[PATCH]: PR29335 use MPFR for builtins fma, fmin and fmax

[Brooks Moses]

Kaveh R. GHAZI wrote:
> Right.  I wish MIN/MAX_EXPR were suitable for fmin/fmax replacements.
> There are a lot of optimizations in the middle-end for these EXPRs that
> I'd rather not duplicate in the builtins.  Instead I'd like to be able to
> lower the builtins into these codes like we do for fabs->ABS_EXPR.  But
> that would require work to change all the backends to honor the new
> behaviors for -0.0 and/or NaN.
> 
> Actually, I'm not sure if fmin/fmax care about -0.0, but I think they do
> care about NaN vs a normal number.  Anyone got the standard handy?

Which standard?  If it's like the Fortran standard, the C standard 
probably doesn't say, which presumably leaves it up to the IEEE 
floating-point standard, if anything.

There was a fairly warm little debate in comp.lang.fortran (and 
crossposted elsewhere, IIRC) a couple of months ago about what the IEEE 
standard required in such cases, and whether it required anything at 
all.  What I recall of the arguments roughly amounted to (a) The IEEE 
standard can be read to say that max(NaN, x) should be x, regardless of 
argument order (though I don't recall how clear this reading was); and 
(b) this behavior is stupid and disconcerting for anyone who's using 
propagation of NaNs to indicate when something's gone bad, and if that's 
what the IEEE standard says then it's wrong.

If you want to come to a specific decision on the matter, the thread may 
be worth digging up and reading; it wasn't at all Fortran-specific.

- Brooks