[fortran, patch] Add Fortran 2003 IEEE intrinsic modules

[Steve Kargl]

On Fri, Nov 22, 2013 at 03:27:05PM +0100, FX wrote:
> > You can add glibc to the list to of libraries that may have issues
> > with its long double math functions.
> 
> Yeah, I?m sure I can find bugs in any math library :) 

It's not bug.  It is implementation details.  For example, e_sqrtl.c
contains

  long double
  __ieee754_sqrtl (long double x)
  {
    fputs ("__ieee754_sqrtl not implemented\n", stderr);
    __set_errno (ENOSYS);
    return 0.0;
  }
  strong_alias (__ieee754_sqrtl, __sqrtl_finite)

  stub_warning (sqrtl)

Yes, gcc probably generates a hardware fpu instruction from sqrtl
(when an fpu is available), but there are other functions that
have "not implemented" messages.

> Also, IIRC, you did some great job in improving FreeBSD?s math
> library some time ago, so that will be an interesting platform
> to test my patch on!

Yes, I've implemented some of the missing C99 math functions for
FreeBSD.  The status of FreeBSD's libm can be found  at
https://wiki.freebsd.org/Numerics.  The wiki lists some functions
that are implement but have yet to be committed.  These include
my sincos[fl], erfl, and erfcl functions.  Along the way, I found
erff and erfcf were somewhat broken and fixed these functions.  The
problems I found in erff and erfcf are still present in glibc.
Unfortunately, as with gfortran I have little time to work on FreeBSD's
libm, so it is getting done in spurts.

PS: FreeBSD implements both Intel 80-bit and IEEE 128-bit long double
formats and the implementations are done with round-to-nearest by
default (except sqrtl which should be correctly rounded in all rounding
mode).

PPS: Thanks for picking up the IEEE-754 support for gfortran.
-- 
Steve