[Bug target/14850] New: -ffloat-store and -mieee-fp aren't doing what I assume they should do

[Bug target/14850] New: -ffloat-store and -mieee-fp aren't doing what I assume they should do

[gonnetp at inf dot ethz dot ch]

I am trying to compute the IEEE-754 unit of least precision (ulp) for 64-bit
double precision numbers on an IA-32 machine with the following C-code:

    int main ( int argc , char *argv ) {

        double d, eps;

        printf("Unit of Least Precision:\n");
        d = 1.0; eps = 0.5;
        while (1.0 + eps/2 > 1.0)
            eps /= 2;
        printf("ulp = %20.16e\n\n",eps);

        }

Compiling with gcc-3.3.1 using

    gcc -O0 -ffloat-store -mieee-fp ieee.c

I get the output

    Unit of Least Precision:
    ulp = 1.0842021724855044e-19

meaning that my program is using the extra bits despite the "-ffloat-store" and
"-mieee-fp" options. Is this correct?

Using the Intel C Compiler v. 7.1 and compiling with

    icc -pc64 -O0 ieee.c

I get the correct result

    Unit of Least Precision:
    ulp = 2.2204460492503131e-16

Is there an option to gcc that forces the use of IEEE-754 64-bit operations as
in the Intel compiler? If not, is there any way to set this behaviour by hand?

Cheers and thanks
Pedro

-- 
           Summary: -ffloat-store and -mieee-fp aren't doing what I assume
                    they should do
           Product: gcc
           Version: 3.3.1
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: target
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: gonnetp at inf dot ethz dot ch
                CC: gcc-bugs at gcc dot gnu dot org
 GCC build triplet: i686-pc-linux-gnu
  GCC host triplet: i686-pc-linux-gnu
GCC target triplet: i686-pc-linux-gnu


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14850

[Bug target/14850] -ffloat-store and -mieee-fp aren't doing what I assume they should do

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-04-05 11:24 -------
If you have a pentium4 or higher (including amd64), you can use the following options: "-mfpmath=sse 
-msse -msse2".

The other thing you can do is manually change the rounding mode which I assume is what the option 
passed to ICC does.

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14850

[Bug target/14850] -ffloat-store and -mieee-fp aren't doing what I assume they should do

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-04-07 03:36 -------
Actually they are documented as what they do.  What you need to do is to change the 
FPU's internal precission to 64bit from the defualt of 80bits.

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14850

[Bug target/14850] -ffloat-store and -mieee-fp aren't doing what I assume they should do

[gonnetp at inf dot ethz dot ch]

------- Additional Comments From gonnetp at inf dot ethz dot ch  2004-04-07 09:11 -------
Subject: Re:  -ffloat-store and -mieee-fp aren't doing
	what I assume they should do


Hmm... Do you know how to do this?

Is there a way of submitting this as a feature request (i.e. for an
icc-like command line option)? The problem is that adding any special
library calls to set the FPU precision will mess-up portability :(

Cheers and many thanks,
Pedro


On Wed, 2004-04-07 at 04:36, pinskia at gcc dot gnu dot org wrote:
> ------- Additional Comments From pinskia at gcc dot gnu dot org  2004-04-07 03:36 -------
> Actually they are documented as what they do.  What you need to do is to change the 
> FPU's internal precission to 64bit from the defualt of 80bits.



-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14850