support more IEEE-like semantics; better algorithms
authorClaude Heiland-Allen <claudiusmaximus@goto10.org>
Mon, 16 Apr 2012 15:20:05 +0000 (16:20 +0100)
committerClaude Heiland-Allen <claudiusmaximus@goto10.org>
Mon, 16 Apr 2012 15:20:05 +0000 (16:20 +0100)
commit753212c417f5a22367a9d8e1f9536e67ce0d3dda
treefd22d252454f77e997b6f82411779ac231fabb76
parent448b64eb2266c3f597e4b44affc4b67e8dacc83c
support more IEEE-like semantics; better algorithms

  * support +inf, -inf, nan; handle zero explictly

  * support special values in recodeFloat

  * instance EffectivePrecision

  * numeric classes extended to special values

  * recip algorithm changed to Newton-Raphson division

  * encodeFloat efficiency improved with ilog2

  * implement exponent, significand, scaleFloat

  * pi efficiency improved with scaleFloat

  * pi convergence criterion set to 2 lsb accuracy

  * sqrt efficiency improved with scaleFloat, rescaling via [1,4), Double initial estimate

  * sqrt convergence criterion set to 2 lsb accuracy

  * log efficiency improved with scaleFloat

  * log convergence criterion set to 2 lsb accuracy

  * checking for overflow/underflow doesn't error/trace
Numeric/VariablePrecision/Float.hs