Data.Approximate.Numerics

These functions provide wildly inaccurate but very fast approximations to common transcendental functions.

The algorithms here are based on Martin Ankerl's optimized pow, http://martin.ankerl.com/2007/10/04/optimized-pow-approximation-for-java-and-c-c/ which is in turn based on http://nic.schraudolph.org/pubs/Schraudolph99.pdf

Documentation

class Floating a => Fast a where

Minimal complete definition

flog, flog_lb, flog_ub, fexp, fexp_lb, fexp_ub, fpow, fpow_lb, fpow_ub

Methods

flog :: a -> a

Calculate an approximate log.

flog_lb :: a -> a

flog_ub :: a -> a

fexp :: a -> a

Calculate an approximate exp.

fexp_lb :: a -> a

fexp_ub :: a -> a

fpow :: a -> a -> a

Calculate an approximate pow.

fpow_lb :: a -> a -> a

fpow_ub :: a -> a -> a

Instances

Fast Double

Methods

flog :: Double -> Double

flog_lb :: Double -> Double

flog_ub :: Double -> Double

fexp :: Double -> Double

fexp_lb :: Double -> Double

fexp_ub :: Double -> Double

fpow :: Double -> Double -> Double

fpow_lb :: Double -> Double -> Double

fpow_ub :: Double -> Double -> Double

Fast Float

Methods

flog :: Float -> Float

flog_lb :: Float -> Float

flog_ub :: Float -> Float

fexp :: Float -> Float

fexp_lb :: Float -> Float

fexp_ub :: Float -> Float

fpow :: Float -> Float -> Float

fpow_lb :: Float -> Float -> Float

fpow_ub :: Float -> Float -> Float

blog :: Floating a => a -> a

Borchardt’s Algorithm from “Dead Reckoning: Calculating without instruments”.

This is a remarkably bad approximate logarithm.

flog had better outperform it! It is provided merely for comparison.