines · Ferruh Yigit
Quoted 77 lines · Dharmik Thakkar
Hi Ferruh,
Quoted 75 lines · Ferruh Yigit
Quoted 56 lines · Dharmik Thakkar
Update documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
v2:
- Use shorter lines, up to 100.
---doc/guides/testpmd_app_ug/testpmd_funcs.rst| 53 +++++++++++++--------1 file changed, 33 insertions(+), 20 deletions(-)doc/guides/testpmd_app_ug/testpmd_funcs.rst +15
-20@@ -351,26 +351,39 @@ Example for the io forwarding engine, with some packet drops on the tx side:: testpmd> show fwd stats all - ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- - RX-packets: 274293770 TX-packets: 274293642 TX-dropped: 128 - - ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- - RX-packets: 274301850 TX-packets: 274301850 TX-dropped: 0 - - ---------------------- Forward statistics for port 0 ---------------------- - RX-packets: 274293802 RX-dropped: 0 RX-total: 274293802 - TX-packets: 274301862 TX-dropped: 0 TX-total: 274301862 - ---------------------------------------------------------------------------- - - ---------------------- Forward statistics for port 1 ---------------------- - RX-packets: 274301894 RX-dropped: 0 RX-total: 274301894 - TX-packets: 274293706 TX-dropped: 128 TX-total: 274293834 - ---------------------------------------------------------------------------- - - +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ - RX-packets: 548595696 RX-dropped: 0 RX-total: 548595696 - TX-packets: 548595568 TX-dropped: 128 TX-total: 548595696 - ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + ------- Forward Stats for RX Port= 0/Queue= 0 -> TX Port= 1/Queue= 0 ------- + RX-packets: 43536504 TX-packets: 43536488 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 0 -> TX Port= 0/Queue= 0 ------- + RX-packets: 149738504 TX-packets: 149738504 TX-dropped: 0 + + ------- Forward Stats for RX Port= 1/Queue= 1 -> TX Port= 0/Queue= 1 ------- + RX-packets: 149753052 TX-packets: 149753052 TX-dropped: 0 + + ---------------------- Forward statistics for port 0 ---------------------- + RX-packets: 43538881 RX-dropped: 72 RX-total: 43538953 + TX-packets: 299491753 TX-dropped: 0 TX-total: 299491753 + ---------------------------------------------------------------------------- + + ---------------------- Forward statistics for port 1 ----------------------doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 378
Quoted 10 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +10
-0+ RX-packets: 299493085 RX-dropped: 8357 RX-total: 299501442 + TX-packets: 43539683 TX-dropped: 0 TX-total: 43539683 + ---------------------------------------------------------------------------- + + +++++++++++++++ Accumulated forward statistics for all ports+++++++++++++++ + RX-packets: 343031966 RX-dropped: 8429 RX-total: 343040395 + TX-packets: 343031436 TX-dropped: 0 TX-total: 343031436 + ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ + + CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz ClockHi Dharmik,
Overall it is OK to show "CPU cycles/packet", but my concern is numbers can be misleading, because for example the numbers I am getting is like below:
Quoted 7 lines
Quoted 4 lines
Quoted 2 lines
CPU cycles/packet=4.38 (total cycles=32871036274 / total RX packets=7511734336) at 2100 MHz Clock
CPU cycles/packet shown above is a part of the example. The numbers are as per the statistics shown in the example.
I know but now you are showing a performance number and it is specific to a platform, my concern it may be misleading and create false expectations for others.
Ok, makes sense.
Quoted 10 lines
Quoted 10 lines
Quoted 9 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst Example for the io forwarding engine, with some packet drops on the tx side:: · line 382
Quoted 7 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +4
-0+ +.. note:: + + Measuring CPU cycles/packet requires enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLESs/TX_CYCLES/CYCLES
Quoted 3 lines
Quoted 1 line
Sorry, didn’t understand this comment.
There is no '
CONFIG_RTE_TEST_PMD_RECORD_CORE_TX_CYCLES' config option.Got it, will update.
Quoted 14 lines
Quoted 8 lines
Quoted 8 lines
Quoted 7 lines
doc/guides/testpmd_app_ug/testpmd_funcs.rst +4
-0+ configuration option. On aarch64 platforms, by default, the cycles are counted using + generic counter which runs at a lower frequency than the CPU clock. To get the cycles/packet + at CPU clock frequency, please scale the cycles/packet to CPU clock frequency. + Alternatively, use the PMU based cycle counter.Quoted 5 lines
Quoted 2 lines
Not sure the "aarch64 platforms" note suits to here, where documents the command. And I can see the platform documentation already updated for this.
This additional comment about aarch64 platforms is to provide complete information to the users with regards to the config option and offer a better understanding of the CPU cycles/packet numbers. IMO, without this note, CPU cycles/packet numbers can be misleading.
Quoted 8 lines
This is testpmd documentation to describe "show fwd stats" command. We are adding a note to say with a config option it may also show performance data, which is OK. But starting to describe the platform differences for this config option looks to much details for the scope of the document. For testpmd I would prefer command usage independent from platforms, otherwise the documentation may become too confusing/complex.
But agree to provide the information for the specific platform, which you are already doing in other patch in the platform documentation.Alright, I will remove platform in the next version.
Quoted 20 lines
Quoted 19 lines
Quoted 10 lines
Combining above two comments, what do you think instead of replacing existing stats, add a note saying enabling '
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES' appends "CPU cycles/packet" and give above two samples, like:"
EnablingRTE_TEST_PMD_RECORD_CORE_CYCLESappends "CPU cycles/packet" stats, like:Quoted 6 lines
CPU cycles/packet=4.38 (total cycles=32871036274 / total RX packets=7511734336) at 2100 MHz Clock
OR
CPU cycles/packet=14.28 (total cycles=4899533541 / total RX packets=343031966) at 200 MHz Clock
"Yes, this is also an option. IMO, numbers shown within CPU cycles/packet example should be consistent with the forward statistics numbers.
Quoted 7 lines
Quoted 3 lines
This part is just the sample of the "CPU cycles/packet" output, I think it provides enough context to be understandable without matching Rx packet count in above part of the stats.
Instead of adding 2 samples, do you think using xxx is best? For example:
CPU cycles/packet=xx.dd (total cycles=xxxx / total RX packets=xxxx) at xxxx MHz clock
+1
PATCH v3 1/2 doc: add cycles per packet in testpmd user guide
In reply to Dharmik ThakkarUpdate documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
---
v3:
- Keep existing stats and add a note providing CPU cycles/packet example (Ferruh)
- Remove platform specific information (Ferruh)
- Remove performance numbers from CPU cycles/packet example (Ferruh)
- Fix typo in CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES (Ferruh)
v2:
- Use shorter lines, up to 100.
--- doc/guides/testpmd_app_ug/testpmd_funcs.rst | 6 ++++++ 1 file changed, 6 insertions(+)
doc/guides/testpmd_app_ug/testpmd_funcs.rst +6 -0
@@ -372,6 +372,12 @@ Example for the io forwarding engine, with some packet drops on the tx side:: TX-packets: 548595568 TX-dropped: 128 TX-total: 548595696 ++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++ +.. note:: + + Enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES appends "CPU cycles/packet" stats, like: + + CPU cycles/packet=xx.dd (total cycles=xxxx / total RX packets=xxxx) at xxx MHz clock + clear fwd ~~~~~~~~~
Signature · 1 line
2.17.1
PATCH v3 2/2 doc: include config options in testpmd user guide
Update testpmd documentation to include RECORD configuration options, CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES and CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.
---
v3:
- Replace config/common_base with build/.config (Thomas)
v2:
- Remove extra '#'.
--- doc/guides/testpmd_app_ug/build_app.rst | 12 ++++++++++++ 1 file changed, 12 insertions(+)
doc/guides/testpmd_app_ug/build_app.rst +12 -0
@@ -21,6 +21,18 @@ The basic compilation steps are: export RTE_TARGET=x86_64-native-linux-gcc +#. If required, enable configuration options. For example: + + .. code-block:: console + + cd to the top-level DPDK directory + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES\)=n,\1=y,' build/.config + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS\)=n,\1=y,' build/.config + + Enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES enables measurement of CPU cycles. + + Enabling CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS enables display of RX and TX bursts. + #. Build the application: .. code-block:: console
Signature · 1 line
2.17.1
Re: [dpdk-dev] [PATCH v3 2/2] doc: include config options in testpmd user guide
Quoted 31 lines · Dharmik Thakkar
Update testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
v3:
- Replace config/common_base with build/.config (Thomas)v2:
- Remove extra '#'.--- doc/guides/testpmd_app_ug/build_app.rst | 12 ++++++++++++ 1 file changed, 12 insertions(+)doc/guides/testpmd_app_ug/build_app.rst +5
-0@@ -21,6 +21,18 @@ The basic compilation steps are: export RTE_TARGET=x86_64-native-linux-gcc +#. If required, enable configuration options. For example: + + .. code-block:: console + + cd to the top-level DPDK directory
doc/guides/testpmd_app_ug/build_app.rst The basic compilation steps are: · line 38
Quoted 10 lines
doc/guides/testpmd_app_ug/build_app.rst +7
-0+ sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES\)=n,\1=y,' build/.config + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS\)=n,\1=y,' build/.config + + Enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES enables measurement of CPU cycles. + + Enabling CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS enables display of RX and TX bursts. + #. Build the application: .. code-block:: console
Hi Dharmik,
This patch waiting to address the Thomas comment on document how to enable those config options in meson. @Thomas, can you please correct if I got it wrong?
Re: [dpdk-dev] [PATCH v3 2/2] doc: include config options in testpmd user guide
17/06/2020 20:21, Ferruh Yigit:
Quoted 40 lines · Dharmik Thakkar
Quoted 39 lines
Update testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
v3:
- Replace config/common_base with build/.config (Thomas)v2:
- Remove extra '#'.--- doc/guides/testpmd_app_ug/build_app.rst | 12 ++++++++++++ 1 file changed, 12 insertions(+)doc/guides/testpmd_app_ug/build_app.rst +12
-0@@ -21,6 +21,18 @@ The basic compilation steps are: export RTE_TARGET=x86_64-native-linux-gcc +#. If required, enable configuration options. For example: + + .. code-block:: console + + cd to the top-level DPDK directory + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES\)=n,\1=y,' build/.config + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS\)=n,\1=y,' build/.config + + Enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES enables measurement of CPU cycles. + + Enabling CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS enables display of RX and TX bursts. + #. Build the application:
Quoted 7 lines
doc/guides/testpmd_app_ug/build_app.rst The basic compilation steps are: · line 38
Quoted 2 lines
doc/guides/testpmd_app_ug/build_app.rst +0
-0.. code-block:: consoleHi Dharmik,
This patch waiting to address the Thomas comment on document how to enable those config options in meson. @Thomas, can you please correct if I got it wrong?
If the doc is only about make, it will be removed in 2 months. So yes, I think it doesn't make sense to be merged as is.
Re: [dpdk-dev] [PATCH v3 2/2] doc: include config options in testpmd user guide
+ Bruce
Quoted 48 lines · Thomas Monjalon
17/06/2020 20:21, Ferruh Yigit:
Quoted 45 lines · Dharmik Thakkar
Quoted 31 lines
Update testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
v3:
- Replace config/common_base with build/.config (Thomas)v2:
- Remove extra '#'.
---doc/guides/testpmd_app_ug/build_app.rst| 12 ++++++++++++1 file changed, 12 insertions(+)doc/guides/testpmd_app_ug/build_app.rst +5
-0@@ -21,6 +21,18 @@ The basic compilation steps are: export RTE_TARGET=x86_64-native-linux-gcc +#. If required, enable configuration options. For example: + + .. code-block:: console + + cd to the top-level DPDK directorydoc/guides/testpmd_app_ug/build_app.rst The basic compilation steps are: · line 38
Quoted 10 lines
doc/guides/testpmd_app_ug/build_app.rst +7
-0+ sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES\)=n,\1=y,' build/.config + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS\)=n,\1=y,' build/.config + + Enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES enables measurement of CPU cycles. + + Enabling CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS enables display of RX and TX bursts. + #. Build the application: .. code-block:: consoleHi Dharmik,
Quoted 6 lines
Quoted 2 lines
This patch waiting to address the Thomas comment on document how to enable those config options in meson. @Thomas, can you please correct if I got it wrong?
If the doc is only about make, it will be removed in 2 months. So yes, I think it doesn't make sense to be merged as is.
Make sense.
With meson, I see multiple ways to enable configuration options:
1] macro in $(RTE_SDK)/config/rte_config.h
2] using dpdk_conf.set() in $(RTE_SDK)/config/meson.build
I’d appreciate some suggestions on the best way to enable these options.
Quoted 2 lines
Re: [dpdk-dev] [PATCH v3 2/2] doc: include config options in testpmd user guide
19/06/2020 17:38, Dharmik Thakkar:
Quoted 57 lines
+ Bruce
Quoted 48 lines · Thomas Monjalon
17/06/2020 20:21, Ferruh Yigit:
Quoted 45 lines · Dharmik Thakkar
Quoted 31 lines
Update testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
v3:
- Replace config/common_base with build/.config (Thomas)v2:
- Remove extra '#'.
---doc/guides/testpmd_app_ug/build_app.rst| 12 ++++++++++++1 file changed, 12 insertions(+)doc/guides/testpmd_app_ug/build_app.rst +5
-0@@ -21,6 +21,18 @@ The basic compilation steps are: export RTE_TARGET=x86_64-native-linux-gcc +#. If required, enable configuration options. For example: + + .. code-block:: console + + cd to the top-level DPDK directorydoc/guides/testpmd_app_ug/build_app.rst The basic compilation steps are: · line 38
Quoted 10 lines
doc/guides/testpmd_app_ug/build_app.rst +7
-0+ sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES\)=n,\1=y,' build/.config + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS\)=n,\1=y,' build/.config + + Enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES enables measurement of CPU cycles. + + Enabling CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS enables display of RX and TX bursts. + #. Build the application: .. code-block:: consoleHi Dharmik,
Quoted 6 lines
Quoted 2 lines
This patch waiting to address the Thomas comment on document how to enable those config options in meson. @Thomas, can you please correct if I got it wrong?
If the doc is only about make, it will be removed in 2 months. So yes, I think it doesn't make sense to be merged as is.
Quoted 8 lines
Make sense.
With meson, I see multiple ways to enable configuration options:
1] macro in $(RTE_SDK)/config/rte_config.h
2] usingdpdk_conf.set()in $(RTE_SDK)/config/meson.buildI’d appreciate some suggestions on the best way to enable these options.
Neither 1 nor 2 because both are source files. Configuration must be done in the build directory.
We use "meson -D" to configure options listed in meson_options.txt The real question is which option we accept to be added.
Re: [dpdk-dev] [PATCH v3 2/2] doc: include config options in testpmd user guide
Quoted 69 lines · Thomas Monjalon
19/06/2020 17:38, Dharmik Thakkar:
Quoted 64 lines
+ Bruce
Quoted 48 lines · Thomas Monjalon
17/06/2020 20:21, Ferruh Yigit:
Quoted 45 lines · Dharmik Thakkar
Quoted 31 lines
Update testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
v3:
- Replace config/common_base with build/.config (Thomas)v2:
- Remove extra '#'.
---doc/guides/testpmd_app_ug/build_app.rst| 12 ++++++++++++1 file changed, 12 insertions(+)doc/guides/testpmd_app_ug/build_app.rst +5
-0@@ -21,6 +21,18 @@ The basic compilation steps are: export RTE_TARGET=x86_64-native-linux-gcc +#. If required, enable configuration options. For example: + + .. code-block:: console + + cd to the top-level DPDK directorydoc/guides/testpmd_app_ug/build_app.rst The basic compilation steps are: · line 38
Quoted 10 lines
doc/guides/testpmd_app_ug/build_app.rst +7
-0+ sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES\)=n,\1=y,' build/.config + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS\)=n,\1=y,' build/.config + + Enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES enables measurement of CPU cycles. + + Enabling CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS enables display of RX and TX bursts. + #. Build the application: .. code-block:: consoleHi Dharmik,
Quoted 6 lines
Quoted 2 lines
This patch waiting to address the Thomas comment on document how to enable those config options in meson. @Thomas, can you please correct if I got it wrong?
If the doc is only about make, it will be removed in 2 months. So yes, I think it doesn't make sense to be merged as is.
Make sense.
With meson, I see multiple ways to enable configuration options:
1] macro in $(RTE_SDK)/config/rte_config.h
2] usingdpdk_conf.set()in $(RTE_SDK)/config/meson.build
Quoted 9 lines
Quoted 1 line
I’d appreciate some suggestions on the best way to enable these options.
Neither 1 nor 2 because both are source files. Configuration must be done in the build directory.
We use "meson -D" to configure options listed in meson_options.txt The real question is which option we accept to be added.
Yes, correct. There are suggestions for making these as runtime configuration functions which can be enabled(or disabled) using ’set’ from within the application. Do you think, till this change is implemented, these options can be included in meson_options.txt?
Quoted 1 line
Re: [dpdk-dev] [PATCH v3 2/2] doc: include config options in testpmd user guide
19/06/2020 18:13, Dharmik Thakkar:
Quoted 68 lines
Quoted 68 lines · Thomas Monjalon
19/06/2020 17:38, Dharmik Thakkar:
Quoted 54 lines
Quoted 47 lines · Thomas Monjalon
17/06/2020 20:21, Ferruh Yigit:
Quoted 45 lines · Dharmik Thakkar
Quoted 31 lines
Update testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
v3:
- Replace config/common_base with build/.config (Thomas)v2:
- Remove extra '#'.
---doc/guides/testpmd_app_ug/build_app.rst| 12 ++++++++++++1 file changed, 12 insertions(+)doc/guides/testpmd_app_ug/build_app.rst +5
-0@@ -21,6 +21,18 @@ The basic compilation steps are: export RTE_TARGET=x86_64-native-linux-gcc +#. If required, enable configuration options. For example: + + .. code-block:: console + + cd to the top-level DPDK directorydoc/guides/testpmd_app_ug/build_app.rst The basic compilation steps are: · line 38
Quoted 10 lines
doc/guides/testpmd_app_ug/build_app.rst +7
-0+ sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES\)=n,\1=y,' build/.config + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS\)=n,\1=y,' build/.config + + Enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES enables measurement of CPU cycles. + + Enabling CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS enables display of RX and TX bursts. + #. Build the application: .. code-block:: consoleHi Dharmik,
Quoted 6 lines
Quoted 2 lines
This patch waiting to address the Thomas comment on document how to enable those config options in meson. @Thomas, can you please correct if I got it wrong?
If the doc is only about make, it will be removed in 2 months. So yes, I think it doesn't make sense to be merged as is.
Quoted 8 lines
Make sense.
With meson, I see multiple ways to enable configuration options:
1] macro in $(RTE_SDK)/config/rte_config.h
2] usingdpdk_conf.set()in $(RTE_SDK)/config/meson.buildI’d appreciate some suggestions on the best way to enable these options.
Neither 1 nor 2 because both are source files. Configuration must be done in the build directory.
Quoted 6 lines
Quoted 4 lines
We use "meson -D" to configure options listed in meson_options.txt The real question is which option we accept to be added.
Yes, correct. There are suggestions for making these as runtime configuration functions which can be enabled(or disabled) using ’set’ from within the application.
I don't understand this proposal.
Is there a patch?
Quoted 1 line
Do you think, till this change is implemented, these options can be included in meson_options.txt?
In general, if you think there is a better alternative, please propose a patch.
I have no opinion what is best to manage such option.
Re: [dpdk-dev] [PATCH v3 2/2] doc: include config options in testpmd user guide
Quoted 80 lines · Thomas Monjalon
19/06/2020 18:13, Dharmik Thakkar:
Quoted 68 lines
Quoted 68 lines · Thomas Monjalon
19/06/2020 17:38, Dharmik Thakkar:
Quoted 54 lines
Quoted 47 lines · Thomas Monjalon
17/06/2020 20:21, Ferruh Yigit:
Quoted 45 lines · Dharmik Thakkar
Quoted 31 lines
Update testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
v3:
- Replace config/common_base with build/.config (Thomas)v2:
- Remove extra '#'.
---doc/guides/testpmd_app_ug/build_app.rst| 12 ++++++++++++1 file changed, 12 insertions(+)doc/guides/testpmd_app_ug/build_app.rst +5
-0@@ -21,6 +21,18 @@ The basic compilation steps are: export RTE_TARGET=x86_64-native-linux-gcc +#. If required, enable configuration options. For example: + + .. code-block:: console + + cd to the top-level DPDK directorydoc/guides/testpmd_app_ug/build_app.rst The basic compilation steps are: · line 38
Quoted 10 lines
doc/guides/testpmd_app_ug/build_app.rst +7
-0+ sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES\)=n,\1=y,' build/.config + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS\)=n,\1=y,' build/.config + + Enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES enables measurement of CPU cycles. + + Enabling CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS enables display of RX and TX bursts. + #. Build the application: .. code-block:: consoleHi Dharmik,
Quoted 6 lines
Quoted 2 lines
This patch waiting to address the Thomas comment on document how to enable those config options in meson. @Thomas, can you please correct if I got it wrong?
If the doc is only about make, it will be removed in 2 months. So yes, I think it doesn't make sense to be merged as is.
Quoted 8 lines
Make sense.
With meson, I see multiple ways to enable configuration options:
1] macro in $(RTE_SDK)/config/rte_config.h
2] usingdpdk_conf.set()in $(RTE_SDK)/config/meson.buildI’d appreciate some suggestions on the best way to enable these options.
Neither 1 nor 2 because both are source files. Configuration must be done in the build directory.
Quoted 6 lines
Quoted 4 lines
We use "meson -D" to configure options listed in meson_options.txt The real question is which option we accept to be added.
Yes, correct. There are suggestions for making these as runtime configuration functions which can be enabled(or disabled) using ’set’ from within the application.
Quoted 7 lines
I don't understand this proposal.
Is there a patch?Quoted 1 line
Do you think, till this change is implemented, these options can be included in meson_options.txt?
In general, if you think there is a better alternative, please propose a patch.
I am not sure if changing it to a runtime function is a better alternative.
I am expecting some performance degradation with the change since the macros will be replaced with ‘if’ statements. I will hack up a patch to see the performance difference and propose if there is not much of a difference.
Quoted 2 lines
I have no opinion what is best to manage such option.
cover app/testpmd: add runtime config
In reply to Dharmik ThakkarMeson build system lacks support for CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES and CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS configuration options.
One solution is to add these options within meson_options.txt
Since adding these as runtime config causes no performance impact, it makes sense to enable these as runtime config alongside other available runtime options within testpmd.
Dharmik Thakkar (5):
app/testpmd: add record-core-cycles runtime config
doc: add record-core-cycles to testpmd funcs doc app/testpmd: add record-burst-stats runtime config
doc: add record-burst-stats to testpmd funcs doc app/testpmd: enable empty polls in 5tswap
Phil Yang (1):
app/testpmd: enable burst stats for noisy vnf mode
app/test-pmd/5tswap.c | 25 ++---- app/test-pmd/cmdline.c | 92 +++++++++++++++++++++ app/test-pmd/config.c | 12 +++ app/test-pmd/csumonly.c | 24 ++---- app/test-pmd/flowgen.c | 21 ++--- app/test-pmd/icmpecho.c | 24 ++---- app/test-pmd/iofwd.c | 26 ++---- app/test-pmd/macfwd.c | 25 ++---- app/test-pmd/macswap.c | 24 ++---- app/test-pmd/noisy_vnf.c | 4 + app/test-pmd/parameters.c | 11 ++- app/test-pmd/rxonly.c | 19 +---- app/test-pmd/testpmd.c | 77 ++++++++--------- app/test-pmd/testpmd.h | 38 +++++++-- app/test-pmd/txonly.c | 20 +---- doc/guides/testpmd_app_ug/run_app.rst | 8 ++ doc/guides/testpmd_app_ug/testpmd_funcs.rst | 30 +++++++ 17 files changed, 272 insertions(+), 208 deletions(-)
Signature · 1 line
2.17.1
PATCH 1/6 app/testpmd: add record-core-cycles runtime config
Convert CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES to a runtime configuration.
--- app/test-pmd/5tswap.c | 16 +++----------- app/test-pmd/cmdline.c | 46 +++++++++++++++++++++++++++++++++++++++ app/test-pmd/config.c | 6 +++++ app/test-pmd/csumonly.c | 16 +++----------- app/test-pmd/flowgen.c | 17 ++++----------- app/test-pmd/icmpecho.c | 16 +++----------- app/test-pmd/iofwd.c | 18 ++++----------- app/test-pmd/macfwd.c | 17 ++++----------- app/test-pmd/macswap.c | 16 +++----------- app/test-pmd/parameters.c | 6 ++++- app/test-pmd/rxonly.c | 15 +++---------- app/test-pmd/testpmd.c | 44 ++++++++++++++++++------------------- app/test-pmd/testpmd.h | 18 +++++++++++++-- app/test-pmd/txonly.c | 16 +++----------- 14 files changed, 125 insertions(+), 142 deletions(-)
app/test-pmd/5tswap.c +3 -13
@@ -105,15 +105,9 @@ pkt_burst_5tuple_swap(struct fwd_stream *fs)uint8_t *byte; } h; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES -uint64_t start_tsc; -uint64_t end_tsc; -uint64_t core_cycles; -#endif +uint64_t start_tsc =0 ; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - start_tsc =rte_rdtsc (); -#endif +get_start_cycles (&start_tsc);/* * Receive a burst of packets and forward them.
@@ -193,11 +187,7 @@ pkt_burst_5tuple_swap(struct fwd_stream *fs)rte_pktmbuf_free (pkts_burst[nb_tx]); }while (++nb_tx < nb_rx); } -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - end_tsc =rte_rdtsc (); - core_cycles = (end_tsc - start_tsc); - fs->core_cycles = (uint64_t ) (fs->core_cycles + core_cycles); -#endif +get_end_cycles (fs, start_tsc); }struct fwd_engine five_tuple_swap_fwd_engine = {
app/test-pmd/cmdline.c +46 -0
64 lines not on this page · raw message
app/test-pmd/config.c +6 -0
@@ -3679,6 +3679,12 @@ set_xstats_hide_zero(uint8_t on_off) xstats_hide_zero = on_off; } +void +set_record_core_cycles (uint8_t on_off) +{ + record_core_cycles = on_off; +} +static inline void print_fdir_mask (struct rte_eth_fdir_masks *mask) {
app/test-pmd/csumonly.c +3 -13
@@ -789,15 +789,9 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)uint16_t nb_segments =0 ;int ret; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES -uint64_t start_tsc; -uint64_t end_tsc; -uint64_t core_cycles; -#endif +uint64_t start_tsc =0 ; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - start_tsc =rte_rdtsc (); -#endif +get_start_cycles (&start_tsc);/* receive a burst of packet */ nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
@@ -1098,11 +1092,7 @@ pkt_burst_checksum_forward(struct fwd_stream *fs) }while (++nb_tx < nb_rx); } -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - end_tsc =rte_rdtsc (); - core_cycles = (end_tsc - start_tsc); - fs->core_cycles = (uint64_t ) (fs->core_cycles + core_cycles); -#endif +get_end_cycles (fs, start_tsc); }struct fwd_engine csum_fwd_engine = {
app/test-pmd/flowgen.c +4 -13
@@ -97,16 +97,10 @@ pkt_burst_flow_gen(struct fwd_stream *fs)uint16_t i;uint32_t retry;uint64_t tx_offloads; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES -uint64_t start_tsc; -uint64_t end_tsc; -uint64_t core_cycles; -#endif +uint64_t start_tsc =0 ;static int next_flow =0 ; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - start_tsc =rte_rdtsc (); -#endif +get_start_cycles (&start_tsc);/* Receive a burst of packets and discard them. */ nb_rx = rte_eth_rx_burst(fs->rx_port, fs->rx_queue, pkts_burst,
@@ -207,11 +201,8 @@ pkt_burst_flow_gen(struct fwd_stream *fs)rte_pktmbuf_free (pkts_burst[nb_tx]); }while (++nb_tx < nb_pkt); } -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - end_tsc =rte_rdtsc (); - core_cycles = (end_tsc - start_tsc); - fs->core_cycles = (uint64_t ) (fs->core_cycles + core_cycles); -#endif + +get_end_cycles (fs, start_tsc); }struct fwd_engine flow_gen_engine = {
app/test-pmd/icmpecho.c +3 -13
@@ -293,15 +293,9 @@ reply_to_icmp_echo_rqsts(struct fwd_stream *fs)uint32_t cksum;uint8_t i;int l2_len; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES -uint64_t start_tsc; -uint64_t end_tsc; -uint64_t core_cycles; -#endif +uint64_t start_tsc =0 ; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - start_tsc =rte_rdtsc (); -#endif +get_start_cycles (&start_tsc);/* * First, receive a burst of packets.
@@ -520,11 +514,7 @@ reply_to_icmp_echo_rqsts(struct fwd_stream *fs) } } -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - end_tsc =rte_rdtsc (); - core_cycles = (end_tsc - start_tsc); - fs->core_cycles = (uint64_t ) (fs->core_cycles + core_cycles); -#endif +get_end_cycles (fs, start_tsc); }struct fwd_engine icmp_echo_engine = {
app/test-pmd/iofwd.c +4 -14
@@ -50,16 +50,9 @@ pkt_burst_io_forward(struct fwd_stream *fs)uint16_t nb_rx;uint16_t nb_tx;uint32_t retry; +uint64_t start_tsc =0 ; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES -uint64_t start_tsc; -uint64_t end_tsc; -uint64_t core_cycles; -#endif - -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - start_tsc =rte_rdtsc (); -#endif +get_start_cycles (&start_tsc);/* * Receive a burst of packets and forward them.
@@ -96,11 +89,8 @@ pkt_burst_io_forward(struct fwd_stream *fs)rte_pktmbuf_free (pkts_burst[nb_tx]); }while (++nb_tx < nb_rx); } -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - end_tsc =rte_rdtsc (); - core_cycles = (end_tsc - start_tsc); - fs->core_cycles = (uint64_t ) (fs->core_cycles + core_cycles); -#endif + +get_end_cycles (fs, start_tsc); }struct fwd_engine io_fwd_engine = {
app/test-pmd/macfwd.c +4 -13
@@ -56,15 +56,9 @@ pkt_burst_mac_forward(struct fwd_stream *fs)uint16_t i;uint64_t ol_flags =0 ;uint64_t tx_offloads; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES -uint64_t start_tsc; -uint64_t end_tsc; -uint64_t core_cycles; -#endif +uint64_t start_tsc =0 ; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - start_tsc =rte_rdtsc (); -#endif +get_start_cycles (&start_tsc);/* * Receive a burst of packets and forward them.
@@ -126,11 +120,8 @@ pkt_burst_mac_forward(struct fwd_stream *fs)rte_pktmbuf_free (pkts_burst[nb_tx]); }while (++nb_tx < nb_rx); } -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - end_tsc =rte_rdtsc (); - core_cycles = (end_tsc - start_tsc); - fs->core_cycles = (uint64_t ) (fs->core_cycles + core_cycles); -#endif + +get_end_cycles (fs, start_tsc); }struct fwd_engine mac_fwd_engine = {
app/test-pmd/macswap.c +3 -13
@@ -57,15 +57,9 @@ pkt_burst_mac_swap(struct fwd_stream *fs)uint16_t nb_rx;uint16_t nb_tx;uint32_t retry; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES -uint64_t start_tsc; -uint64_t end_tsc; -uint64_t core_cycles; -#endif +uint64_t start_tsc =0 ; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - start_tsc =rte_rdtsc (); -#endif +get_start_cycles (&start_tsc);/* * Receive a burst of packets and forward them.
@@ -105,11 +99,7 @@ pkt_burst_mac_swap(struct fwd_stream *fs)rte_pktmbuf_free (pkts_burst[nb_tx]); }while (++nb_tx < nb_rx); } -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - end_tsc =rte_rdtsc (); - core_cycles = (end_tsc - start_tsc); - fs->core_cycles = (uint64_t ) (fs->core_cycles + core_cycles); -#endif +get_end_cycles (fs, start_tsc); }struct fwd_engine mac_swap_engine = {
app/test-pmd/parameters.c +5 -1
@@ -70,7 +70,7 @@ usage(char* progname)"--rxpt= | --rxht= | --rxwt= | --rxfreet= | " "--txpt= | --txht= | --txwt= | --txfreet= | " "--txrst= | --tx-offloads= | | --rx-offloads= | " -"--vxlan-gpe-port= ]\n" , +"--vxlan-gpe-port= | --record-core-cycles]\n" , progname);#ifdef RTE_LIBRTE_CMDLINE printf (" --interactive: run in interactive mode.\n" );
@@ -216,6 +216,7 @@ usage(char* progname) "valid only with --mp-alloc=anon\n");printf (" --rx-mq-mode=0xX: hexadecimal bitmask of RX mq mode can be " "enabled\n" ); +printf (" --record-core-cycles: enable measurement of CPU cycles.\n" ); }#ifdef RTE_LIBRTE_CMDLINE
@@ -677,6 +678,7 @@ launch_args_parse(int argc, char** argv) { "noisy-lkup-num-reads-writes", 1, 0, 0 }, { "no-iova-contig", 0, 0, 0 }, { "rx-mq-mode", 1, 0, 0 }, + { "record-core-cycles", 0, 0, 0 }, { 0, 0, 0, 0 }, };
@@ -1381,6 +1383,8 @@ launch_args_parse(int argc, char** argv)"rx-mq-mode must be >= 0 and <= %d\n" , ETH_MQ_RX_VMDQ_DCB_RSS); } +if (!strcmp (lgopts[opt_idx].name,"record-core-cycles" )) + record_core_cycles =1 ;break ;case 'h' :usage (argv[0 ]);
app/test-pmd/rxonly.c +3 -12
@@ -49,14 +49,9 @@ pkt_burst_receive(struct fwd_stream *fs)struct rte_mbuf *pkts_burst[MAX_PKT_BURST ];uint16_t nb_rx;uint16_t i; +uint64_t start_tsc =0 ; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES -uint64_t start_tsc; -uint64_t end_tsc; -uint64_t core_cycles; - - start_tsc =rte_rdtsc (); -#endif +get_start_cycles (&start_tsc);/* * Receive a burst of packets.
@@ -73,11 +68,7 @@ pkt_burst_receive(struct fwd_stream *fs)for (i =0 ; i < nb_rx; i++)rte_pktmbuf_free (pkts_burst[i]); -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - end_tsc =rte_rdtsc (); - core_cycles = (end_tsc - start_tsc); - fs->core_cycles = (uint64_t ) (fs->core_cycles + core_cycles); -#endif +get_end_cycles (fs, start_tsc); }struct fwd_engine rx_only_engine = {
app/test-pmd/testpmd.c +22 -22
79 lines not on this page · raw message
app/test-pmd/testpmd.h +16 -2
@@ -129,9 +129,7 @@ struct fwd_stream {uint64_t rx_bad_outer_l4_csum;/**< received packets has bad outer l4 checksum */ unsigned int gro_times;/**< GRO operation times */ -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES uint64_t core_cycles;/**< used for RX and TX processing */ -#endif #ifdef RTE_TEST_PMD_RECORD_BURST_STATS struct pkt_burst_stats rx_burst_stats;struct pkt_burst_stats tx_burst_stats;
@@ -301,6 +299,7 @@ extern uint16_t nb_rx_queue_stats_mappings;extern uint8_t xstats_hide_zero;/**< Hide zero values for xstats display */ /* globals used for configuration */ +extern uint8_t record_core_cycles;/**< Enables measurement of CPU cycles */ extern uint16_t verbose_level;/**< Drives messages being displayed, if any. */ extern int testpmd_logtype;/**< Log type for testpmd logs */ extern uint8_t interactive;
@@ -681,6 +680,20 @@ port_pci_reg_write(struct rte_port *port, uint32_t reg_off, uint32_t reg_v)#define port_id_pci_reg_write (pt_id, reg_off, reg_value) \ port_pci_reg_write(&ports[(pt_id)], (reg_off), (reg_value)) +static inline void +get_start_cycles (uint64_t *start_tsc) +{ +if (record_core_cycles) + *start_tsc =rte_rdtsc (); +} + +static inline void +get_end_cycles (struct fwd_stream *fs,uint64_t start_tsc) +{ +if (record_core_cycles) + fs->core_cycles +=rte_rdtsc () - start_tsc; +} +/* Prototypes */ unsigned int parse_item_list(char* str, const char* item_name, unsigned int max_items,
@@ -772,6 +785,7 @@ void set_qmap(portid_t port_id, uint8_t is_rx, uint16_t queue_id, uint8_t map_vavoid set_xstats_hide_zero (uint8_t on_off); +void set_record_core_cycles (uint8_t on_off);void set_verbose_level (uint16_t vb_level);void set_tx_pkt_segments (unsigned *seg_lengths,unsigned nb_segs);void show_tx_pkt_segments (void );
app/test-pmd/txonly.c +3 -13
@@ -294,15 +294,9 @@ pkt_burst_transmit(struct fwd_stream *fs)uint32_t retry;uint64_t ol_flags =0 ;uint64_t tx_offloads; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES -uint64_t start_tsc; -uint64_t end_tsc; -uint64_t core_cycles; -#endif +uint64_t start_tsc =0 ; -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - start_tsc =rte_rdtsc (); -#endif +get_start_cycles (&start_tsc); mbp =current_fwd_lcore ()->mbp; txp = &ports[fs->tx_port];
@@ -391,11 +385,7 @@ pkt_burst_transmit(struct fwd_stream *fs) }while (++nb_tx < nb_pkt); } -#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES - end_tsc =rte_rdtsc (); - core_cycles = (end_tsc - start_tsc); - fs->core_cycles = (uint64_t ) (fs->core_cycles + core_cycles); -#endif +get_end_cycles (fs, start_tsc); } static void
Signature · 1 line
2.17.1
PATCH 2/6 doc: add record-core-cycles to testpmd funcs doc
In reply to Dharmik ThakkarUpdate documentation to include description of 'record-core-cycles' runtime configuration.
--- doc/guides/testpmd_app_ug/run_app.rst | 6 ++++++ doc/guides/testpmd_app_ug/testpmd_funcs.rst | 15 +++++++++++++++ 2 files changed, 21 insertions(+)
doc/guides/testpmd_app_ug/run_app.rst +6 -0
@@ -489,3 +489,9 @@ The command line options are: The default value is 0x7:: ETH_MQ_RX_RSS_FLAG | ETH_MQ_RX_DCB_FLAG | ETH_MQ_RX_VMDQ_FLAG + +* ``--record-core-cycles`` + + Enable measurement of CPU cycles per packet. + +
doc/guides/testpmd_app_ug/testpmd_funcs.rst +15 -0
@@ -710,6 +710,21 @@ Set the forwarding ports hexadecimal mask:: This is equivalent to the ``--portmask`` command-line option. +set record-core-cycles +~~~~~~~~~~~~~~~~~~~~~~ + +Set the recording of CPU cycles:: + + testpmd> set record-core-cycles (on|off) + +Where: + +* ``on`` enables measurement of CPU cycles per packet. + +* ``off`` disables measurement of CPU cycles per packet. + +This is equivalent to the ``--record-core-cycles command-line`` option. + set burst ~~~~~~~~~
Signature · 1 line
2.17.1
PATCH 3/6 app/testpmd: add record-burst-stats runtime config
In reply to Dharmik ThakkarConvert CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS to a runtime configuration.
--- app/test-pmd/5tswap.c | 8 ++----- app/test-pmd/cmdline.c | 46 +++++++++++++++++++++++++++++++++++++++ app/test-pmd/config.c | 6 +++++ app/test-pmd/csumonly.c | 8 ++----- app/test-pmd/flowgen.c | 4 +--- app/test-pmd/icmpecho.c | 8 ++----- app/test-pmd/iofwd.c | 8 ++----- app/test-pmd/macfwd.c | 8 ++----- app/test-pmd/macswap.c | 8 ++----- app/test-pmd/parameters.c | 7 +++++- app/test-pmd/rxonly.c | 4 +--- app/test-pmd/testpmd.c | 33 ++++++++++++++-------------- app/test-pmd/testpmd.h | 20 +++++++++++++---- app/test-pmd/txonly.c | 4 +--- 14 files changed, 106 insertions(+), 66 deletions(-)
app/test-pmd/5tswap.c +2 -6
@@ -117,9 +117,7 @@ pkt_burst_5tuple_swap(struct fwd_stream *fs)if (unlikely (nb_rx ==0 ))return ; -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->rx_burst_stats.pkt_burst_spread[nb_rx]++; -#endif +inc_rx_burst_stats (fs, nb_rx); fs->rx_packets += nb_rx; txp = &ports[fs->tx_port];
@@ -178,9 +176,7 @@ pkt_burst_5tuple_swap(struct fwd_stream *fs) } } fs->tx_packets += nb_tx; -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->tx_burst_stats.pkt_burst_spread[nb_tx]++; -#endif +inc_tx_burst_stats (fs, nb_tx);if (unlikely (nb_tx < nb_rx)) { fs->fwd_dropped += (nb_rx - nb_tx); do {
app/test-pmd/cmdline.c +46 -0
64 lines not on this page · raw message
app/test-pmd/config.c +6 -0
@@ -3685,6 +3685,12 @@ set_record_core_cycles(uint8_t on_off) record_core_cycles = on_off; } +void +set_record_burst_stats (uint8_t on_off) +{ + record_burst_stats = on_off; +} +static inline void print_fdir_mask (struct rte_eth_fdir_masks *mask) {
app/test-pmd/csumonly.c +2 -6
@@ -796,9 +796,7 @@ pkt_burst_checksum_forward(struct fwd_stream *fs)/* receive a burst of packet */ nb_rx =rte_eth_rx_burst (fs->rx_port, fs->rx_queue, pkts_burst, nb_pkt_per_burst); -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->rx_burst_stats.pkt_burst_spread[nb_rx]++; -#endif +inc_rx_burst_stats (fs, nb_rx);if (unlikely (nb_rx ==0 ))return ;
@@ -1082,9 +1080,7 @@ pkt_burst_checksum_forward(struct fwd_stream *fs) fs->rx_bad_l4_csum += rx_bad_l4_csum; fs->rx_bad_outer_l4_csum += rx_bad_outer_l4_csum; -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->tx_burst_stats.pkt_burst_spread[nb_tx]++; -#endif +inc_tx_burst_stats (fs, nb_tx);if (unlikely (nb_tx < nb_rx)) { fs->fwd_dropped += (nb_rx - nb_tx); do {
app/test-pmd/flowgen.c +1 -3
@@ -188,9 +188,7 @@ pkt_burst_flow_gen(struct fwd_stream *fs) } fs->tx_packets += nb_tx; -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->tx_burst_stats.pkt_burst_spread[nb_tx]++; -#endif +inc_tx_burst_stats (fs, nb_tx);if (unlikely (nb_tx < nb_pkt)) {/* Back out the flow counter. */ next_flow -= (nb_pkt - nb_tx);
app/test-pmd/icmpecho.c +2 -6
@@ -302,9 +302,7 @@ reply_to_icmp_echo_rqsts(struct fwd_stream *fs)*/ nb_rx =rte_eth_rx_burst (fs->rx_port, fs->rx_queue, pkts_burst, nb_pkt_per_burst); -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->rx_burst_stats.pkt_burst_spread[nb_rx]++; -#endif +inc_rx_burst_stats (fs, nb_rx);if (unlikely (nb_rx ==0 ))return ;
@@ -503,9 +501,7 @@ reply_to_icmp_echo_rqsts(struct fwd_stream *fs) } } fs->tx_packets += nb_tx; -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->tx_burst_stats.pkt_burst_spread[nb_tx]++; -#endif +inc_tx_burst_stats (fs, nb_tx);if (unlikely (nb_tx < nb_replies)) { fs->fwd_dropped += (nb_replies - nb_tx); do {
app/test-pmd/iofwd.c +2 -6
@@ -59,9 +59,7 @@ pkt_burst_io_forward(struct fwd_stream *fs)*/ nb_rx =rte_eth_rx_burst (fs->rx_port, fs->rx_queue, pkts_burst, nb_pkt_per_burst); -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->rx_burst_stats.pkt_burst_spread[nb_rx]++; -#endif +inc_rx_burst_stats (fs, nb_rx);if (unlikely (nb_rx ==0 ))return ; fs->rx_packets += nb_rx;
@@ -80,9 +78,7 @@ pkt_burst_io_forward(struct fwd_stream *fs) } } fs->tx_packets += nb_tx; -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->tx_burst_stats.pkt_burst_spread[nb_tx]++; -#endif +inc_tx_burst_stats (fs, nb_tx);if (unlikely (nb_tx < nb_rx)) { fs->fwd_dropped += (nb_rx - nb_tx); do {
app/test-pmd/macfwd.c +2 -6
@@ -65,9 +65,7 @@ pkt_burst_mac_forward(struct fwd_stream *fs)*/ nb_rx =rte_eth_rx_burst (fs->rx_port, fs->rx_queue, pkts_burst, nb_pkt_per_burst); -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->rx_burst_stats.pkt_burst_spread[nb_rx]++; -#endif +inc_rx_burst_stats (fs, nb_rx);if (unlikely (nb_rx ==0 ))return ;
@@ -111,9 +109,7 @@ pkt_burst_mac_forward(struct fwd_stream *fs) } fs->tx_packets += nb_tx; -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->tx_burst_stats.pkt_burst_spread[nb_tx]++; -#endif +inc_tx_burst_stats (fs, nb_tx);if (unlikely (nb_tx < nb_rx)) { fs->fwd_dropped += (nb_rx - nb_tx); do {
app/test-pmd/macswap.c +2 -6
@@ -66,9 +66,7 @@ pkt_burst_mac_swap(struct fwd_stream *fs)*/ nb_rx =rte_eth_rx_burst (fs->rx_port, fs->rx_queue, pkts_burst, nb_pkt_per_burst); -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->rx_burst_stats.pkt_burst_spread[nb_rx]++; -#endif +inc_rx_burst_stats (fs, nb_rx);if (unlikely (nb_rx ==0 ))return ;
@@ -90,9 +88,7 @@ pkt_burst_mac_swap(struct fwd_stream *fs) } } fs->tx_packets += nb_tx; -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->tx_burst_stats.pkt_burst_spread[nb_tx]++; -#endif +inc_tx_burst_stats (fs, nb_tx);if (unlikely (nb_tx < nb_rx)) { fs->fwd_dropped += (nb_rx - nb_tx); do {
app/test-pmd/parameters.c +6 -1
@@ -70,7 +70,8 @@ usage(char* progname)"--rxpt= | --rxht= | --rxwt= | --rxfreet= | " "--txpt= | --txht= | --txwt= | --txfreet= | " "--txrst= | --tx-offloads= | | --rx-offloads= | " -"--vxlan-gpe-port= | --record-core-cycles]\n" , +"--vxlan-gpe-port= | --record-core-cycles | " +"--record-burst-stats]\n" , progname);#ifdef RTE_LIBRTE_CMDLINE printf (" --interactive: run in interactive mode.\n" );
@@ -217,6 +218,7 @@ usage(char* progname)printf (" --rx-mq-mode=0xX: hexadecimal bitmask of RX mq mode can be " "enabled\n" );printf (" --record-core-cycles: enable measurement of CPU cycles.\n" ); +printf (" --record-burst-stats: enable display of RX and TX bursts.\n" ); }#ifdef RTE_LIBRTE_CMDLINE
@@ -679,6 +681,7 @@ launch_args_parse(int argc, char** argv) { "no-iova-contig", 0, 0, 0 }, { "rx-mq-mode", 1, 0, 0 }, { "record-core-cycles", 0, 0, 0 }, + { "record-burst-stats", 0, 0, 0 }, { 0, 0, 0, 0 }, };
@@ -1385,6 +1388,8 @@ launch_args_parse(int argc, char** argv) }if (!strcmp (lgopts[opt_idx].name,"record-core-cycles" )) record_core_cycles =1 ; +if (!strcmp (lgopts[opt_idx].name,"record-burst-stats" )) + record_burst_stats =1 ;break ;case 'h' :usage (argv[0 ]);
app/test-pmd/rxonly.c +1 -3
@@ -58,9 +58,7 @@ pkt_burst_receive(struct fwd_stream *fs)*/ nb_rx =rte_eth_rx_burst (fs->rx_port, fs->rx_queue, pkts_burst, nb_pkt_per_burst); -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->rx_burst_stats.pkt_burst_spread[nb_rx]++; -#endif +inc_rx_burst_stats (fs, nb_rx);if (unlikely (nb_rx ==0 ))return ;
app/test-pmd/testpmd.c +17 -16
@@ -480,6 +480,11 @@ uint8_t xstats_hide_zero;*/ uint8_t record_core_cycles; +/* +* Display of RX and TX bursts disabled by default +*/ +uint8_t record_burst_stats; +unsigned int num_sockets =0 ;unsigned int socket_ids[RTE_MAX_NUMA_NODES ];
@@ -1684,7 +1689,6 @@ init_fwd_streams(void)return 0 ; } -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS static void pkt_burst_stats_display (const char *rx_tx,struct pkt_burst_stats *pbs) {
@@ -1751,7 +1755,6 @@ pkt_burst_stats_display(const char *rx_tx, struct pkt_burst_stats *pbs) sburstp += burst_percent[i]; } } -#endif /* RTE_TEST_PMD_RECORD_BURST_STATS */ static void fwd_stream_stats_display (streamid_t stream_id)
@@ -1782,10 +1785,10 @@ fwd_stream_stats_display(streamid_t stream_id)printf ("\n" ); } -#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); -#endif +if (record_burst_stats) { +pkt_burst_stats_display ("RX" , &fs->rx_burst_stats); +pkt_burst_stats_display ("TX" , &fs->tx_burst_stats); + } }void
@@ -1915,14 +1918,14 @@ fwd_stats_display(void) stats.opackets + ports_stats[pt_id].tx_dropped); } -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS -if (ports_stats[pt_id].rx_stream) -pkt_burst_stats_display ("RX" , - &ports_stats[pt_id].rx_stream->rx_burst_stats); -if (ports_stats[pt_id].tx_stream) -pkt_burst_stats_display ("TX" , - &ports_stats[pt_id].tx_stream->tx_burst_stats); -#endif +if (record_burst_stats) { +if (ports_stats[pt_id].rx_stream) +pkt_burst_stats_display ("RX" , + &ports_stats[pt_id].rx_stream->rx_burst_stats); +if (ports_stats[pt_id].tx_stream) +pkt_burst_stats_display ("TX" , + &ports_stats[pt_id].tx_stream->tx_burst_stats); + }if (port->rx_queue_stats_mapping_enabled) {printf ("\n" );
@@ -2004,10 +2007,8 @@ fwd_stats_reset(void) fs->rx_bad_l4_csum =0 ; fs->rx_bad_outer_l4_csum =0 ; -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS memset (&fs->rx_burst_stats,0 ,sizeof (fs->rx_burst_stats));memset (&fs->tx_burst_stats,0 ,sizeof (fs->tx_burst_stats)); -#endif fs->core_cycles =0 ; } }
app/test-pmd/testpmd.h +16 -4
@@ -76,7 +76,6 @@ enum {/**< allocate mempool natively, use rte_pktmbuf_pool_create_extbuf */ }; -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS /** * The data structure associated with RX and TX packet burst statistics * that are recorded for each forwarding stream.
@@ -84,7 +83,6 @@ enum {struct pkt_burst_stats {unsigned int pkt_burst_spread[MAX_PKT_BURST ]; }; -#endif /** Information for a given RSS type. */ struct rss_type_info {
@@ -130,10 +128,8 @@ struct fwd_stream {/**< received packets has bad outer l4 checksum */ unsigned int gro_times;/**< GRO operation times */ uint64_t core_cycles;/**< used for RX and TX processing */ -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS struct pkt_burst_stats rx_burst_stats;struct pkt_burst_stats tx_burst_stats; -#endif };/** Descriptor for a single flow. */
@@ -300,6 +296,7 @@ extern uint8_t xstats_hide_zero; /**< Hide zero values for xstats display *//* globals used for configuration */ extern uint8_t record_core_cycles;/**< Enables measurement of CPU cycles */ +extern uint8_t record_burst_stats;/**< Enables display of RX and TX bursts */ extern uint16_t verbose_level;/**< Drives messages being displayed, if any. */ extern int testpmd_logtype;/**< Log type for testpmd logs */ extern uint8_t interactive;
@@ -694,6 +691,20 @@ get_end_cycles(struct fwd_stream *fs, uint64_t start_tsc) fs->core_cycles +=rte_rdtsc () - start_tsc; } +static inline void +inc_rx_burst_stats (struct fwd_stream *fs,uint16_t nb_rx) +{ +if (record_burst_stats) + fs->rx_burst_stats.pkt_burst_spread[nb_rx]++; +} + +static inline void +inc_tx_burst_stats (struct fwd_stream *fs,uint16_t nb_tx) +{ +if (record_burst_stats) + fs->tx_burst_stats.pkt_burst_spread[nb_tx]++; +} +/* Prototypes */ unsigned int parse_item_list(char* str, const char* item_name, unsigned int max_items,
@@ -786,6 +797,7 @@ void set_qmap(portid_t port_id, uint8_t is_rx, uint16_t queue_id, uint8_t map_vavoid set_xstats_hide_zero (uint8_t on_off);void set_record_core_cycles (uint8_t on_off); +void set_record_burst_stats (uint8_t on_off);void set_verbose_level (uint16_t vb_level);void set_tx_pkt_segments (unsigned *seg_lengths,unsigned nb_segs);void show_tx_pkt_segments (void );
app/test-pmd/txonly.c +1 -3
@@ -369,9 +369,7 @@ pkt_burst_transmit(struct fwd_stream *fs)if (txonly_multi_flow)RTE_PER_LCORE (_ip_var) -= nb_pkt - nb_tx; -#ifdef RTE_TEST_PMD_RECORD_BURST_STATS - fs->tx_burst_stats.pkt_burst_spread[nb_tx]++; -#endif +inc_tx_burst_stats (fs, nb_tx);if (unlikely (nb_tx < nb_pkt)) {if (verbose_level >0 && fs->fwd_dropped ==0 ) printf("port %d tx_queue %d - drop "
Signature · 1 line
2.17.1
PATCH 4/6 doc: add record-burst-stats to testpmd funcs doc
In reply to Dharmik ThakkarUpdate documentation to include description of 'record-burst-stats' runtime configuration.
--- doc/guides/testpmd_app_ug/run_app.rst | 2 ++ doc/guides/testpmd_app_ug/testpmd_funcs.rst | 15 +++++++++++++++ 2 files changed, 17 insertions(+)
doc/guides/testpmd_app_ug/run_app.rst +2 -0
@@ -494,4 +494,6 @@ The command line options are: Enable measurement of CPU cycles per packet. +* ``--record-burst-stats`` + Enable display of RX and TX burst stats.
doc/guides/testpmd_app_ug/testpmd_funcs.rst +15 -0
@@ -725,6 +725,21 @@ Where: This is equivalent to the ``--record-core-cycles command-line`` option. +set record-burst-stats +~~~~~~~~~~~~~~~~~~~~~~ + +Set the displaying of RX and TX bursts:: + + testpmd> set record-burst-stats (on|off) + +Where: + +* ``on`` enables display of RX and TX bursts. + +* ``off`` disables display of RX and TX bursts. + +This is equivalent to the ``--record-burst-stats command-line`` option. + set burst ~~~~~~~~~
Signature · 1 line
2.17.1
PATCH 5/6 app/testpmd: enable burst stats for noisy vnf mode
In reply to Dharmik ThakkarFrom: Phil Yang <[email protected]>
Add burst stats for noisy vnf mode.
--- app/test-pmd/noisy_vnf.c | 4 ++++ 1 file changed, 4 insertions(+)
app/test-pmd/noisy_vnf.c +4 -0
@@ -154,6 +154,7 @@ pkt_burst_noisy_vnf(struct fwd_stream *fs) nb_rx =rte_eth_rx_burst (fs->rx_port, fs->rx_queue, pkts_burst, nb_pkt_per_burst); +inc_rx_burst_stats (fs, nb_rx);if (unlikely (nb_rx ==0 )) goto flush; fs->rx_packets += nb_rx;
@@ -164,6 +165,7 @@ pkt_burst_noisy_vnf(struct fwd_stream *fs) pkts_burst, nb_rx);if (unlikely (nb_tx < nb_rx) && fs->retry_enabled) nb_tx +=do_retry (nb_rx, nb_tx, pkts_burst, fs); +inc_tx_burst_stats (fs, nb_tx); fs->tx_packets += nb_tx; fs->fwd_dropped +=drop_pkts (pkts_burst, nb_rx, nb_tx);return ;
@@ -187,6 +189,7 @@ pkt_burst_noisy_vnf(struct fwd_stream *fs) nb_deqd);if (unlikely (nb_tx < nb_rx) && fs->retry_enabled) nb_tx +=do_retry (nb_rx, nb_tx, tmp_pkts, fs); +inc_tx_burst_stats (fs, nb_tx); fs->fwd_dropped +=drop_pkts (tmp_pkts, nb_deqd, nb_tx); } }
@@ -211,6 +214,7 @@ pkt_burst_noisy_vnf(struct fwd_stream *fs) tmp_pkts, nb_deqd);if (unlikely (sent < nb_deqd) && fs->retry_enabled) nb_tx +=do_retry (nb_rx, nb_tx, tmp_pkts, fs); +inc_tx_burst_stats (fs, nb_tx); fs->fwd_dropped +=drop_pkts (tmp_pkts, nb_deqd, sent); ncf->prev_time =rte_get_timer_cycles (); }
Signature · 1 line
2.17.1
PATCH 6/6 app/testpmd: enable empty polls in 5tswap
In reply to Dharmik ThakkarEnable empty polls in burst stats within 5tswap.c
--- app/test-pmd/5tswap.c | 3 +-- 1 file changed, 1 insertion(+), 2 deletions(-)
app/test-pmd/5tswap.c +1 -2
@@ -114,11 +114,10 @@ pkt_burst_5tuple_swap(struct fwd_stream *fs)*/ nb_rx =rte_eth_rx_burst (fs->rx_port, fs->rx_queue, pkts_burst, nb_pkt_per_burst); +inc_rx_burst_stats (fs, nb_rx);if (unlikely (nb_rx ==0 ))return ; -inc_rx_burst_stats (fs, nb_rx); - fs->rx_packets += nb_rx; txp = &ports[fs->tx_port]; ol_flags =ol_flags_init (txp->dev_conf.txmode.offloads);
Signature · 1 line
2.17.1
Re: [dpdk-dev] [PATCH 6/6] app/testpmd: enable empty polls in 5tswap
Quoted 9 lines
-----Original Message-----
From: dev <[email protected]> On Behalf Of Dharmik Thakkar
Sent: Wednesday, July 15, 2020 5:51 AM
To: Wenzhuo Lu <[email protected]>; Beilei Xing <[email protected]>; Bernard Iremonger <[email protected]>
Cc: [email protected]; nd <[email protected]>; Dharmik Thakkar <[email protected]>
Subject: [dpdk-dev] [PATCH 6/6] app/testpmd: enable empty polls in 5tswap
Quoted 4 lines
Enable empty polls in burst stats within
5tswap.cSigned-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>
Re: [dpdk-dev] [PATCH 0/6] app/testpmd: add runtime config
In reply to Dharmik ThakkarQuoted 7 lines · Dharmik Thakkar
Meson build system lacks support for
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATSconfiguration options.One solution is to add these options within meson_options.txt
Since adding these as runtime config causes no performance impact,
Hi Dharmik,
These are on the datapath, and even disable there will be additional checks, isn't it expected to have some impact?
Did you do any measurements for it?
Quoted 2 lines
it makes sense to enable these as runtime config alongside other available runtime options within testpmd.
If there is performance impact,
another option can be still using the compile time config with meson. As Jerin suggested previously, or Ed is doing in his patch right now, something like:
#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
const int record_core_cycles = 1;
#else
const int record_core_cycles = 0;
#endif
static inline void
get_start_cycles(uint64_t *start_tsc)
{
if (record_core_cycles)
*start_tsc = rte_rdtsc();
}
When meson compiled by default, compiler will optimize out the code during compile time.
To enable it, can provide the config option with CFLAGS, CFLAGS="-DRTE_TEST_PMD_RECORD_CORE_CYCLES" meson build
Quoted 30 lines
Dharmik Thakkar (5):
app/testpmd: add record-core-cycles runtime config
doc: add record-core-cycles to testpmd funcs doc app/testpmd: add record-burst-stats runtime config
doc: add record-burst-stats to testpmd funcs doc app/testpmd: enable empty polls in 5tswapPhil Yang (1):
app/testpmd: enable burst stats for noisy vnf modeapp/test-pmd/5tswap.c | 25 ++---- app/test-pmd/cmdline.c | 92 +++++++++++++++++++++ app/test-pmd/config.c | 12 +++ app/test-pmd/csumonly.c | 24 ++---- app/test-pmd/flowgen.c | 21 ++--- app/test-pmd/icmpecho.c | 24 ++---- app/test-pmd/iofwd.c | 26 ++---- app/test-pmd/macfwd.c | 25 ++---- app/test-pmd/macswap.c | 24 ++---- app/test-pmd/noisy_vnf.c | 4 + app/test-pmd/parameters.c | 11 ++- app/test-pmd/rxonly.c | 19 +---- app/test-pmd/testpmd.c | 77 ++++++++--------- app/test-pmd/testpmd.h | 38 +++++++-- app/test-pmd/txonly.c | 20 +---- doc/guides/testpmd_app_ug/run_app.rst | 8 ++ doc/guides/testpmd_app_ug/testpmd_funcs.rst | 30 +++++++ 17 files changed, 272 insertions(+), 208 deletions(-)
Re: [dpdk-dev] [PATCH 0/6] app/testpmd: add runtime config
Quoted 7 lines · Ferruh Yigit
Quoted 6 lines · Dharmik Thakkar
Meson build system lacks support for
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATSconfiguration options.One solution is to add these options within meson_options.txt
Quoted 8 lines
Quoted 1 line
Since adding these as runtime config causes no performance impact,
Hi Dharmik,
These are on the datapath, and even disable there will be additional checks, isn't it expected to have some impact? Did you do any measurements for it?
Branches that always predict the same way can be very cheap, and unless proven to be a problem, I'd see no issue with having a few on the datapath
- especially if it just one or two per burst. If we start seeing a significant number, or ones that occur for every packet, then we perhaps need to be more cautious.
I also think that using lots of different CFLAGS for turning things on and off is as bad - or even worse - than using the old build-time config options, just this time we don't have a single list of them somewhere! Therefore, I think we really need to start converting these to runtime options where we can.
Regards,
/Bruce
Re: [dpdk-dev] [PATCH 0/6] app/testpmd: add runtime config
Quoted 12 lines · Bruce Richardson
Quoted 7 lines · Ferruh Yigit
Quoted 6 lines · Dharmik Thakkar
Meson build system lacks support for
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATSconfiguration options.One solution is to add these options within meson_options.txt
Quoted 7 lines
Quoted 7 lines
Quoted 1 line
Since adding these as runtime config causes no performance impact,
Hi Dharmik,
These are on the datapath, and even disable there will be additional checks, isn't it expected to have some impact? Did you do any measurements for it?
Hi Ferruh,
In my measurements, I saw a maximum performance degradation of 0.3% for rx and tx throughput (pps) with runtime option (both disabled and enabled cases) when compared to compile time option.. There was no difference in cycles per packet measurement.
I did these measurements with Mellanox-ConnectX-5 card on N1SDP server with ’set fwd mac retry’ option.
Quoted 15 lines
Quoted 1 line
Branches that always predict the same way can be very cheap, and unless proven to be a problem, I'd see no issue with having a few on the datapath
- especially if it just one or two per burst. If we start seeing a significant number, or ones that occur for every packet, then we perhaps need to be more cautious.I also think that using lots of different CFLAGS for turning things on and off is as bad - or even worse - than using the old build-time config options, just this time we don't have a single list of them somewhere! Therefore, I think we really need to start converting these to runtime options where we can.
Regards,
/Bruce
Re: [dpdk-dev] [PATCH 0/6] app/testpmd: add runtime config
Quoted 20 lines · Dharmik Thakkar
Quoted 12 lines · Bruce Richardson
Quoted 7 lines · Ferruh Yigit
Quoted 6 lines · Dharmik Thakkar
Meson build system lacks support for
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATSconfiguration options.One solution is to add these options within meson_options.txt
Quoted 7 lines
Quoted 7 lines
Quoted 1 line
Since adding these as runtime config causes no performance impact,
Hi Dharmik,
These are on the datapath, and even disable there will be additional checks, isn't it expected to have some impact? Did you do any measurements for it?
Quoted 7 lines
Hi Ferruh,
In my measurements, I saw a maximum performance degradation of 0.3% for rx and tx throughput (pps) with runtime option (both disabled and enabled cases) when compared to compile time option.. There was no difference in cycles per packet measurement.
I did these measurements with Mellanox-ConnectX-5 card on N1SDP server with ’set fwd mac retry’ option.
Thanks Dharmik,
0.3% does not look a lot, I also don't get any recognizable drop when the stats are disabled (there is a drop (up to %15) when stats are enabled but that is OK). And there is a benefit to have the stats runtime configurable.
This minor drop for runtime configuration looks OK to me, but lets wait a little more if there is any strong objection to it, we can proceed afterwards.
Quoted 19 lines
Quoted 16 lines
Quoted 2 lines
Branches that always predict the same way can be very cheap, and unless proven to be a problem, I'd see no issue with having a few on the datapath
- especially if it just one or two per burst. If we start seeing a significant number, or ones that occur for every packet, then we perhaps need to be more cautious.I also think that using lots of different CFLAGS for turning things on and off is as bad - or even worse - than using the old build-time config options, just this time we don't have a single list of them somewhere! Therefore, I think we really need to start converting these to runtime options where we can.
Regards,
/Bruce
Ferruh Yigit
Reviewed-byRe: [dpdk-dev] [PATCH 0/6] app/testpmd: add runtime config
Quoted 29 lines · Ferruh Yigit
Quoted 20 lines · Dharmik Thakkar
Quoted 12 lines · Bruce Richardson
Quoted 7 lines · Ferruh Yigit
Quoted 6 lines · Dharmik Thakkar
Meson build system lacks support for
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATSconfiguration options.One solution is to add these options within meson_options.txt
Quoted 7 lines
Quoted 7 lines
Quoted 1 line
Since adding these as runtime config causes no performance impact,
Hi Dharmik,
These are on the datapath, and even disable there will be additional checks, isn't it expected to have some impact? Did you do any measurements for it?
Quoted 7 lines
Hi Ferruh,
In my measurements, I saw a maximum performance degradation of 0.3% for rx and tx throughput (pps) with runtime option (both disabled and enabled cases) when compared to compile time option.. There was no difference in cycles per packet measurement.
I did these measurements with Mellanox-ConnectX-5 card on N1SDP server with ’set fwd mac retry’ option.
Quoted 8 lines
Thanks Dharmik,
0.3% does not look a lot, I also don't get any recognizable drop when the stats are disabled (there is a drop (up to %15) when stats are enabled but that is OK). And there is a benefit to have the stats runtime configurable.
This minor drop for runtime configuration looks OK to me, but lets wait a little more if there is any strong objection to it, we can proceed afterwards.
There seems no objection to switching runtime configuration.
For series,
Series applied to dpdk-next-net/main, thanks.
(doc patches merged to the code while merging.)
Re: [dpdk-dev] [PATCH 0/6] app/testpmd: add runtime config
In reply to Bruce RichardsonQuoted 22 lines · Bruce Richardson
Quoted 7 lines · Ferruh Yigit
Quoted 6 lines · Dharmik Thakkar
Meson build system lacks support for
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATSconfiguration options.One solution is to add these options within meson_options.txt
Quoted 8 lines
Quoted 1 line
Since adding these as runtime config causes no performance impact,
Hi Dharmik,
These are on the datapath, and even disable there will be additional checks, isn't it expected to have some impact? Did you do any measurements for it?
Branches that always predict the same way can be very cheap, and unless proven to be a problem, I'd see no issue with having a few on the datapath
- especially if it just one or two per burst. If we start seeing a significant number, or ones that occur for every packet, then we perhaps need to be more cautious.
Quoted 5 lines
I also think that using lots of different CFLAGS for turning things on and off is as bad - or even worse - than using the old build-time config options, just this time we don't have a single list of them somewhere! Therefore, I think we really need to start converting these to runtime options where we can.
+1 using config options via 'CFLAGS' is worse. But if we want it to be compile time option because of performance, this is an option.
Re: [dpdk-dev] [PATCH v3 1/2] doc: add cycles per packet in testpmd user guide
In reply to Dharmik ThakkarQuoted 6 lines · Dharmik Thakkar
Update documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>
Ferruh Yigit
AppliedRe: [dpdk-dev] [PATCH v3 1/2] doc: add cycles per packet in testpmd user guide
Quoted 10 lines · Ferruh Yigit
Quoted 6 lines · Dharmik Thakkar
Update documentation for 'show fwd' testpmd runtime function to show CPU cycles/packet example.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>Reviewed-by Ferruh Yigit <[email protected]>
Applied to dpdk-next-net/master, thanks.
(note that not all patches in the patchset applied, only 1/2 applied)
PATCH v2 4/5 doc: include config options in testpmd user guide
In reply to Dharmik ThakkarUpdate testpmd documentation to include RECORD configuration options, CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES and CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.
---
v2:
- Remove extra '#'.
--- doc/guides/testpmd_app_ug/build_app.rst | 12 ++++++++++++ 1 file changed, 12 insertions(+)
doc/guides/testpmd_app_ug/build_app.rst +12 -0
@@ -21,6 +21,18 @@ The basic compilation steps are: export RTE_TARGET=x86_64-native-linux-gcc +#. If required, enable configuration options. For example: + + .. code-block:: console + + cd to the top-level DPDK directory + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES\)=n,\1=y,' config/common_base + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS\)=n,\1=y,' config/common_base + + Enabling CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES enables measurement of CPU cycles. + + Enabling CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS enables display of RX and TX bursts. + #. Build the application: .. code-block:: console
Signature · 1 line
2.20.1
Re: [dpdk-dev] [PATCH v2 4/5] doc: include config options in testpmd user guide
Quoted 12 lines
-----Original Message-----
From: Dharmik Thakkar <[email protected]>
Sent: Friday, May 8, 2020 11:38 PM
To: Lu, Wenzhuo <[email protected]>; Wu, Jingjing <[email protected]>; Iremonger, Bernard <[email protected]>; Mcnamara, John <[email protected]>; Kovacevic, Marko <[email protected]>
Cc: [email protected]; [email protected]; Dharmik Thakkar <[email protected]>
Subject: [PATCH v2 4/5] doc: include config options in testpmd user guide
Quoted 7 lines
Update testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>
Re: [dpdk-dev] [PATCH v2 4/5] doc: include config options in testpmd user guide
Quoted 9 lines
Quoted 7 lines
Update testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>
How these options are managed with meson?
Re: [dpdk-dev] [PATCH v2 4/5] doc: include config options in testpmd user guide
Quoted 8 lines · Thomas Monjalon
Quoted 5 lines
Quoted 5 lines
Update testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.
Quoted 9 lines
Quoted 5 lines
Quoted 3 lines
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>How these options are managed with meson?
As per my understanding, currently, this is not implemented with meson.
With ‘make’, the configuration options are saved within ./build/include/rte_config.h ( which gets generated during make config …). But this file (rte_config.h) does not get generated when using meson.
Re: [dpdk-dev] [PATCH v2 4/5] doc: include config options in testpmd user guide
20/05/2020 00:58, Dharmik Thakkar:
Quoted 9 lines
Quoted 9 lines · Thomas Monjalon
Quoted 8 lines
Quoted 7 lines
Update testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>
Quoted 7 lines
Quoted 3 lines
Quoted 1 line
How these options are managed with meson?
As per my understanding, currently, this is not implemented with meson.
With ‘make’, the configuration options are saved within./build/include/rte_config.h( which gets generated during make config …). But this file (rte_config.h) does not get generated when using meson.
That's also my understanding.
There is a gap which needs to be fixed, not sure what is the best approach. Can it be made a runtime option?
Re: [dpdk-dev] [PATCH v2 4/5] doc: include config options in testpmd user guide
Quoted 16 lines · Thomas Monjalon
20/05/2020 00:58, Dharmik Thakkar:
Quoted 11 lines
Quoted 11 lines · Thomas Monjalon
Quoted 9 lines
Quoted 7 lines
Update testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>
Quoted 8 lines
Quoted 5 lines
Quoted 1 line
How these options are managed with meson?
As per my understanding, currently, this is not implemented with meson.
With ‘make’, the configuration options are saved within./build/include/rte_config.h( which gets generated during make config …). But this file (rte_config.h) does not get generated when using meson.That's also my understanding.
There is a gap which needs to be fixed, not sure what is the best approach.
IMO, for now, it is best to replicate what is being done for ‘make’.
Quoted 1 line
Can it be made a runtime option?
Yes, it can be made a runtime option but that will be a separate patch.
Quoted 1 line
Re: [dpdk-dev] [PATCH v2 4/5] doc: include config options in testpmd user guide
In reply to Dharmik Thakkar09/05/2020 00:38, Dharmik Thakkar:
Quoted 8 lines
Update testpmd documentation to include RECORD configuration options,
CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLESandCONFIG_RTE_TEST_PMD_RECORD_BURST_STATS.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>---
doc/guides/testpmd_app_ug/build_app.rst The basic compilation steps are: · line 30
Quoted 7 lines
doc/guides/testpmd_app_ug/build_app.rst +7
-0+#. If required, enable configuration options. For example: + + .. code-block:: console + + cd to the top-level DPDK directory + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_CORE_CYCLES\)=n,\1=y,' config/common_base + sed -i 's,\(CONFIG_RTE_TEST_PMD_RECORD_BURST_STATS\)=n,\1=y,' config/common_base
Changing source code is wrong.
Only the generated .config file should be changed.
Not even speaking about meson...
Why we don't have any review on documentation patches?
Patch dropped from next-net pulling.
PATCH v2 5/5 doc: add aarch64 generic counter section
In reply to Dharmik ThakkarAdd a separate section for low-resolution generic counter for ARM64 profiling methods.
---
v2:
- Add Acked-by tag
--- doc/guides/prog_guide/profile_app.rst | 11 ++++++++--- 1 file changed, 8 insertions(+), 3 deletions(-)
doc/guides/prog_guide/profile_app.rst +8 -3
@@ -57,11 +57,16 @@ For more derails refer to the `ARM64 specific PMU events enumeration <http://infocenter.arm.com/help/index.jsp?topic=/com.arm.doc.100095_0002_04_en/way1382543438508.html>`_. -High-resolution cycle counter -~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ +Low-resolution generic counter +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The default ``cntvct_el0`` based ``rte_rdtsc()`` provides a portable means to -get a wall clock counter in user space. Typically it runs at <= 100MHz. +get a wall clock counter in user space. Typically it runs at a lower clock frequency than the CPU clock frequency. +Cycles counted using this method should be scaled to CPU clock frequency. + + +High-resolution cycle counter +~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ The alternative method to enable ``rte_rdtsc()`` for a high resolution wall clock counter is through the ARMv8 PMU subsystem. The PMU cycle counter runs
Signature · 1 line
2.20.1
Quoted 10 lines
-----Original Message-----
From: Dharmik Thakkar <[email protected]>
Sent: Friday, May 8, 2020 11:38 PM
To: Lu, Wenzhuo <[email protected]>; Wu, Jingjing <[email protected]>; Iremonger, Bernard <[email protected]>
Cc: [email protected]; [email protected]; Dharmik Thakkar <[email protected]>
Subject: [PATCH v2 1/5] app/testpmd: print clock with CPU cycles per pkt
Quoted 8 lines
On aarch64 platforms, the cycles are counted using either a low-resolution generic counter or a high-resolution PMU cycle counter.
Print the clock frequency along with CPU cycles/packet to identify which cycle counter is being used.Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>
Quoted 8 lines · Dharmik Thakkar
On aarch64 platforms, the cycles are counted using either a low-resolution generic counter or a high-resolution PMU cycle counter. Print the clock frequency along with CPU cycles/packet to identify which cycle counter is being used.
Signed-off-by Dharmik Thakkar <[email protected]>Reviewed-by Honnappa Nagarahalli <[email protected]>Reviewed-by Phil Yang <[email protected]>
Since 3/5 is independent patch, to not cause that patch hold the patchset, I will merge rest while we can keep discuss the 3/5.
Except 3/5
Series applied to dpdk-next-net/master, thanks.