Float to Int Conversions

Float to Int Conversions

[Stephen Pisuk]

Hello,

I have a question regarding the conversion of floats
to integers. The problem is that GCC (version 2.95
3.0.2)  produce different results than Microsoft's Visual Studio
6.0.  First of all, its a matter of who is right, and then
second can they be made consistent.


For example
double x = pow(2,32) + 10;
unsigned int    y = ((unsigned int) x);

Under MS VS y = 10
however under linux with GCC y=2^31 or 0x80000000

I assume the difference is in how the FPU control
word is set with the various error masks.

Is there any way so alter the functionality of GCC
to produce the first result?

Any help would be greatly appreciated. We write code
under both platforms, and its less of an issue of who
is right or wrong, but rather consistency.

Thanks
__________________________________________________
Stephen Pisuk
MIT Lincoln Laboratory, Room C-472
244 Wood St., Lexington, MA   02420-9108
voice: (781)-981-3682   fax: (781)-981-4583
e-mail:   smp@ll.mit.edu

Re: Float to Int Conversions

[John Love-Jensen]

Hi Stephen,

Have you set IEEE 754 compliance on both compilers?

GCC -ffloat-store

I don't know what MSVC uses.

--Eljay

Re: Float to Int Conversions

[Stephen Pisuk]

No I had not, but it didn't help

If the result of the conversion is an INT
the conversion saturates at 0x80000000
however if the result is an UNSIGNED INT
the conversion is correct;


> Hi Stephen,
>
> Have you set IEEE 754 compliance on both compilers?
>
> GCC -ffloat-store
>
> I don't know what MSVC uses.
>
> --Eljay
>
>

__________________________________________________
Stephen Pisuk
MIT Lincoln Laboratory, Room C-472
244 Wood St., Lexington, MA   02420-9108
voice: (781)-981-3682   fax: (781)-981-4583
e-mail:   smp@ll.mit.edu

building on MacOS 10.1.3?

[sbloch]

I'm not sure if this message got to the list when I tried it before.

I just upgraded my Mac to MacOS 10.1.3.  I downloaded the standard
developer package from Apple, including "make", "cc", etc. but I've
been unable to build GNU make, gcc, et al.

GNU make: I get compiler errors on the source file gettext.c because
LC_MESSAGES is undefined.  It's referred to in some places, protected by
an #ifdef LC_MESSAGES, but in other places there is no such protection.

GCC: I get compiler errors on the source file function.c because
DEFAULT_ABI is undefined.  I haven't poked around to see what's going
on here.

Any suggestions?  I have no idea what these symbols are supposed to
mean, so I'm hesitant to go around defining them myself.

-- 
                                                 Stephen Bloch
                                               sbloch@adelphi.edu
                                         http://www.adelphi.edu/sbloch/
                                        Math/CS Dept, Adelphi University

Re: Float to Int Conversions

[John Love-Jensen]

Hi Stephen,

According to C++ Programming Language by Stroustrup, section C.6.2.6 --
the behavior is UNDEFINED (implementation dependent) if the truncated
value cannot be represented in the destionation type.

Not what you wanted to hear.  If you want consistent behavior, you'll
have to write a helper subroutine that catches the boundary cases.

--Eljay