Andreas Schwab wrote:
> > Then there are 3 bugs in glibc-2.3.6 for ia64.
>
> Definitely not. This is not an NaN, according to the hardware.
This is a ridiculous argumentation: I showed you a program that contains
- hardware instructions for testing X == X and X >= 0 || X <= 0
(fcmp.eq and fcmp.ge, fcmp.le); both evaluate to false;
- a library call to the isnanl function (note the use of -fno-builtin!);
it evaluates to false.
IEEE 754 does not say that a NaN can be recognized by a
"fclass.m p6,p7=f8,0x1c0" instruction. But it says that a NaN can be
recognized by the fact that it does not compare equal to itself.
Bruno
