Size of "long long" in 128-bit machine?

Question

In a 128-bit RISC-V (or other 128-bit machine), how big are "long" and "long long" data types in C/C++?

To clarify: what are the sizes that an implementer of a compiler might be expected to use when writing the limits.h file for such a machine, in the absence of other implementations to conform to?

Answer 1

They can be any size the compiler writer wanted, subject to them being at least as large as their predecessor type (long int for long long int and int for long int) and satisfying the minimum ranges set out in the standards.

See, for example, 5.2.4.2 Numerical limits in C11, which states the minimum range required:

long int
    LONG_MIN           -2147483647 // −(2^31 − 1)
    LONG_MAX           +2147483647 //   2^31 − 1
long long int
    LLONG_MIN -9223372036854775807 // −(2^63 − 1)
    LLONG_MAX +9223372036854775807 //   2^63 − 1

Note that these aren't your full two's complement range since they have to account for the other two encoding schemes as well, ones' complement and sign-magnitude, both of which have the concept of negative zero.

If you really want to know, you can just look up those values in the limits.h header file, or compile and run:

#include <stdio.h>
#include <limits.h>

int main (void) {
    printf ("BITS/CHAR %d\n", CHAR_BIT);
    printf ("CHARS/SHORT %d\n", sizeof(short));
    printf ("CHARS/INT %d\n", sizeof(int));
    printf ("CHARS/LONG %d\n", sizeof(long));
    printf ("CHARS/LLONG %d\n", sizeof(long long));
    putchar ('\n');

    printf ("SHORT MIN %d\n", SHRT_MIN);
    printf ("SHORT MAX %d\n", SHRT_MAX);
    printf ("INT MIN %d\n", INT_MIN);
    printf ("INT MAX %d\n", INT_MAX);
    printf ("LONG MIN %ld\n", LONG_MIN);
    printf ("LONG MAX %ld\n", LONG_MAX);
    printf ("LLONG MIN %lld\n", LLONG_MIN);
    printf ("LLONG MAX %lld\n", LLONG_MAX);
    return 0;
}

On my system, a fairly bog-standard one (and slightly reformatted to look pretty):

BITS/CHAR      8
CHARS/SHORT    2
CHARS/INT      4
CHARS/LONG     4
CHARS/LLONG    8

SHORT MIN                 -32768
SHORT MAX                  32767
INT MIN              -2147483648
INT MAX               2147483647
LONG MIN             -2147483648
LONG MAX              2147483647
LLONG MIN   -9223372036854775808
LLONG MAX    9223372036854775807

So it looks like, on this system, I have two's complement (the 8/7 mismatch on the last digit of negative/positive numbers), no padding bits, 16-bit short int, 32-bit int and long int and 64-bit long long int.

If you run similar code in your own environment, that should be able to tell you similar information for it.