[2/5] app/testpmd: print fractional part in CPU cycles
Checks
Commit Message
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)
Signed-off-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
Reviewed-by: Phil Yang <phil.yang@arm.com>
---
app/test-pmd/testpmd.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
Comments
> -----Original Message-----
> From: dev <dev-bounces@dpdk.org> On Behalf Of Dharmik Thakkar
> Sent: Wednesday, May 6, 2020 10:59 PM
> To: Lu, Wenzhuo <wenzhuo.lu@intel.com>; Wu, Jingjing <jingjing.wu@intel.com>; Iremonger, Bernard <bernard.iremonger@intel.com>
> Cc: dev@dpdk.org; nd@arm.com; Dharmik Thakkar <dharmik.thakkar@arm.com>
> Subject: [dpdk-dev] [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)
>
> Signed-off-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
> Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
> Reviewed-by: Phil Yang <phil.yang@arm.com>
> ---
> app/test-pmd/testpmd.c | 4 ++--
> 1 file changed, 2 insertions(+), 2 deletions(-)
>
> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
> index 9a8cbbd6fc7c..9444a730a153 100644
> --- a/app/test-pmd/testpmd.c
> +++ b/app/test-pmd/testpmd.c
> @@ -1955,9 +1955,9 @@ fwd_stats_display(void)
> #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
> #define CYC_PER_MHZ 1E6
> if (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.
> fwd_cycles, total_recv, (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));
> #endif
> }
> --
> 2.20.1
Hi Konstantin,
> On May 7, 2020, at 4:50 AM, Ananyev, Konstantin <konstantin.ananyev@intel.com> wrote:
>
>
>
>> -----Original Message-----
>> From: dev <dev-bounces@dpdk.org> On Behalf Of Dharmik Thakkar
>> Sent: Wednesday, May 6, 2020 10:59 PM
>> To: Lu, Wenzhuo <wenzhuo.lu@intel.com>; Wu, Jingjing <jingjing.wu@intel.com>; Iremonger, Bernard <bernard.iremonger@intel.com>
>> Cc: dev@dpdk.org; nd@arm.com; Dharmik Thakkar <dharmik.thakkar@arm.com>
>> Subject: [dpdk-dev] [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)
>>
>> Signed-off-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
>> Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>> ---
>> app/test-pmd/testpmd.c | 4 ++--
>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>
>> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
>> index 9a8cbbd6fc7c..9444a730a153 100644
>> --- a/app/test-pmd/testpmd.c
>> +++ b/app/test-pmd/testpmd.c
>> @@ -1955,9 +1955,9 @@ fwd_stats_display(void)
>> #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
>> #define CYC_PER_MHZ 1E6
>> if (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?
>
>> fwd_cycles, total_recv, (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));
>> #endif
>> }
>> --
>> 2.20.1
> Hi Konstantin,
>
> > On May 7, 2020, at 4:50 AM, Ananyev, Konstantin <konstantin.ananyev@intel.com> wrote:
> >
> >
> >
> >> -----Original Message-----
> >> From: dev <dev-bounces@dpdk.org> On Behalf Of Dharmik Thakkar
> >> Sent: Wednesday, May 6, 2020 10:59 PM
> >> To: Lu, Wenzhuo <wenzhuo.lu@intel.com>; Wu, Jingjing <jingjing.wu@intel.com>; Iremonger, Bernard <bernard.iremonger@intel.com>
> >> Cc: dev@dpdk.org; nd@arm.com; Dharmik Thakkar <dharmik.thakkar@arm.com>
> >> Subject: [dpdk-dev] [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)
> >>
> >> Signed-off-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
> >> Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
> >> Reviewed-by: Phil Yang <phil.yang@arm.com>
> >> ---
> >> app/test-pmd/testpmd.c | 4 ++--
> >> 1 file changed, 2 insertions(+), 2 deletions(-)
> >>
> >> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
> >> index 9a8cbbd6fc7c..9444a730a153 100644
> >> --- a/app/test-pmd/testpmd.c
> >> +++ b/app/test-pmd/testpmd.c
> >> @@ -1955,9 +1955,9 @@ fwd_stats_display(void)
> >> #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
> >> #define CYC_PER_MHZ 1E6
> >> if (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?
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
?
>
> >
> >> fwd_cycles, total_recv, (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));
> >> #endif
> >> }
> >> --
> >> 2.20.1
> On May 8, 2020, at 12:17 PM, Ananyev, Konstantin <konstantin.ananyev@intel.com> wrote:
>
>> Hi Konstantin,
>>
>>> On May 7, 2020, at 4:50 AM, Ananyev, Konstantin <konstantin.ananyev@intel.com> wrote:
>>>
>>>
>>>
>>>> -----Original Message-----
>>>> From: dev <dev-bounces@dpdk.org> On Behalf Of Dharmik Thakkar
>>>> Sent: Wednesday, May 6, 2020 10:59 PM
>>>> To: Lu, Wenzhuo <wenzhuo.lu@intel.com>; Wu, Jingjing <jingjing.wu@intel.com>; Iremonger, Bernard <bernard.iremonger@intel.com>
>>>> Cc: dev@dpdk.org; nd@arm.com; Dharmik Thakkar <dharmik.thakkar@arm.com>
>>>> Subject: [dpdk-dev] [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)
>>>>
>>>> Signed-off-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
>>>> Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
>>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
>>>> ---
>>>> app/test-pmd/testpmd.c | 4 ++--
>>>> 1 file changed, 2 insertions(+), 2 deletions(-)
>>>>
>>>> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
>>>> index 9a8cbbd6fc7c..9444a730a153 100644
>>>> --- a/app/test-pmd/testpmd.c
>>>> +++ b/app/test-pmd/testpmd.c
>>>> @@ -1955,9 +1955,9 @@ fwd_stats_display(void)
>>>> #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
>>>> #define CYC_PER_MHZ 1E6
>>>> if (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?
>
> 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
>
>
>
>>
>>>
>>>> fwd_cycles, total_recv, (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));
>>>> #endif
>>>> }
>>>> --
>>>> 2.20.1
> -----Original Message-----
> From: Dharmik Thakkar <Dharmik.Thakkar@arm.com>
> Sent: Friday, May 8, 2020 6:37 PM
> To: Ananyev, Konstantin <konstantin.ananyev@intel.com>
> Cc: Lu, Wenzhuo <wenzhuo.lu@intel.com>; Wu, Jingjing <jingjing.wu@intel.com>; Iremonger, Bernard <bernard.iremonger@intel.com>;
> dev@dpdk.org; nd <nd@arm.com>
> Subject: Re: [dpdk-dev] [PATCH 2/5] app/testpmd: print fractional part in CPU cycles
>
>
>
> > On May 8, 2020, at 12:17 PM, Ananyev, Konstantin <konstantin.ananyev@intel.com> wrote:
> >
> >> Hi Konstantin,
> >>
> >>> On May 7, 2020, at 4:50 AM, Ananyev, Konstantin <konstantin.ananyev@intel.com> wrote:
> >>>
> >>>
> >>>
> >>>> -----Original Message-----
> >>>> From: dev <dev-bounces@dpdk.org> On Behalf Of Dharmik Thakkar
> >>>> Sent: Wednesday, May 6, 2020 10:59 PM
> >>>> To: Lu, Wenzhuo <wenzhuo.lu@intel.com>; Wu, Jingjing <jingjing.wu@intel.com>; Iremonger, Bernard
> <bernard.iremonger@intel.com>
> >>>> Cc: dev@dpdk.org; nd@arm.com; Dharmik Thakkar <dharmik.thakkar@arm.com>
> >>>> Subject: [dpdk-dev] [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)
> >>>>
> >>>> Signed-off-by: Dharmik Thakkar <dharmik.thakkar@arm.com>
> >>>> Reviewed-by: Honnappa Nagarahalli <honnappa.nagarahalli@arm.com>
> >>>> Reviewed-by: Phil Yang <phil.yang@arm.com>
> >>>> ---
> >>>> app/test-pmd/testpmd.c | 4 ++--
> >>>> 1 file changed, 2 insertions(+), 2 deletions(-)
> >>>>
> >>>> diff --git a/app/test-pmd/testpmd.c b/app/test-pmd/testpmd.c
> >>>> index 9a8cbbd6fc7c..9444a730a153 100644
> >>>> --- a/app/test-pmd/testpmd.c
> >>>> +++ b/app/test-pmd/testpmd.c
> >>>> @@ -1955,9 +1955,9 @@ fwd_stats_display(void)
> >>>> #ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
> >>>> #define CYC_PER_MHZ 1E6
> >>>> if (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?
> >
> > 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.
>
> >
> >
> >
> >>
> >>>
> >>>> fwd_cycles, total_recv, (uint64_t)(rte_get_tsc_hz() / CYC_PER_MHZ));
> >>>> #endif
> >>>> }
> >>>> --
> >>>> 2.20.1
@@ -1955,9 +1955,9 @@ fwd_stats_display(void)
#ifdef RTE_TEST_PMD_RECORD_CORE_CYCLES
#define CYC_PER_MHZ 1E6
if (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
}