[parisc-linux] Re: Next step in glibc test math failures analysis ; -)

[Carlos O'Donell]

> So my next question is: is some of you know a similar implementation of cacosf()?
> or what would be the results of cacosf() for various limit figure with nan, 0.0 and infinity?

Is the problem the setting of the V-bit by the operation?

The value of  "cacos (inf + NaN i)" is  "NaN + inf i", the problem is
that after the test the processor has turned on the V bit to indicate
that the operation is invalid.

In 10-9 of the PA 2.0 book there is a complete description of what
should and should
not trigger an "invalid" exception.

Do we have a case of:
1. Any arithmetic operation on a signaling NaN except for conversion
to integer formats.
2. Magnitude subtraction of intinities like (+inf + -inf)
3. Multiplication of 0 and inf
4. The division operations 0/0 and inf/inf
5. Square root if the operand is less than zero
6. Inproper comparison operations
7. Conversion to integer format as in table 10-6.

Cheers,
Carlos.