This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]

Re: Fourth Draft "Unsafe fp optimizations" project description.


And Linus Torvalds writes:
 - 
 - Assuming they don't have a fdiv-like bug ;)

They don't.  It'd have to be an entirely different type of bug.  ;)

 - So I seriously doubt that you'd get better accuracy in the library
 - routines even if you tried _really_ hard. So I still maintain that the
 - only thing that a x86-based library routine can do is

I believe the current most-accurate-and-still-fast amount is around
0.57 ulp in round-to-nearest, but I don't have a citation handy.  
It'd be either in Markstein's "IA-64 and Elementary Functions" or 
Muller's "Elementary Functions".

 - I think that ridiculous arguments can get ridiculous results. Garbage in,
 - garbage out. But hey, maybe somebody has an algorithm that cares.

See K.C. Ng's paper on argument reduction under 
http://www.validlab.com .  Sun does have this in some libraries,
and some people do care.  Large arguments can appear in spectral
PDE solvers, although it's fair to question their origin.  The
argument reduction cost is huge, but apparently certain radar
scattering simulation folks are willing to pay for it...

Jason


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]