Skip to content

AdaptivePredicates.jl: Port of Shewchuk's robust predicates into Julia.

License

Notifications You must be signed in to change notification settings

JuliaGeometry/AdaptivePredicates.jl

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

27 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

AdaptivePredicates

A Julia port of "Routines for Arbitrary Precision Floating-point Arithmetic and Fast Robust Geometric Predicates" by Jonathan Richard Shewchuk. https://www.cs.cmu.edu/~quake/robust.html

The package provides four predicates. In the functions below, all the inputs should be NTuples with either Float64 or Float32 coordinates; complex inputs can be used for orient2 and incircle.

  • orient2(pa, pb, pc): Given three points pa, pb, and pc in two dimensions, returns a positive value if the points are in counter-clockwise order; a negative value if they occur in clockwise order; and zero if they are collinear. Equivalently, returns a positive value if pc is left of the oriented line from pa to pb; a negative value if pc is right of this line; and zero if they are collinear.
  • orient3(pa, pb, pc, pd): Given four points pa, pb, pc, and pd in three dimensions, define the oriented plane on which the triangle (pa, pb, pc) is positively oriented. Returns a positive value if pd is below this plane; a negative value if pd is above this plane; and zero if the points are coplanar.
  • incircle(pa, pb, pc, pd): Given four points pa, pb, pc, and pd in two dimensions, returns a positive value if pd is inside the circle through pa, pb, and pc; a negative value if pd is outside this circle; and zero if pd is on the circle.
  • insphere(pa, pb, pc, pd, pe): Given five points pa, pb, pc, pd, and pe in three dimensions, returns a positive value if pe inside of the sphere through pa, pb, pc, and pd; a negative value if pe is outside this sphere; and zero if pe is on the sphere.

We also define the functions orient2p, orient3p, incirclep, and inspherep which simply return the sign of the corresponding predicate. For example,

julia> using AdaptivePredicates

julia> pa, pb, pc = (0.2, 0.3), (0.1, -0.5), (0.7, 0.3);

julia> orient2(pa, pb, pc)
0.39999999999999997

julia> orient2p(pa, pb, pc)
1

julia> pa, pb, pc, pd, pe = (0.3f0, 0.3f0, 0.17f0), (-0.3f0, 1.71f0, 0.0f0), (0.0f0, 0.0f0, 5.0f0), (1.1f0, -0.53f0, 1.2f0), (0.5f0, 0.50f0, 0.5f0);

julia> insphere(pa, pb, pc, pd, pe)
-5.021922f0

julia> inspherep(pa, pb, pc, pd, pe)
-1

Installation

If you want to use the package, you can do

using Pkg
Pkg.add("AdaptivePredicates")
using AdaptivePredicates

Other Functions

All the functions from the predicates.c file from Shewchuk's original code have been included in this package. This includes,

  • All macros have been implemented as functions, e.g. Fast_Two_Sum and Four_Four_Sum.
  • All arithmetic functions have been implemented, e.g. grow_expansion and scale_expansion_zeroelim.
  • All the predicates have been implemented. In particular, not only have orient2, orient3, incircle, and insphere been implemented, but also the forms with the suffixes fast, exact, and slow (and adapt, but this is what orient2, orient3, incircle, and insphere use anyway).

Only the functions orient2, orient3, incircle, and insphere have been marked as public, as well as their p and fast counterparts.

Caveats

Shewchuk's original paper gives no analysis in the presence of underflow or overflow. The only mention of it is:

This article does not address issues of overflow and underflow, so I allow the exponent to be an integer in the range $[-\infty, \infty]$. (Fortunately, many applications have inputs whose exponents fall within a circumscribed range. The four predicates implemented for this article will not overflow nor underflow if their inputs have exponents in the range $[-142, 201]$ and IEEE 754 double precision arithmetic is used.)

  • Richard Shewchuk, J. Adaptive Precision Floating-Point Arithmetic and Fast Robust Geometric Predicates. Discrete Comput Geom 18(3), 305–363 (1997)

Note that this range comes from the insphere predicate. The number range is much wider for orient2, for example, since it requires far fewer additions, subtractions, and multiplications.

Thus, for some numbers, the values returned from these predicates may be invalid due to underflow or overflow. In ranges where this is a concern, you should use ExactPredicates.jl instead. If you need the values of the predicates and not just their signs, but are outside of the range valid for AdaptivePredicates.jl, you are unfortunately out of luck.

If you need more information about how these predicates work, you should refer to Shewchuk's paper.

License

The original code is in the public domain and this Julia port is under the MIT License.