Re:1/0 = 0
by js8 on Saturday August 11, 2018 @03:35AM
(#17738760)
As an aside, I have thought recently that integers should also have NaN value, just like floating point values have. It's useful in statistics to differentiate between not having a value and value of 0.
And I was thinking that perhaps 0x80000000 (in two's complement arithmetic) would be a good value for integer NaN. This value is already fixed point with respect to negation, so why not make it into an exception on its own right?
But it would mean to rethink all the hardware which I don't think is an option.
In any case, I think to set 1/0 to 0 is a bad idea, but I could imagine setting x/0 to 0x80000000 (interpreted as NaN) to make more sense.
And I was thinking that perhaps 0x80000000 (in two's complement arithmetic) would be a good value for integer NaN. This value is already fixed point with respect to negation, so why not make it into an exception on its own right?
But it would mean to rethink all the hardware which I don't think is an option.
In any case, I think to set 1/0 to 0 is a bad idea, but I could imagine setting x/0 to 0x80000000 (interpreted as NaN) to make more sense.