News for hackers. Stuff that matters.
Re:A Radical Approach to Computation with Real Numbers [pdf] by dnautics on Saturday April 02, 2016 @05:08PM (#11413446)
I'm currently working with John on producing a Julia-based prototype of the original unums (work has stalled until I am getting paid again). A lot of the decision choices in this are based on annoying things that came up during the development. For example: subnormal numbers are a horrifying pain in the butt (but there are also really good reasons for having them)

I think there might be something to this "new unum" especially if there can be a shortcut that trims the size of the lookup tables when going to higher precision.

PS if anyone has some interesting math-ey contract work let me know.

Re:A Radical Approach to Computation with Real Numbers [pdf] by DavidSJ on Saturday April 02, 2016 @10:07PM (#11414592)
Do you know if there's been any thought on extending this to arbitrary algebras? Seems like you've already got the mechanism to define a partition of the space, and the lookup table describing the behavior of an operation with respect to the subsets of that partition, so this might have utility beyond arithmetic.
Re:A Radical Approach to Computation with Real Numbers [pdf] by jacobolus on Sunday April 03, 2016 @12:51AM (#11414975)
Huh. I wonder how fast the lookup tables for various operations could be if they were built using some kind of programmable logic. Then you could theoretically just interpret the bit patterns to be whatever kind of number-like object you want, and program the lookup tables accordingly.
Re:A Radical Approach to Computation with Real Numbers [pdf] by DavidSJ on Sunday April 03, 2016 @01:43AM (#11415082)
Right, now that I think about it, it's only coincidental that the objects in question represent the subsets in a partition of the reals.

It seems what we have here is a finite set A (in arithmetic, A would be a finite collection of sets which constitute a partition of the reals) and two operations: f, g: P(A) × P(A) -> P(A), taking a pair of subsets of A to another subset of A, with the property that f(W ∪ X, Y) = f(W, Y) ∪ f(X, Y), and so on. This can all be represented as a bitmask table of size |A|^3, or half if commutativity is enforced.

We also have the "inverse" which takes X and Z, and returns the set {a ∈ A : f(X, {a}) ∩ Z ≠ Ø}.

If I'm not mistaken, the mechanisms for doing all of this in hardware would come for free, and it represents quite a variety of problem domains beyond arithmetic. Perhaps some arithmetic-specific optimization would be ruled out (e.g. the bit twiddling to get additive or multiplicative inverses).

Re:A Radical Approach to Computation with Real Numbers [pdf] by trsohmers on Sunday April 03, 2016 @04:25AM (#11415397)
So, instead of ROM, it would be SRAM... no "reprogrammable logic" required. The difference between ROM bit cell and a RAM bit cell is around ~3x larger, and uses a lot more power wise for read and writes.
Re:A Radical Approach to Computation with Real Numbers [pdf] by dnautics on Sunday April 03, 2016 @01:50AM (#11415092)
Agreed. Interesting algebras like z/n[i] or z/n[√5] exist, but I'm not sure what the utility outside of theoretical algebra is.
Re:A Radical Approach to Computation with Real Numbers [pdf] by jacobolus on Sunday April 03, 2016 @02:18AM (#11415150)
The algebraic integers in Q(√5) — namely, numbers of the form aτ + b, where a and be are integers and the golden ratio τ = (1+√5)/2 — are quite interesting. If you make vectors of them, you can make a dodecahedron, iconsahedron, 120 cell, 600 cell, etc.

This is great practically for modeling quasicrystals, as well as things like buckyballs, certain types of viruses, architectural structures, ...

If you want to play with the symmetries, I recommend buying some Zometool construction toys, https://www.zometool.com

Re:A Radical Approach to Computation with Real Numbers [pdf] by trsohmers on Saturday April 02, 2016 @08:06PM (#11414200)
Here's a link to the prototype Julia implementation of the original (Unum's 1.0)... https://github.com/REX-Computing/unumjl
Re:A Radical Approach to Computation with Real Numbers [pdf] by carterschonwald on Saturday April 02, 2016 @11:02PM (#11414723)
is this slide deck the only specification of the unum 2.0 ideas?