What is the precision of long double in C++?

Question

Does anyone know how to find out the precision of long double on a specific platform? I appear to be losing precision after 17 decimal digits, which is the same as when I just use double. I would expect to get more, since double is represented with 8 bytes on my platform, while long double is 12 bytes.

Before you ask, this is for Project Euler, so yes I do need more than 17 digits. :)

EDIT: Thanks for the quick replies. I just confirmed that I can only get 18 decimal digits by using long double on my system.

Erasure

Is it possible that it's converting things to regular double at some point? – Ogpu

Yes, that's definitely possible. I was up until 2AM checking my code for that last night, and I'm pretty sure it's long double all the way through, though. – Erasure

Answer 1

You can find out with std::numeric_limits:

#include <iostream>     // std::cout
#include <limits>       // std::numeric_limits
int main(){
    std::cout << std::numeric_limits<long double>::digits10 << std::endl;
}

Savarin

will that tell you the precision, or just the limits? – Ogpu

it tell you the precision. i.e how many digits you can have in the double. min() and max() tell you the limit – Savarin

More specifically it'll tell you how many you can round into a long double and not lose the digit. – Octodecimo

yeah, while min() and max() will tell you the limit you can have even if you loose precision. max is 1.18973e+4932 here, while digits10 is 18. – Savarin

Thanks for the reply. I'm getting the same results, 18 decimal digits. Now I'm wondering why 4 extra bytes of storage are needed to get one more decimal digit? – Erasure

On x86 it tends to be the 80-bit extended format: 64bits of mantissa, 15bits of exponent, probably going to be around 18 round safe digits. Double is 53bit mantissa or 16 round safe digits. – Octodecimo

What about max_digits10 vs digits10? – Scarito

Answer 2

You can use <cfloat>. Specifically:

LDBL_DIG

Octodecimo

Thanks for the link. It's LDBL_DIG that gives the decimal digits. LDBL_MANT_DIG gave me the binary digits. – Erasure