1. 7
hpcwire.com
1. 7

Isn’t this just universal numbers? https://en.m.wikipedia.org/wiki/Unum_(number_format)

1. 1

I would love to know the validity of this claim. It seems fishy that a patent was filed but no white paper was submitted to journal for peer review (that I can find). If anyone with more expertise can provide their take on the matter, I would greatly enjoy it!

1. 19

The inventor has a website called boundedfloatingpoint.com. There he describes it in a bit more detail than the article, but not much.

Note carefully how he describes it:

This invention provides a device that performs floating point operations while calculating and retaining a bound on floating point error.

And “[t]his invention provides error notification by comparing the lost bits”.

It’s a solution to the problem of “unreported errors”. His solution provides extra fields in the floating point representation to carry information about ’lost bits” and allows the operator to specify how many significant digits must be retained before an error is flagged.

This is an advantage over the current technology that does not permit any control on the allowable error. This invention, not only permits the detection of loss of significant bits, but also allows the number of required retained significant digits to be specified.

At a cursory glance one might be inclined to think he’s solved the problem of floating point, but the reality is he’s developed a standard for communicating error in floating-point operations that can be implemented in hardware.

Not to detract from his solution, but it doesn’t seem like he’s invented anything that will surprise hardware designers.

1. 7

Thank you for that analysis. This is a real problem with floating point numbers, but hardly the only one.

People who haven’t seen it might be interested in this post from last year about a new number representation called “posits”, which addresses some completely orthogonal issues with fixed-size number representations. :)

1. 1

Nice! Thanks for the link.

2. 1

It’s a solution to the problem of “unreported errors”. His solution provides extra fields in the floating point representation to carry information about ’lost bits” and allows the operator to specify how many significant digits must be retained before an error is flagged.

``````SIGNAL ON LOSTDIGITS;
NUMERIC DIGITS 10;
NUMERIC FUZZ 2;
``````

We just need to do all our math in REXX.