[Mono-devel-list] floating point compares

[Paolo Molaro]

Posted just to mono-devel-list, there is not point in cross-posting.

On 03/06/03 yoros at wanadoo.es wrote:
> using System;
> 
> public class EntryPoint {
> 
> 	public static void Main (string[] args) {
> 
> 		double a = Double.NaN;
> 		double b = 1.0;
> 
> 		if (a == a)
> 			Console.WriteLine("NAN == NAN");
> 		if (a < a)
> 			Console.WriteLine("NAN < NAN");
> 		if (a > a)
> 			Console.WriteLine("NAN > NAN");
> 		if (a <= a)
> 			Console.WriteLine("NAN <= NAN");
> 		if (a >= a)
> 			Console.WriteLine("NAN >= NAN");
[...]

There are at least two bugs, one in mcs and one in the runtime:

mcs generates the following code to implement bool val = a <= a:
        IL_0052: ldloc.0 
        IL_0053: ldloc.0 
        IL_0054: cgt 
        IL_0056: ldc.i4.0 
        IL_0057: ceq 
while cvs generates:
        IL_004d: ldstr "NAN <= NAN: {0}"
        IL_0052: ldloc.0 
        IL_0053: ldloc.0 
        IL_0054: cgt.un 
        IL_0056: ldc.i4.0 
        IL_0057: ceq 
and for bool val = a >= a:
        IL_0068: ldloc.0 
        IL_0069: ldloc.0 
        IL_006a: clt 
        IL_006c: ldc.i4.1 
        IL_006d: sub 
versus the csc output:
        IL_0068: ldloc.0 
        IL_0069: ldloc.0 
        IL_006a: clt.un 
        IL_006c: ldc.i4.0 
        IL_006d: ceq 

The output from the ms runtime when running the code compiled by mcs
is the same as with the mono runtime.

I have filed http://bugzilla.ximian.com/show_bug.cgi?id=39314
to track the bug.

lupus

-- 
-----------------------------------------------------------------
lupus at debian.org                                     debian/rules
lupus at ximian.com                             Monkeys do it better