2 Dec
2010
2 Dec
'10
9:44 p.m.
On 2 Dec 2010 at 21:15, Brent Hilpert wrote:
I was just thinking/conjecturing that if the
implementors were doing
inherent floating point conversions, they better have confirmed that
the integer divide result would be correct for all (legal)
combinations of operand values, with no (say) off-by-1 situations or
what-have-you.
I suspect that in the original example the system had a special
representation for NAN or infinity in the case of DBZ and the
generated code didn't test for it.
--Chuck