Rounding errors are a common problem in floating point arithmetic. For a pair of
floating point numbers `a`

, `b`

, an error-free transformation (EFT) is an
operation that maps `(a, b)`

into `(x, y)`

, where `a ○ b = x + y`

and ```
x = fl(a
○ b)
```

with `○`

being either addition, subtraction or multiplication. Standard
EFTs are the `TwoSum`

algorithm by Knuth (implemented as `err_add`

) and the
`TwoProd`

algorithm by Dekker (implemented as `err_mul`

).

This package implements several EFTs from the scientific literature.

This project is still very much a work in progress. It requires a recent build of Julia 0.4 to work.

01/11/2015

over 6 years ago

18 commits