Re: Avoiding floating point problems on Pentium Xeon platform w/GCC 3.2

[Kelley Cook <kcook34 at ford dot com>]

> Below is one of the situations I ran into.
> 	int main()
> 	{
> 	        double d1, d2;
> 	        int i1, i2;
>
> 	        d1 = 0.04;
> 	        i1 = 100;
> 	        i2 = 100;
>
> 	        d2 = i1 / d1;
> 	        i1 = d2; 
> 	        i2 = i2 / d1;
>
> 	        printf("%d %d\n", i1, i2);
> 	        return 0;
> 	}
>
> On Alpha, the above application results in "2500 2500", but on Intel Xeon,
> it results in "2500 2499".

First of all the obvious answer that your code is highly suspect for an engineering app, since you don't want to ever convert from double -> integer without using either lround() or lrint().

On the other hand, this "problem" goes away if you use "-march=pentium4 -mfpmath=sse" or (due to happenstance in this particular instance) if you enable optimization with "-O2".

Kelley Cook