| Thread Links | Date Links | ||||
|---|---|---|---|---|---|
| Thread Prev | Thread Next | Thread Index | Date Prev | Date Next | Date Index |
On 09.09.2015 15:49, Walter Mascarenhas wrote:
Dear Colleagues, Regarding the uncertain form for interval literals in item 9.7.4 of the (full) standard, 1) What is the meaning of the strings 0.00?1, 0?1, 0?1u, 0?1d
0.00?1 = [-0.01, +0.01] 0?1 = [-1, +1] 0?1u = [0, 1] 0?1d = [-1, 0]
or any string in which the center of the corresponding interval is zero? I would assign the interval [0,0] to all of them, but I believe the standard could be more clear about this case. Anyway, is [0,0] the correct interval in these cases?
No, the number after the question mark counts the uncertainty in ULPs and ULPs of zero do not equal zero.
Off-topic question: Which programming language do you use for you implementation? You could use the interval testing framework ITF1788, which currently supports C++, Octave, and Julia (and can easily be extended to other languages). There already exist many unit tests for IEEE Std 1788-2015 arithmetic, which can be of benefit for your project.
There are also test cases with decorations (most taken from libieeep1788 by Marco Nehmeier).
Oliver