posit: Posit Numbers

[ bsd3, library, math, numeric ] [ Propose Tags ] [ Report a vulnerability ]

The Posit Number format attempting to conform to the Posit Standard Versions 3.2 and 2022. Where Real numbers are approximated by `Maybe Rational` and sampled in a similar way to the projective real line.


[Skip to Readme]

Flags

Manual Flags

NameDescriptionDefault
do-no-storable-random

Build without Storable or Random/Uniform Class support

Disabled
do-liquid

Build with Liquid Haskell checking

Disabled
do-rewrite

Build with Rewrite Rules for Fused Operations.

Disabled

Use -f <flag> to enable a flag, or -f -<flag> to disable that flag. More info

Downloads

Maintainer's Corner

Package maintainers

For package maintainers and hackage trustees

Candidates

  • No Candidates
Versions [RSS] 3.2.0.0, 3.2.0.1, 3.2.0.2, 3.2.0.3, 3.2.0.4, 3.2.0.5, 2022.0.0.0, 2022.0.0.1, 2022.0.1.0, 2022.0.1.1, 2022.0.1.2, 2022.0.1.3, 2022.0.1.4, 2022.1.0.0, 2022.2.0.0
Change log ChangeLog.md
Dependencies base (>=4.7 && <5), data-dword, deepseq (>=1.1 && <2), random, scientific [details]
Tested with ghc ==8.10.7, ghc ==9.0.2, ghc ==9.2.8, ghc ==9.4.8, ghc ==9.6.3
License BSD-3-Clause
Copyright 2021-2023 Nathan Waivio
Author Nathan Waivio
Maintainer nathan.waivio@gmail.com
Category Numeric, Math
Home page https://github.com/waivio/posit#readme
Bug tracker https://github.com/waivio/posit/issues
Source repo head: git clone https://github.com/waivio/posit
Uploaded by nwaivio at 2023-12-08T14:37:30Z
Distributions
Downloads 650 total (59 in the last 30 days)
Rating (no votes yet) [estimated by Bayesian average]
Your Rating
  • λ
  • λ
  • λ
Status Docs available [build log]
Last success reported on 2023-12-08 [all 1 reports]

Readme for posit-2022.2.0.0

[back to package description]

posit 2022.2.0.0

The Posit Standard 2022, and Posit Standard 3.2, where Real numbers are approximated by Maybe Rational. The Posit Numbers are a drop in replacement for Float or Double mapped to a 2's complement integer type; smoothly and with tapering precision, in a similar way to the projective real line. The 'posit' library implements the following standard classes:

  • Show
  • Eq
  • Ord -- compare as an integer representation
  • Num -- Addition, subtraction, multiplication, and other operations
  • Enum -- Successor and Predecessor
  • Fractional -- division, divide by zero is Not a Real (NaR) number
  • Real
  • Bounded
  • FusedOps -- dot product and others
  • Convertable -- Conversions between different posit formats
  • AltShow
  • Read
  • Storable -- Formats for binary data, for computation and data interchange
  • Random
  • Uniform
  • RealFrac
  • RealFloat
  • Floating -- Mathematical functions such as logarithm, exponential, trigonometric, and hyperbolic functions. Warning! May induce trance.

The Posits are indexed by the type (es :: ES) where exponent size and word size are related. In posit-3.2 es is instantiated as Z, I, II, III, IV, V. In posit-2022 es is instantiated as Z_2022, I_2022, II_2022, III_2022, IV_2022, V_2022. The word size (in bits) of the value is = 8 * 2^es, that is 2^es bytes. The Types: 'Posit8', 'Posit16', 'Posit32', 'Posit64', 'Posit128', and 'Posit256' as well as, 'P8', 'P16', 'P32', 'P64', 'P128', and 'P256' are implemented and include a couple of auxiliary classes, like AltShow, AltFloating, and FusedOps. So, 3.2 scales by dynamic range, 2022 scales by precision.

class AltShow a where
  -- Display the Posit in its Binary Representation
  displayBinary :: a -> String
  -- Display the Posit in its Integral Representation
  displayIntegral :: a -> String
  -- Display the Posit as a Rational
  displayRational :: a -> String
  -- Display the Posit as a Decimal until the Repented occurs
  displayDecimal :: a -> String
class AltFloating p where
  machEps :: p  -- Machine Epsilon near 1.0
  approxEq :: p -> p -> Bool
  goldenRatio :: p
  hypot2 :: p -> p -> p
  hypot3 :: p -> p -> p -> p
  hypot4 :: p -> p -> p -> p -> p

The 'FusedOps' class has been updated with rewrite rules, enabled by the --flag posit:do-rewrite flag. When the flag is enabled, the ghc compiler will match the pattern and then automatically rewrite with the fused operation.

class Num a => FusedOps a where
  -- |Fused Multiply Add: (a * b) + c
  fma :: a -> a -> a -> a
  -- |Fused Add Multiply: (a + b) * c
  fam :: a -> a -> a -> a
  -- |Fused Multiply Multiply Subtract: (a * b) - (c * d)
  fmms :: a -> a -> a -> a -> a
  -- |Fused Sum of 3 values: a + b + c
  fsum3 :: a -> a -> a -> a
  -- |Fused Sum of 4 values: a + b + c + d
  fsum4 :: a -> a -> a -> a -> a
  -- |Fused Sum of a List of Posits
  fsumL :: Foldable t => t a -> a
  -- |Fused Dot Product of 3 element vector: (a1 * b1) + (a2 * b2) + (a3 * b3)
  fdot3 :: a -> a -> a -> a -> a -> a -> a
  -- |Fused Dot Product of 4 element veector: (a0 * b0) + (a1 * b1) + (a2 * b2) + (a3 * b3)
  fdot4 :: a -> a -> a -> a -> a -> a -> a -> a -> a
  -- |Fused Dot Product of Two Lists
  fdotL :: Foldable t => t a -> t a -> a
  -- |Fused Subtract Multiply: a - (b * c)
  fsm :: a -> a -> a -> a

The Posit type is 'Convertible' between other Posit lengths.

class Convertible a b where
  convert :: a -> b

The Posit Library is built on top of two of the most excellent libraries: data-dword, and scientific. The 'data-dword' library provides the underlining machine word representation, it can provide 2^es word size, 2's complement fixed length integers. The 'scientific' library provides 'read' and 'show' instances.

The Elementary Functions for this posit library are implemented natively using the next larger posit number. P16 and Posit16 numbers are tested exhaustively against Double Precision numbers as the orical. The results are below:

Number of Accurate Bits exp: numberOfAccurateBits exp

Number of Accurate Bits log: numberOfAccuarateBits log

Number of Accurate Bits sqrt: numberOfAccuarateBits sqrt

Number of Accurate Bits sin: numberOfAccuarateBits sin

Number of Accurate Bits cos: numberOfAccuarateBits cos

Number of Accurate Bits asin: numberOfAccuarateBits asin

Number of Accurate Bits acos: numberOfAccuarateBits acos

Number of Accurate Bits atan: numberOfAccuarateBits atan

Number of Accurate Bits sinh: numberOfAccuarateBits sinh

Number of Accurate Bits cosh: numberOfAccuarateBits cosh

Number of Accurate Bits asinh: numberOfAccuarateBits asinh

Number of Accurate Bits acosh: numberOfAccuarateBits acosh

Number of Accurate Bits atanh: numberOfAccuarateBits atanh