Numeric.ExpExtended

Extend floating point types with a larger exponent range.

Documentation

class RealFloat a => ExpExtendable a where

Extend floating point types with a larger exponent range.

Implementors need only implement:

• the ExpExtended data type, isomorphic to a strict pair (a, Int)
• its constructor unsafeExpExtended
• its destructor withExpExtended

Using a data family allows the UNPACK optimisation.

Minimal complete definition

unsafeExpExtended, withExpExtended

Associated Types

data ExpExtended a

Associated data.

Instances: Compensable, Enum, Eq, Floating, Fractional, Num, Ord, Precise, Read, Real, RealFloat, RealFrac, Show

Methods

withExpExtended :: ExpExtended a -> (a -> Int -> r) -> r

Deconstruct into basic value and exponent.

unsafeExpExtended :: a -> Int -> ExpExtended a

Construct from a basic value and an exponent, without checking the invariant. Use expExtended instead.

cache :: Cache a

Cache of magic values. Stored once per instance to avoid recomputation.

expExtended :: a -> Int -> ExpExtended a

Construct from a basic value and an exponent, ensuring that the result establishes the internal invariant:

m == significand m && ((m == 0 || isInfinite m || isNaN m) ==> e == 0)

Also handles overflow to infinity, and underflow to zero.

Instances

ExpExtendable Double

ExpExtendable Float

expExtended' :: ExpExtendable a => a -> ExpExtended a

Extend the exponent range while preserving the value.

expExtended' x == expExtended x 0

unExpExtendable :: ExpExtendable a => ExpExtended a -> Bool

Does the extended value fit in the base type without over/underflow?

unExpExtended :: ExpExtendable a => ExpExtended a -> a

Scale to the base type (possibly overflowing to infinity or underflowing to zero).

type EDouble = ExpExtended Double

type EFloat = ExpExtended Float