app/dma-perf: add average latency per worker

Message ID 20231219164021.858-1-vipin.varghese@amd.com (mailing list archive)
State Superseded, archived
Delegated to: Thomas Monjalon
Headers
Series app/dma-perf: add average latency per worker |

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/loongarch-compilation success Compilation OK
ci/loongarch-unit-testing success Unit Testing PASS
ci/github-robot: build success github build: passed
ci/Intel-compilation success Compilation OK
ci/intel-Testing success Testing PASS
ci/intel-Functional success Functional PASS
ci/iol-mellanox-Performance success Performance Testing PASS
ci/iol-intel-Performance success Performance Testing PASS
ci/iol-intel-Functional success Functional Testing PASS
ci/iol-broadcom-Performance success Performance Testing PASS
ci/iol-broadcom-Functional success Functional Testing PASS
ci/iol-abi-testing success Testing PASS
ci/iol-unit-amd64-testing success Testing PASS
ci/iol-unit-arm64-testing success Testing PASS
ci/iol-compile-arm64-testing success Testing PASS
ci/iol-compile-amd64-testing success Testing PASS
ci/iol-sample-apps-testing success Testing PASS

Commit Message

Vipin Varghese Dec. 19, 2023, 4:40 p.m. UTC
  Modify the user display data with total average latency per worker.

Signed-off-by: Vipin Varghese <vipin.varghese@amd.com>
---
 app/test-dma-perf/benchmark.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)
  

Comments

Chengwen Feng Feb. 23, 2024, 9:45 a.m. UTC | #1
Hi Vipin,

On 2023/12/20 0:40, Vipin Varghese wrote:
> Modify the user display data with total average latency per worker.
> 
> Signed-off-by: Vipin Varghese <vipin.varghese@amd.com>
> ---
>  app/test-dma-perf/benchmark.c | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
> 
> diff --git a/app/test-dma-perf/benchmark.c b/app/test-dma-perf/benchmark.c
> index 9b1f58c78c..8b6886af62 100644
> --- a/app/test-dma-perf/benchmark.c
> +++ b/app/test-dma-perf/benchmark.c
> @@ -470,7 +470,8 @@ mem_copy_benchmark(struct test_configure *cfg, bool is_dma)
>  		bandwidth_total += bandwidth;
>  		avg_cycles_total += avg_cycles;
>  	}
> -	printf("\nTotal Bandwidth: %.3lf Gbps, Total MOps: %.3lf\n", bandwidth_total, mops_total);
> +	printf("\nAverage Cycles/op: %.2lf, Total Bandwidth: %.3lf Gbps, Total MOps: %.3lf\n",
> +			(float) avg_cycles_total / nb_workers, bandwidth_total, mops_total);

Because this is total stats, suggest add Total prefix, e.g. "Total Average Cycles/op"

I think print format keep one-digit precision is enough. Also please modify CSV_TOTAL_LINE_FMT
make sure the csv also have same precision of Cycles/op.

Thanks

>  	snprintf(output_str[MAX_WORKER_NB], MAX_OUTPUT_STR_LEN, CSV_TOTAL_LINE_FMT,
>  			cfg->scenario_id, nr_buf, memory * nb_workers,
>  			avg_cycles_total / nb_workers, bandwidth_total, mops_total);
>
  
Vipin Varghese Feb. 27, 2024, 9:50 a.m. UTC | #2
On 2/23/2024 3:15 PM, fengchengwen wrote:
> Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.
>
>
> Hi Vipin,
>
> On 2023/12/20 0:40, Vipin Varghese wrote:
>> Modify the user display data with total average latency per worker.
>>
>> Signed-off-by: Vipin Varghese <vipin.varghese@amd.com>
>> ---
>>   app/test-dma-perf/benchmark.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/app/test-dma-perf/benchmark.c b/app/test-dma-perf/benchmark.c
>> index 9b1f58c78c..8b6886af62 100644
>> --- a/app/test-dma-perf/benchmark.c
>> +++ b/app/test-dma-perf/benchmark.c
>> @@ -470,7 +470,8 @@ mem_copy_benchmark(struct test_configure *cfg, bool is_dma)
>>                bandwidth_total += bandwidth;
>>                avg_cycles_total += avg_cycles;
>>        }
>> -     printf("\nTotal Bandwidth: %.3lf Gbps, Total MOps: %.3lf\n", bandwidth_total, mops_total);
>> +     printf("\nAverage Cycles/op: %.2lf, Total Bandwidth: %.3lf Gbps, Total MOps: %.3lf\n",
>> +                     (float) avg_cycles_total / nb_workers, bandwidth_total, mops_total);

thanks for the suggestions, please find my observations below

> Because this is total stats, suggest add Total prefix, e.g. "Total Average Cycles/op"
I did not follow this, so please let me try to explain my understanding. 
For `n` operation we count the average cycles, then
we add the cycles to form `total average cycles`; this is then divide by 
`n` operations. Making this per operation what is the
average cycles taken for the round trip time. Hence `Total Average 
Cyeles/op` does not sound right, but `Average Cycles / op` does.
> I think print format keep one-digit precision is enough. Also please modify CSV_TOTAL_LINE_FMT
> make sure the csv also have same precision of Cycles/op.

We have checked the CSV formatting a find
1. the precision for average cycle/op is 2 digits precision.
2. already the CVS format has average cycles/op integrated.

Hence no change is required.

>
> Thanks
>
>>        snprintf(output_str[MAX_WORKER_NB], MAX_OUTPUT_STR_LEN, CSV_TOTAL_LINE_FMT,
>>                        cfg->scenario_id, nr_buf, memory * nb_workers,
>>                        avg_cycles_total / nb_workers, bandwidth_total, mops_total);
>>
  
Chengwen Feng Feb. 27, 2024, 1:09 p.m. UTC | #3
Hi Vipin,

On 2024/2/27 17:50, Varghese, Vipin wrote:
> 
> On 2/23/2024 3:15 PM, fengchengwen wrote:
>> Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.
>>
>>
>> Hi Vipin,
>>
>> On 2023/12/20 0:40, Vipin Varghese wrote:
>>> Modify the user display data with total average latency per worker.
>>>
>>> Signed-off-by: Vipin Varghese <vipin.varghese@amd.com>
>>> ---
>>>   app/test-dma-perf/benchmark.c | 3 ++-
>>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/app/test-dma-perf/benchmark.c b/app/test-dma-perf/benchmark.c
>>> index 9b1f58c78c..8b6886af62 100644
>>> --- a/app/test-dma-perf/benchmark.c
>>> +++ b/app/test-dma-perf/benchmark.c
>>> @@ -470,7 +470,8 @@ mem_copy_benchmark(struct test_configure *cfg, bool is_dma)
>>>                bandwidth_total += bandwidth;
>>>                avg_cycles_total += avg_cycles;
>>>        }
>>> -     printf("\nTotal Bandwidth: %.3lf Gbps, Total MOps: %.3lf\n", bandwidth_total, mops_total);
>>> +     printf("\nAverage Cycles/op: %.2lf, Total Bandwidth: %.3lf Gbps, Total MOps: %.3lf\n",
>>> +                     (float) avg_cycles_total / nb_workers, bandwidth_total, mops_total);
> 
> thanks for the suggestions, please find my observations below
> 
>> Because this is total stats, suggest add Total prefix, e.g. "Total Average Cycles/op"
> I did not follow this, so please let me try to explain my understanding. For `n` operation we count the average cycles, then
> we add the cycles to form `total average cycles`; this is then divide by `n` operations. Making this per operation what is the
> average cycles taken for the round trip time. Hence `Total Average Cyeles/op` does not sound right, but `Average Cycles / op` does.

OK
My intention was to differentiate (since they have the same beginning).

>> I think print format keep one-digit precision is enough. Also please modify CSV_TOTAL_LINE_FMT
>> make sure the csv also have same precision of Cycles/op.
> 
> We have checked the CSV formatting a find
> 1. the precision for average cycle/op is 2 digits precision.
> 2. already the CVS format has average cycles/op integrated.
> 
> Hence no change is required.

It's weird.

We can see more clear when add together:
#define CSV_TOTAL_LINE_FMT "Scenario %u Summary, , , , , ,%u,%.2lf,%u,%.3lf,%.3lf\n"
	snprintf(output_str[MAX_WORKER_NB], MAX_OUTPUT_STR_LEN, CSV_TOTAL_LINE_FMT,
			cfg->scenario_id, nr_buf, memory * nb_workers,
			avg_cycles_total / nb_workers, bandwidth_total, mops_total);
The bandwidth_total, mops_total both are float, which take last two "%.31f", then "%u" is for "avg_cycles_total / nb_workers"

Also, avg_cycles_total / nb_workers both are u32, and its result is u32 if not with force convert.

You could modify with "avg_cycles_total*1.0 / nb_workers", then compile will output warning:

[2035/3727] Compiling C object app/dpdk-test-dma-perf.p/test-dma-perf_benchmark.c.o
../../dpdk/app/test-dma-perf/benchmark.c: In function ‘mem_copy_benchmark’:
../../dpdk/app/test-dma-perf/benchmark.c:26:28: warning: format ‘%u’ expects argument of type ‘unsigned int’, but argument 7 has type ‘double’ [-Wformat=]
 #define CSV_TOTAL_LINE_FMT "Scenario %u Summary, , , , , ,%u,%.2lf,%u,%.3lf,%.3lf\n"

Thanks
                            ^
> 
>>
>> Thanks
>>
>>>        snprintf(output_str[MAX_WORKER_NB], MAX_OUTPUT_STR_LEN, CSV_TOTAL_LINE_FMT,
>>>                        cfg->scenario_id, nr_buf, memory * nb_workers,
>>>                        avg_cycles_total / nb_workers, bandwidth_total, mops_total);
>>>
> .
  
Vipin Varghese Feb. 28, 2024, 3:05 a.m. UTC | #4
On 2/23/2024 3:15 PM, fengchengwen wrote:
> Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.
>
>
> Hi Vipin,
>
> On 2023/12/20 0:40, Vipin Varghese wrote:
>> Modify the user display data with total average latency per worker.
>>
>> Signed-off-by: Vipin Varghese <vipin.varghese@amd.com>
>> ---
>>   app/test-dma-perf/benchmark.c | 3 ++-
>>   1 file changed, 2 insertions(+), 1 deletion(-)
>>
>> diff --git a/app/test-dma-perf/benchmark.c b/app/test-dma-perf/benchmark.c
>> index 9b1f58c78c..8b6886af62 100644
>> --- a/app/test-dma-perf/benchmark.c
>> +++ b/app/test-dma-perf/benchmark.c
>> @@ -470,7 +470,8 @@ mem_copy_benchmark(struct test_configure *cfg, bool is_dma)
>>                bandwidth_total += bandwidth;
>>                avg_cycles_total += avg_cycles;
>>        }
>> -     printf("\nTotal Bandwidth: %.3lf Gbps, Total MOps: %.3lf\n", bandwidth_total, mops_total);
>> +     printf("\nAverage Cycles/op: %.2lf, Total Bandwidth: %.3lf Gbps, Total MOps: %.3lf\n",
>> +                     (float) avg_cycles_total / nb_workers, bandwidth_total, mops_total);
> Because this is total stats, suggest add Total prefix, e.g. "Total Average Cycles/op"
>
> I think print format keep one-digit precision is enough. Also please modify CSV_TOTAL_LINE_FMT
> make sure the csv also have same precision of Cycles/op.

Thank you Chengwen, let me share v2 shortly.


>
> Thanks
>
>>        snprintf(output_str[MAX_WORKER_NB], MAX_OUTPUT_STR_LEN, CSV_TOTAL_LINE_FMT,
>>                        cfg->scenario_id, nr_buf, memory * nb_workers,
>>                        avg_cycles_total / nb_workers, bandwidth_total, mops_total);
>>
  
Vipin Varghese Feb. 28, 2024, 3:06 a.m. UTC | #5
Addressed
On 2/27/2024 6:39 PM, fengchengwen wrote:
> Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.
>
>
> Hi Vipin,
>
> On 2024/2/27 17:50, Varghese, Vipin wrote:
>> On 2/23/2024 3:15 PM, fengchengwen wrote:
>>> Caution: This message originated from an External Source. Use proper caution when opening attachments, clicking links, or responding.
>>>
>>>
>>> Hi Vipin,
>>>
>>> On 2023/12/20 0:40, Vipin Varghese wrote:
>>>> Modify the user display data with total average latency per worker.
>>>>
>>>> Signed-off-by: Vipin Varghese <vipin.varghese@amd.com>
>>>> ---
>>>>    app/test-dma-perf/benchmark.c | 3 ++-
>>>>    1 file changed, 2 insertions(+), 1 deletion(-)
>>>>
>>>> diff --git a/app/test-dma-perf/benchmark.c b/app/test-dma-perf/benchmark.c
>>>> index 9b1f58c78c..8b6886af62 100644
>>>> --- a/app/test-dma-perf/benchmark.c
>>>> +++ b/app/test-dma-perf/benchmark.c
>>>> @@ -470,7 +470,8 @@ mem_copy_benchmark(struct test_configure *cfg, bool is_dma)
>>>>                 bandwidth_total += bandwidth;
>>>>                 avg_cycles_total += avg_cycles;
>>>>         }
>>>> -     printf("\nTotal Bandwidth: %.3lf Gbps, Total MOps: %.3lf\n", bandwidth_total, mops_total);
>>>> +     printf("\nAverage Cycles/op: %.2lf, Total Bandwidth: %.3lf Gbps, Total MOps: %.3lf\n",
>>>> +                     (float) avg_cycles_total / nb_workers, bandwidth_total, mops_total);
>> thanks for the suggestions, please find my observations below
>>
>>> Because this is total stats, suggest add Total prefix, e.g. "Total Average Cycles/op"
>> I did not follow this, so please let me try to explain my understanding. For `n` operation we count the average cycles, then
>> we add the cycles to form `total average cycles`; this is then divide by `n` operations. Making this per operation what is the
>> average cycles taken for the round trip time. Hence `Total Average Cyeles/op` does not sound right, but `Average Cycles / op` does.
> OK
> My intention was to differentiate (since they have the same beginning).
>
>>> I think print format keep one-digit precision is enough. Also please modify CSV_TOTAL_LINE_FMT
>>> make sure the csv also have same precision of Cycles/op.
>> We have checked the CSV formatting a find
>> 1. the precision for average cycle/op is 2 digits precision.
>> 2. already the CVS format has average cycles/op integrated.
>>
>> Hence no change is required.
> It's weird.
>
> We can see more clear when add together:
> #define CSV_TOTAL_LINE_FMT "Scenario %u Summary, , , , , ,%u,%.2lf,%u,%.3lf,%.3lf\n"
>          snprintf(output_str[MAX_WORKER_NB], MAX_OUTPUT_STR_LEN, CSV_TOTAL_LINE_FMT,
>                          cfg->scenario_id, nr_buf, memory * nb_workers,
>                          avg_cycles_total / nb_workers, bandwidth_total, mops_total);
> The bandwidth_total, mops_total both are float, which take last two "%.31f", then "%u" is for "avg_cycles_total / nb_workers"
>
> Also, avg_cycles_total / nb_workers both are u32, and its result is u32 if not with force convert.
>
> You could modify with "avg_cycles_total*1.0 / nb_workers", then compile will output warning:
>
> [2035/3727] Compiling C object app/dpdk-test-dma-perf.p/test-dma-perf_benchmark.c.o
> ../../dpdk/app/test-dma-perf/benchmark.c: In function ‘mem_copy_benchmark’:
> ../../dpdk/app/test-dma-perf/benchmark.c:26:28: warning: format ‘%u’ expects argument of type ‘unsigned int’, but argument 7 has type ‘double’ [-Wformat=]
>   #define CSV_TOTAL_LINE_FMT "Scenario %u Summary, , , , , ,%u,%.2lf,%u,%.3lf,%.3lf\n"

thank you Chengwen, let me share v2 for this patch with both print and 
csv in same format.


>
> Thanks
>                              ^
>>> Thanks
>>>
>>>>         snprintf(output_str[MAX_WORKER_NB], MAX_OUTPUT_STR_LEN, CSV_TOTAL_LINE_FMT,
>>>>                         cfg->scenario_id, nr_buf, memory * nb_workers,
>>>>                         avg_cycles_total / nb_workers, bandwidth_total, mops_total);
>>>>
>> .
  

Patch

diff --git a/app/test-dma-perf/benchmark.c b/app/test-dma-perf/benchmark.c
index 9b1f58c78c..8b6886af62 100644
--- a/app/test-dma-perf/benchmark.c
+++ b/app/test-dma-perf/benchmark.c
@@ -470,7 +470,8 @@  mem_copy_benchmark(struct test_configure *cfg, bool is_dma)
 		bandwidth_total += bandwidth;
 		avg_cycles_total += avg_cycles;
 	}
-	printf("\nTotal Bandwidth: %.3lf Gbps, Total MOps: %.3lf\n", bandwidth_total, mops_total);
+	printf("\nAverage Cycles/op: %.2lf, Total Bandwidth: %.3lf Gbps, Total MOps: %.3lf\n",
+			(float) avg_cycles_total / nb_workers, bandwidth_total, mops_total);
 	snprintf(output_str[MAX_WORKER_NB], MAX_OUTPUT_STR_LEN, CSV_TOTAL_LINE_FMT,
 			cfg->scenario_id, nr_buf, memory * nb_workers,
 			avg_cycles_total / nb_workers, bandwidth_total, mops_total);