How to write a floating point constant

Question

I'm not aware and cannot quicly find the right way to enter floating point constant in C++.

If i want 2^-52, what should i write ? And, what does << with float ? Is that correct ?

const double pres = 1>>52

using unions to shift "mantissa" part of a float is undefined behaviour. — huseyin tugrul buyukisik, Aug 07 '12 at 06:44
Answered here(duplicate?) http://stackoverflow.com/questions/11806581/forms-of-constants-for-high-performance-addition-and-multiplication-for-double/11806593#11806593 — huseyin tugrul buyukisik, Aug 07 '12 at 06:46
No - that's not a dupe - for constants such as this there a better solutions than unions and bit fiddling. — Paul R, Aug 07 '12 at 06:58

Henrik · Accepted Answer · 2012-08-07T06:54:37.293

5

Looks like you really want the precision of double representation. In this case, don't use magic constants. Instead you can use this:

const double pres = std::numeric_limits<double>::epsilon();

edited Aug 07 '12 at 06:54

answered Aug 07 '12 at 06:49

Henrik

score 3 · Answer 2 · answered Aug 07 '12 at 06:46

3

You can use hex float representation for this:

const double pres = 0x1p-52;

answered Aug 07 '12 at 06:46

Paul R

Hexadecimal floats are in C, but not in C++, for some reason. – R. Martinho Fernandes Aug 07 '12 at 06:53
Oh - that's annoying - I'd never really noticed - presumably g++ and other compilers support this in C++ as an extension ? – Paul R Aug 07 '12 at 06:56
1

Yes, I believe g++ supports that, but MSVC doesn't. – R. Martinho Fernandes Aug 07 '12 at 06:57
2

This is by far the single most obnoxious omission in C++11. – Stephen Canon Aug 07 '12 at 12:40

score 0 · Answer 3 · answered Aug 07 '12 at 06:50

0

#include<math.h>
double pres = 1/pow(2,52);

answered Aug 07 '12 at 06:50

rosjo

does gcc compute this statically ? – Enjolras Aug 07 '12 at 06:57
@Enjolras Chances are, it does. But that’s not guaranteed. – Konrad Rudolph Aug 07 '12 at 06:58

3 Answers3