DeFloat

Floating Point Decomposition

A small implementation of a portable floating point format (i.e. with no loss of precision across floating point implementations).

Examples

Object-Oriented style:

if __name__ == '__main__':
        from defloat import DeFloat

        a = 5.32515
        b = b_z, b_e = DeFloat.decomp(a)
        c = float(b)  # 1 way to do it
        d = b.recomp()  # the other way to do it

        print(f"decomposition:   fraction={b_z}, exponent={b_e}")
        print(f"recomposition 1: {c}")
        print(f"recomposition 2: {d}")
        print(f"check:           {a}")

        assert c == d == a

        """
        expected output:
                decomposition:   fraction=5995585888922999, exponent=3
                recomposition 1: 5.32515
                recomposition 2: 5.32515
                check:           5.32515
        """

Procedural style:

if __name__ == '__main__':
        import defloat

        a = 5.32515
        b_z, b_e = defloat.decomp(a)
        c = defloat.recomp(b_z, b_e)

        print(f"decomposition: fraction={b_z}, exponent={b_e}")
        print(f"recomposition: {c}")
        print(f"check:         {a}")

        assert c == a

        """
        expected output:
                decomposition: fraction=5995585888922999, exponent=3
                recomposition: 5.32515
                check:         5.32515
        """

Notes

Recomposition on hardware with lower precision may/will result in data/precision loss. Precision is only guaranteed within the decomposed format, and recomposition to the same implementation as decomposed from.

License

DeFloat is licensed under Mozilla Public License.