[dpdk-dev] [PATCH v5 1/3] examples/eventdev_pipeline: added sample app

[Hunt, David]

Hi Jerin,

On 6/7/2017 11:04 AM, Hunt, David wrote:
>
> On 6/7/2017 4:31 AM, Jerin Jacob wrote:
>> -----Original Message-----
>>>
--snip--
>> I checked v7 it looks to OK to merge. Can you fix following minor 
>> issue with
>> check patch and check-git-log.sh
>>
>> check-git-log.sh
>> -----------------
>> Wrong headline lowercase:
>>     doc: add sw eventdev pipeline to sample app ug
>
> Will Do. Change sw to SW
>
>> ### examples/eventdev_pipeline: added sample app
>
> Will Do. Add _sw_pmd
>

Both of these will be in next patch.

>> Note:
>> Change application to new name.
>>
>> checkpatch.sh
>> -----------------
>>
>> WARNING:EMBEDDED_FUNCTION_NAME: Prefer using '"%s...", __func__' to
>> using 'consumer', this function's name, in a string
>> #294: FILE: examples/eventdev_pipeline_sw_pmd/main.c:178:
>> +        printf("# consumer RX=%"PRIu64", time %"PRIu64 "ms, "
>>
>> WARNING:EMBEDDED_FUNCTION_NAME: Prefer using '"%s...", __func__' to
>> using 'worker', this function's name, in a string
>> #453: FILE: examples/eventdev_pipeline_sw_pmd/main.c:337:
>> +        printf("  worker %u thread done. RX=%zu TX=%zu\n",
>>
>> total: 0 errors, 2 warnings, 1078 lines checked
>
> These are false positives. The text in the messages are not meant to 
> be the function name.
> If anything, I would prefer to change the function names to have " 
> _thread"?
>

Having looked at this a bit more, and unable to reproduce with my 
original kernel version checkpatch, and the patchwork version does not 
show, and the 4.11.9 stable kernel version does not show, I suggest we 
mark these down as false positives, as the string is not intended to 
show the function name.


>> I will give pull request Thomas on Friday morning. I will include 
>> this change set
>> in the pull request.
>>
>> Regarding the performance drop, Can you add __rte_cache_aligned on those
>> variable which creates regression in moving to rte_malloc area. The
>> cache line could be shared? If not fixing then its fine we will look 
>> into that latter.
>
> I will investigate and post a new patch in a few hours.
>

Of the 4 variables I am attempting to move into fastpath structure, no 
matter whether I move them one at a time or all at once, with 
__rte_cache_align or not, I still see a significant performance 
degradation. I suggest looking into this later.

I will push a patch in the next couple of hours with the first two 
changes mentioned above.  OK with you?

Regards,
Dave.