4 ms·
What's really needed is an error term that can be computed in parallel with your normal floating point operation. Floating point is impossible to write correct
by slededit 10y ago
What's really needed is an error term that can be computed in parallel with your normal floating point operation. Floating point is impossible to write correctly in practice if you don't consider the accumulated error.
- erichocean 10y agoThere are interval arithmetic libraries for C++ that do that, and are fast in the common case where the error fits within floating point norms.