[vlc-devel] [PATCH 09/10] vlc_tick: add msftime_t to express 100ns time

[Rémi Denis-Courmont]

As I already pointed out, CLOCK_FREQ must remain a multiple of one million and a divisor of one billion. Too much problems with overflow and rounding otherwise.

Besides POSIX, ISO and Microsoft all agree there.

Le 22 juin 2018 14:28:08 GMT+01:00, Steve Lhomme <robux4 at ycbcr.xyz> a écrit :
>On 2018-06-22 3:17 PM, Romain Vimont wrote:
>> On Fri, Jun 22, 2018 at 02:53:26PM +0200, Steve Lhomme wrote:
>>> Such values currently have a better resolution than vlc_tick_t
>>> Also provide conversion macros/functions to/from vlc_tick_t
>>> ---
>>>   include/vlc_tick.h | 18 ++++++++++++++++++
>>>   1 file changed, 18 insertions(+)
>>>
>>> diff --git a/include/vlc_tick.h b/include/vlc_tick.h
>>> index 9e7f909e149..6953ba346d4 100644
>>> --- a/include/vlc_tick.h
>>> +++ b/include/vlc_tick.h
>>> @@ -138,6 +138,24 @@ static inline int64_t
>ns_from_vlc_tick(vlc_tick_t vtk)
>>>       return vtk * INT64_C(1000000000) / CLOCK_FREQ;
>>>   }
>>>   
>>> +/**
>>> + * Typically Microsoft time with 100ns resolutions
>>> + */
>>> +typedef int64_t msftime_t;
>>> +#define MSFTIME_FROM_VLC_TICK(vtk)  ((msftime_t)((vtk) *
>(INT64_C(10000000) / CLOCK_FREQ)))
>>> +#define VLC_TICK_FROM_MSFTIME(msft) ((vlc_tick_t)((msft) /
>(INT64_C(10000000) / CLOCK_FREQ)))
>>> +#define MSFTIME_FROM_SEC(sec)       ((msftime_t)((sec) *
>INT64_C(10000000)))  /* seconds in msftime_t */
>>> +
>>> +static inline vlc_tick_t vlc_tick_from_msftime(msftime_t t)
>>> +{
>>> +    return t * CLOCK_FREQ / INT64_C(10000000);
>> The order of operations is not the same as in the macro, so in theory
>> the results (in extreme cases) might be different.
>>
>> If 10000000 is a multiple of CLOCK_FREQ, the macro implementation is
>> always better (less overflows). In practice, this is probably not
>> important.
>
>It doesn't overflow because of the current value of CLOCK_FREQ. If 
>CLOCK_FREQ changes (to 100,000,000 for example) then the macro is doing
>
>division by 0. The function is correct except it might overflow.
>
>To avoid this I would like the add static asserts in the various 
>functions added by these patches. But that means moving the code. Or 
>putting the assert somewhere else (with relevant explanation).
>
>We could also do some #if depending on CLOCK_FREQ > divider or not and 
>do the operations the other way around in that case.
>
>>
>>> +}
>>> +
>>> +static inline msftime_t msftime_from_vlc_tick(vlc_tick_t vtk)
>>> +{
>>> +    return vtk * INT64_C(10000000) / CLOCK_FREQ;
>>> +}
>>> +
>>>  
>/*****************************************************************************
>>>    * MSTRTIME_MAX_SIZE: maximum possible size of mstrtime
>>>   
>*****************************************************************************
>>> -- 
>>> 2.17.0
>>>
>>> _______________________________________________
>>> vlc-devel mailing list
>>> To unsubscribe or modify your subscription options:
>>> https://mailman.videolan.org/listinfo/vlc-devel
>> _______________________________________________
>> vlc-devel mailing list
>> To unsubscribe or modify your subscription options:
>> https://mailman.videolan.org/listinfo/vlc-devel
>
>_______________________________________________
>vlc-devel mailing list
>To unsubscribe or modify your subscription options:
>https://mailman.videolan.org/listinfo/vlc-devel

-- 
Envoyé de mon appareil Android avec Courriel K-9 Mail. Veuillez excuser ma brièveté.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mailman.videolan.org/pipermail/vlc-devel/attachments/20180622/ec011198/attachment-0001.html>