4 ms·
What's wrong with using double floats as the only numerical type?
- acqq 16y agoThe author writes: "This is not true in floating point — also (a+b)+c may not be a+(b+c)." But in the case he mentions, for integral numbers up to 2^53 double floating point would also preserve the above identity." The doubles are not bad as long as you understand them.