[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[Gcl-devel] Re: new and improved results.

From: Camm Maguire
Subject: [Gcl-devel] Re: new and improved results.
Date: 05 Jun 2002 17:25:42 -0400


address@hidden writes:

> the awk fix for libbfd and libiberty is ok.


> the test for awk isn't, you need to test for nawk first

why?  If awk is not available, should then test for nawk.

> if you are on solaris, maybe a test for gawk before either
> of these is a good idea.

???  incompatibilities?

> second, isnormal is not defined on solaris, following

isnormal is defined on posix systems under an extension, which is
turned on with the macro _GNU_SOURCE.  Obviously won't be there for
solaris, but maybe something analogous is?  This is the only sensible
way to distinguish between the least-normalized and least-unnormalized
float constants.  Although we might get by with !isnormal ==
fabs(f)<FLT_MIN.  Please dig for the equivalent test on your box, and
we can implement.

For  the record, normailized floats have a mantissa and exponent, and
typically imply a hidden bit.  when the exponent is completely empty,
the mantissa alone implies an effective exponent which extends the
float range down to the unnormalized mon, losing precision gradually
along the way.

Take care,

> is a excerpt from the man page:
> man isnan
> C Library Functions                                     isnan(3C)
>      isnan, isnand, isnanf, finite, fpclass, unordered  -  deter-
>      mine type of floating-point number
>      #include <ieeefp.h>
>      int isnand(double dsrc);
>      int isnanf(float fsrc);
>      int finite(double dsrc);
>      fpclass_t fpclass(double dsrc);
>      int unordered(double dsrc1, double dsrc2);
>      #include <math.h>
>      int isnan(double dsrc);
>      The  functionalty  of  isnan()  is  identical  to  that   of
>      isnand().
>      isnanf()  is  implemented  as  a  macro  included   in   the
>      <ieeefp.h> header.
>      fpclass() returns the class the dsrc  belongs  to.   The  10
>      possible classes are as follows: 
>           FP_SNAN             signaling NaN
>           FP_QNAN             quiet NaN
>           FP_NINF             negative infinity
>           FP_PINF             positive infinity
>           FP_NDENORM          negative denormalized non-zero
>           FP_PDENORM          positive denormalized non-zero
>           FP_NZERO            negative zero
>           FP_PZERO            positive zero
>           FP_NNORM            negative normalized non-zero
>           FP_PNORM            positive normalized non-zero
>      None of these routines generate any exception, even for sig-
>      naling NaNs.
> Is isnormal defined on linux, or is it a posix standard?
> Dan

Camm Maguire                                            address@hidden
"The earth is but one country, and mankind its citizens."  --  Baha'u'llah

reply via email to

[Prev in Thread] Current Thread [Next in Thread]