[FFmpeg-cvslog] r21735 - trunk/libavutil/internal.h

[Michael Kostylev]

On Sat Feb 20 12:24:22 2010
M?ns Rullg?rd wrote:

>>>>>>>>> Define missing llrint() as macro instead of inline function
>>>>>>>>> 
>>>>>>>>> This fixes building on some broken systems.
>>>>>>>>> 
>>>>>>>>> Modified:
>>>>>>>>>    trunk/libavutil/internal.h
>>>>>>>>> 
>>>>>>>>> Modified: trunk/libavutil/internal.h
>>>>>>>>> ==============================================================================
>>>>>>>>> --- trunk/libavutil/internal.h	Tue Feb  9 23:55:13 2010	(r21734)
>>>>>>>>> +++ trunk/libavutil/internal.h	Tue Feb  9 23:55:16 2010	(r21735)
>>>>>>>>> @@ -225,10 +225,8 @@ static inline av_const unsigned int ff_s
>>>>>>>>>  #endif /* HAVE_EXP2F */
>>>>>>>>>  
>>>>>>>>>  #if !HAVE_LLRINT
>>>>>>>>> -static av_always_inline av_const long long llrint(double x)
>>>>>>>>> -{
>>>>>>>>> -    return rint(x);
>>>>>>>>> -}
>>>>>>>>> +#undef llrint
>>>>>>>>> +#define llrint(x) rint(x)
>>>>>>>>
>>>>>>>> The dos machine is unhappy with this macro.
>>>>>>> 
>>>>>>> Please elaborate.
>>>>>>
>>>>>> A bunch of tests from the regression suite and the idroq-video-encode
>>>>>> test no longer pass since r21735 (reverting it or applying a quick
>>>>>> hack attached in the previous message make them pass again).
>>>>> 
>>>>> Weird.  Which call to llrint() is causing the trouble?
>>>>
>>>> libavcodec/opt.c, line 67:
>>>>     case FF_OPT_TYPE_INT:   *(int       *)dst= llrint(num/den)*intnum; break;
>>> 
>>> Applying your cast to that expression should have no effect at all
>>> since the usual arithmetic conversions already specify that action.
>>
>> Under dos the argument of rint might be too big, e.g. 0xa0080004.0/1 while
>> normally it should be equal to 0x80004.0/1.
> 
> How the hell does casting the *result* fix a problem with the *argument*?

Nohow - it was shown where casting has some effect. 
Thanks for r21922.

Michael