Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

This is a lovely little library, but it is not "an implementation of IEEE 754-1985". The most important of the innumerable major differences:

- The representation is totally different (twos-complement signed significand and exponent instead of unsigned significand, biased exponent, and sign bit), one less bit of precision, no inf/nan encodings, and support for unnormalized encodings. - No rounding. Results are chopped. - Traps instead of defined results for all computations. - No guard bit[s] for subtraction.

If you're using "IEEE 754" to colloquially refer to "floating point", we can go back considerably earlier; at least to the Z3.



Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: