Re: Why is malloc being defined as rpl_malloc ??

[Paul Eggert]

"Dr. David Kirkby" <address@hidden> writes:

> > I long ago lost interest in porting to any machines that don't conform
> > to the IEEE-754 floating point standard (ISO/IEC 559:1989).  
> 
> That does rather exclude some common machines, like the Alphas. 

But the Alpha uses IEEE-754 arithmetic.  It's only the older designs
(IBM and Unisys mainframes, VAX, Crays) that lack IEEE.

> Is there an autoconf macro that checks for IEEE-754 conformance? 

No.  C99 says that implementations predefine the symbol __STDC_IEC_559__
if they conform to IEC 559 (i.e. IEEE-754).  However, few implementations
do that yet, since few conform to C99 yet, so I would suggest code like
the following instead:

   #ifdef STDC_HEADERS
   # include <float.h>
   #endif

   #if (__STDC_IEC_559__
        || (FLT_RADIX == 2 && FLT_MANT_DIG == 24 \
            && FLT_MIN_EXP == -125 && FLT_MAX_EXP == 128))
   # define IEEE_FLOATING_POINT 1
   #else
   # define IEEE_FLOATING_POINT 0
   #endif

This works on all implementations that I know of, and code like this
has been in GNU Emacs for many years (I contributed it in 1996).

> http://www.netbsd.org/People/Pages/ross-essays.html

Some of that web page is quite inaccurate, at least by today's
standards.  For example, it says that IEEE-754 floating point
performance is slower than non-IEEE, but if you look at Spec CFP2000,
say, the fastest single-CPU score is the Bull NovaScale 4040, which
scores 1391 (it is a 1 GHz Itanium 2 running 64-bit GNU/Linux, by the
way).  This tops even the Alpha (HP AlphaSever GS1280, scoring 1124),
and is miles ahead of any non-IEEE-754 box that I know of.

Also, I'm generally not concerned with every last little detail of
IEEE-754.  After all, even the x86, which does conform to the
standard, conforms to it in a tricky way that makes accurate
floating-point programming difficult in common high-level languages
like C and Fortran.  What I am concerned with is whether the platform
conforms to the spirit of IEEE-754.  All modern platforms do so,
including the Alpha -- though some are more grudging than others.


> My attitude is that if I can produce code that produces
> 'sensible/useful' results on hardware not conforming to that standard,
> without sacrificing performance on IEEE-754 machines, I will do so. 

Hmm, even if the code is noticeably harder to maintain?  You'd do that
just for IBM mainframes and Crays?  I'm not sure I'd bother, myself.