Floating point trouble with x86's extended precision

[Bradley Lucier]

On Thursday, August 21, 2003, at 03:44  PM, Volker Reichelt wrote:

>> The problem isn't that the value is rounded.  The problem is that the
>> value is rounded at unpredictable locations,
>
> Agreed. But that's not a bug IMHO.

OK, this is the crux of the issue.  GCC currently does not, under any 
current combination of input source code, flags, anything, allow the 
programmer to have a predictable model of a program's floating-point 
behavior on x86.  You agree with this statement.

You don't think it's a bug.

I, and other people, think that the inability to get predictable 
behavior from GCC (even from an option that you might not like to use 
because of speed or other issues) is a bug.

And this is just about the end of the story.  I, and at least some 
other people, think that GCC should include at least one model of 
floating-point arithmetic for the x86 that allows one to deduce the 
types and precisions of all variables, operations, etc., by examining 
the source code.  It does not (and it probably will not for the 
foreseeable future).  Telling people that not having a predictable 
model of what their program will do is a feature, not a bug, doesn't 
cut it.

Brad