From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <gcc-help-return-23547-listarch-gcc-help=gcc.gnu.org@gcc.gnu.org>
Received: (qmail 14766 invoked by alias); 7 May 2006 17:42:59 -0000
Received: (qmail 14750 invoked by uid 22791); 7 May 2006 17:42:56 -0000
X-Spam-Check-By: sourceware.org
Received: from wr-out-0506.google.com (HELO wr-out-0506.google.com) (64.233.184.224)     by sourceware.org (qpsmtpd/0.31) with ESMTP; Sun, 07 May 2006 17:41:45 +0000
Received: by wr-out-0506.google.com with SMTP id 36so845065wra         for <gcc-help@gcc.gnu.org>; Sun, 07 May 2006 10:41:44 -0700 (PDT)
Received: by 10.54.118.4 with SMTP id q4mr1381790wrc;         Sun, 07 May 2006 10:40:35 -0700 (PDT)
Received: from ?192.168.2.51? ( [141.157.250.229])         by mx.gmail.com with ESMTP id 67sm1077453wra.2006.05.07.10.41.42;         Sun, 07 May 2006 10:41:42 -0700 (PDT)
Message-ID: <445E3155.20209@gmail.com>
Date: Sun, 07 May 2006 17:42:00 -0000
From: Jim Cromie <jim.cromie@gmail.com>
User-Agent: Thunderbird 1.5.0.2 (X11/20060420)
MIME-Version: 1.0
To: "Nelson H. F. Beebe" <beebe@math.utah.edu>
CC:  gcc-help@gcc.gnu.org
Subject: Re: compile-time conversion of floating-point expressions to long  longs
References: <CMM.0.95.0.1146697025.beebe@psi.math.utah.edu>
In-Reply-To: <CMM.0.95.0.1146697025.beebe@psi.math.utah.edu>
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit
Mailing-List: contact gcc-help-help@gcc.gnu.org; run by ezmlm
Precedence: bulk
List-Archive: <http://gcc.gnu.org/ml/gcc-help/>
List-Post: <mailto:gcc-help@gcc.gnu.org>
List-Help: <mailto:gcc-help-help@gcc.gnu.org>
Sender: gcc-help-owner@gcc.gnu.org
X-SW-Source: 2006-05/txt/msg00077.txt.bz2

Nelson H. F. Beebe wrote:
> Jim Cromie <jim.cromie@gmail.com> asks on Wed, 03 May 2006 12:17:38 -0400:
>
>   
>>> ...
>>> Are there any macros that can tear into a floating point number and
>>> pull out the exponent and mantissa ?
>>> ...
>>>       
>
> Yes, the C89 and C99 ISO C Standards include ldexp(x,n), which forms x
> * 2**n, and "f = frexp(x,&n);", which returns the fraction as a value
> in [1/2,1) (or 0 if x is zero) and the exponent of 2 in n.  Both are
> exact, and can be implemented reasonably efficiently.
>   
thank you for the thorough answer.

yes, frexp() is almost what I want.
heres basically what I tried to do with it:

#include <stdio.h>
#include <math.h>
#include <float.h>

/* evaluate these constants at compile-time ?!? */
float f = 3.14159 * 2 * 100;
int myexp;
double mant = frexp(f,&myexp);

/* this confirms gcc's unwillingness to run libmath routines at 
compile-time */
float ff = ldexp(2,8);

int main(int c, char** v)
{
    printf("float %g mantissa %f exp %d \n", f, mant, myexp);
}


as Id feared, it wouldnt compile;
$ make fp-comp
cc -g    fp-comp.c   -o fp-comp
fp-comp.c:37: error: initializer element is not constant
make: *** [fp-comp] Error 1

(the code above is slightly edited for clarity - err line numbers dont 
match, but result does ;-(

I also tried these additions to compile cmd:
   -fkeep-static-consts  -fmerge-constants  -fmerge-all-constants
    -fsingle-precision-constant

So it seems that the compiler just wont call the function for me at 
compile-time,
and fold the constant into the code for me.  Id love to hear differently.

tia
jimc