value of double vs long double in C++

Question

in C++ I'm defining the following types:

double doubleType = 1.6e-300;
long double longDoubleType = 1.6e-300;

I'll then print the values using:

cout << "double is of size " << sizeof(doubleType) << " and value is " << doubleType << endl;
cout << "long double is of size " << sizeof(longDoubleType) << " and value is " << longDoubleType << endl;

my output reads:

double is of size 8 and value is 1.6e-300
long double is of size 12 and value is -1.43863e-264

Whats causing the difference in the interpretation of the values?

Well, [can't reproduce](https://ideone.com/Cv1ZVA). Probably platform dependent... — Rakete1111, Jan 16 '17 at 19:49
Note that long double literals use the suffix L. It should be `long double longDoubleType = 1.6e-300L;`. — François Andrieux, Jan 16 '17 at 19:50
*"long double is of size 12"* That's unusual. What compiler on what platform do you use? — Baum mit Augen, Jan 16 '17 at 19:50
@FrançoisAndrieux But the implicit conversion from `double` literal to `long double` should not lose precision. — Baum mit Augen, Jan 16 '17 at 19:52
[I can reproduce this with printf](http://ideone.com/dzUv0X), by providing the wrong format specifier, which suggests that you've somehow got (part of) the bytes of the long double interpreted as a double. Again, are you sure you used `cout` in your actual test? — user2357112, Jan 16 '17 at 19:52
@BaummitAugen: Why is that unusual? I'd call that very usual on x86. — Kerrek SB, Jan 16 '17 at 19:53
Also I get this output; `double is of size 8 and value is 1.6e-300` `long double is of size 16 and value is 1.6e-300`. — Person, Jan 16 '17 at 19:53
@KerrekSB iirc I have only ever seen 64 bit or 80 bit `long double`. — NathanOliver, Jan 16 '17 at 19:56
@NathanOliver: You're confusing value bits with object bits. — Kerrek SB, Jan 16 '17 at 19:57
@KerrekSB Why would you only use 80 value bits but store it in 96 bits? — NathanOliver, Jan 16 '17 at 19:59
@NathanOliver: Why is a the size of a struct not the sum of the sizes of its members? — Kerrek SB, Jan 16 '17 at 20:00
That's padding but I do not see why there would be padding in a built in type. a long double is not the same as a class type. — NathanOliver, Jan 16 '17 at 20:01
@NathanOliver: Every and any type can have padding (except `char`). The hardware processes 80-bit floats, but they need to be stored 4-byte aligned. — Kerrek SB, Jan 16 '17 at 20:02
@KerrekSB Interesting. I have never thought about it that way. TIL — NathanOliver, Jan 16 '17 at 20:08

score 0 · Answer 1 · answered Jan 16 '17 at 20:10

0

It doesn't seem likely that you got this behavior with std::cout. More likely, you were using printf with the wrong format specifier, something like:

printf("%g", longDoubleType);

This is undefined behavior. It's permitted to do just about anything, but one of the common ways it plays out is that some of the bits of longDoubleType are interpreted as a double instead of a long double.

With printf, I can reproduce your output. With std::cout, I can't.

answered Jan 16 '17 at 20:10

user2357112

260,549
28
431
505

I'm definitely using cout. I'm also using the minGW (minimalist GNU for Windows) compiler. My build command is "g++ main.cpp", nothing fancy. I also tried using the "-std=c++11" option and I get the same results. Can you check your build options, maybe I'm missing something you're including. – user2600449 Jan 16 '17 at 23:03
@user2600449: The minGW information helps. It looks like this is a [known problem](http://stackoverflow.com/questions/27537775/cout-long-double-issue) with minGW, stemming from a mismatch between the `long double` implementations used by minGW and the underlying Microsoft libraries. Other search results suggest it might be fixed in more recent versions (but it's probably not). – user2357112 Jan 16 '17 at 23:21
I saw some similar reports on some websites. I just installed the latest minGW a few weeks ago, so its probably not fixed yet. This seems like the most likely explanation. :/ – user2600449 Jan 16 '17 at 23:26

value of double vs long double in C++

1 Answers1