[dpdk-dev] [PATCH 0/5] remove usage of register keyword in C

Ferruh Yigit ferruh.yigit at intel.com
Tue Oct 9 11:19:13 CEST 2018


On 8/23/2018 2:07 PM, Ferruh Yigit wrote:
> On 8/1/2018 10:03 PM, Stephen Hemminger wrote:
>> On Wed, 1 Aug 2018 18:03:04 +0000
>> Yongseok Koh <yskoh at mellanox.com> wrote:
>>
>>>> On Jul 31, 2018, at 11:07 AM, Stephen Hemminger <stephen at networkplumber.org> wrote:
>>>>
>>>> On Tue, 31 Jul 2018 18:48:40 +0200
>>>> Adrien Mazarguil <adrien.mazarguil at 6wind.com> wrote:
>>>>   
>>>>> On Tue, Jul 31, 2018 at 09:30:54AM -0700, Stephen Hemminger wrote:  
>>>>>> Declaring variables as register in C is a leftover from an earlier
>>>>>> era (like cassette tape decks in cars).    
>>>>>
>>>>> I don't agree here. It's a hint for compilers and developers that the
>>>>> address of such variables won't be needed (and cannot be taken) to enable
>>>>> whatever optimizations are possible knowing this.
>>>>>
>>>>> Somewhat like inline functions, it's not a forced optimization, just a
>>>>> useful hint that shouldn't hurt if used wisely.
>>>>>
>>>>> Besides, cassette decks are not dead yet :)  
>>>>
>>>> If you look at the code, that is not how register is being used (ie. don't take
>>>> address of this). It seems like an attempt at optimization.  
>>>
>>> I know compilers are smart enough and the occurrences in mlx4/5 were made from
>>> my old fashioned habit. But, I don't see any urgency to push this patch in RC
>>> stage even though I'm 99% sure that it is harmless. And in general I don't even
>>> understand why we can't live with that if it isn't harmful (or a violation) but
>>> informative. I mean no badness but at least one goodness :-)
>>>
>>> Thanks,
>>> Yongseok
>>>
>>
>> Sure, this is intended for next release not rc stage.
>> Just trying to clean up code base where I see it.
> 
> I agree with Yongseok, at worst they show the intention of the developer, I
> don't see motivation to remove them unless they are doing something wrong, which
> seems not the reason of this patch.
> 
> And although I found some information that says "register" ignored completely
> for gcc, I can see it differs when optimization disabled.
> I am not saying practically it differs, since we enable optimization expect from
> debugging, most probably there is no practical difference between having the
> keyword or not, but what I am trying to say is it not completely ignored either.

I am for marking this set as rejected, any objection?


More information about the dev mailing list