3 ms·
Floating Point Numbers
- kqr2 12y agoObligatory link to What Every Computer Scientist Should Know About Floating-Point Arithmetic http://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.html http://docs.oracle.com/cd/E19957-01/806-3568/ncg_goldberg.ht...
- acqq 12y agoIt seems there's an error in the text: "must be representable in binary using at most 52 bits for double precision and 23 bits for single recision. In other words, for double, 2^52f is an integer in the interval 0≤2^52f<2^52" 52 bits are used in the double but the highest 53rd bit is implicit for all IEEE754 doubles which aren't denormals. And denormals use less bits as they approach 0. The text misses both important aspects of the standard. And what's with that "I've always known him by his nickname "Velvel"" and then using the nickname from that point on? Sounds like Monty Python sketch: http://m.youtube.com/watch?v=9svCBKWpCRQ http://m.youtube.com/watch?v=9svCBKWpCRQ