were they a stroke of genius or a mistake? and how come no-one ever tried implementing rational numbers in hardware?
Like a long double?
>>107709755Pure genius.
>>107709755rationals are already really fast in software. if you constraint the denominator, all operations become int operations which hardware can already do. This is typically what you do to represent money, e.g : int / 100.ieee754 is really good, but it has some weird features, like subnormals. it's also extremely non trivial to implement transcendental functions like sin, tan, log, atan etc.. because you either have to use higher precision or rely on weird tricks like IBM's math library does.
>>107710266forgot linkshttps://csclub.uwaterloo.ca/~pbarfuss/dekker1971.pdfhttps://www.researchgate.net/publication/220492903_An_accurate_elementary_mathematical_library_for_the_IEEE_floating_point_standard
>>107710266>This is typically what you do to represent money, e.g : int / 100.wrong
LLM clanker brainrot is only possible because GPUs that support this insane datatype can crunch billions of numbers very fast, so I think it is overall a bad thing, because in all mission critical software, people use fixed point arithmetic without a FPU at all.
>>107709755>no-one ever tried implementing rational numbers in hardwareThat's what floating point is, technically, due simply to the fact of limited precision.
>>107709755they let me play quake
>>107709755how else would you implement rational numbers?
>>107710966>how else would you implement rational numbers?I'm sure there's some rational solution. You know, like using a numerator and denominator or something.
>float accountBalance;
