[FFmpeg-devel] [RFC] remove lrintf fallback implementation
Sun Oct 21 23:19:13 CEST 2007
Rich Felker <dalias at aerifal.cx> writes:
> On Sun, Oct 21, 2007 at 12:51:19PM +0100, M?ns Rullg?rd wrote:
>> >> If I never see another mail complaining about missing llrint and
>> >> a broken patch which try to implement it I'm actually in favour
>> >> of this patch.
>> > I'm against optimized cpu-specific implementations to work around
>> > broken systems with missing llrint, but is there something wrong with
>> > just implementing it as (long long)rint(x) ?? llrint is a C99 feature
>> > and C99 math features are notoriously poorly supported...
>> rint() raises the inexact floating-point exception if the argument
>> isn't an integer. nearbyint() does not raise the exception, but is
>> also a C99 feature IIRC.
> And libavcodec is not using floating point exceptions so it's not
> relevant. Surely there's no way the state of exceptions could be
> expected to be maintained across calls to lavc functions. It's not
> even maintained by standard math library functions.
I was thinking of the extra time required for the useless exception
mans at mansr.com
More information about the ffmpeg-devel