Thu May 13 22:33:54 CEST 2010
Jason Garrett-Glaser <darkshikari at gmail.com> writes:
>> does C really mandate the implementation to use double precision?
>> i mean it can mandate the high level type be double but what is
>> preventing the compiler from using a very slightly lower precission
>> float implemenattion.
>> i ve heared intels c compiler will even use trigonometric identities
>> to optimize float code ...
> That's what -ffast-math is for.
Note that we cannot use -ffast-math in FFmpeg. It enables
-ffinite-math-only, which might make things fail if NaNs or
infinities are used, and we use both.
mans at mansr.com
More information about the ffmpeg-devel