www.digitalmars.com         C & C++   DMDScript  

digitalmars.D.bugs - optimizer bug

reply xs0 <xs0 xs0.com> writes:
Content-Type: text/plain; charset=ISO-8859-1; format=flowed
Content-Transfer-Encoding: 7bit

If the attached code is compiled without -o, it works fine; when 
optimized, however, it always produces -nan (regardless of the actual 
coordinates used). If I change all doubles to reals, it also works fine..

Sorry there's so much code, I couldn't make it shorter while maintaining 
the bug. The problem is in the line starting with "double tmp=...".


xs0
Aug 11 2005
parent xs0 <xs0 xs0.com> writes:
Well, I did manage to make it shorter :) I was using constants, and they 
obviously got folded during compilation.. The code below exhibits the 
same problem.. Again, changing doubles to reals makes it work..


----

import std.stdio;

import std.math;
import std.random;


void main()
{
	double x10=rand()*0.000001;
	double x11=rand()*0.000001;
	double x20=rand()*0.000001;
	double x21=rand()*0.000001;
	double y10=rand()*0.000001;
	double y11=rand()*0.000001;
	double y20=rand()*0.000001;
	double y21=rand()*0.000001;

	double tmp=(x10*x10 + x11*x11 + x20*x20 + x10*(x11 - 2.0*x20 - x21) + 
x20*x21 + x21*x21 - x11*(x20 + 2.0*x21) + y10*y10 + y10*y11 + y11*y11 - 
2.0*y10*y20 - y11*y20 + y20*y20 - y10*y21 - 2.0*y11*y21 + y20*y21 + 
y21*y21);

	writefln(tmp);
}


xs0
Aug 11 2005