how to print __uint128_t number using gcc?

No there isn't support in the library for printing these types. They aren't even extended integer types in the sense of the C standard.

Your idea for starting the printing from the back is a good one, but you could use much larger chunks. In some tests for P99 I have such a function that uses

uint64_t const d19 = UINT64_C(10000000000000000000);

as the largest power of 10 that fits into an uint64_t.

As decimal, these big numbers get unreadable very soon so another, easier, option is to print them in hex. Then you can do something like

  uint64_t low = (uint64_t)x;
  // This is UINT64_MAX, the largest number in 64 bit
  // so the longest string that the lower half can occupy
  char buf[] = { "18446744073709551615" };
  sprintf(buf, "%" PRIX64, low);

to get the lower half and then basically the same with

  uint64_t high = (x >> 64);

for the upper half.

The GCC 4.7.1 manual says:

6.8 128-bits integers

As an extension the integer scalar type __int128 is supported for targets having an integer mode wide enough to hold 128-bit. Simply write __int128 for a signed 128-bit integer, or unsigned __int128 for an unsigned 128-bit integer. There is no support in GCC to express an integer constant of type __int128 for targets having long long integer with less then [sic] 128 bit width.

Interestingly, although that does not mention __uint128_t, that type is accepted, even with stringent warnings set:

#include <stdio.h>

int main(void)
{
    __uint128_t u128 = 12345678900987654321;
    printf("%llx\n", (unsigned long long)(u128 & 0xFFFFFFFFFFFFFFFF));
    return(0);
}

Compilation:

$ gcc -O3 -g -std=c99 -Wall -Wextra -pedantic xxx.c -o xxx  
xxx.c: In function ‘main’:
xxx.c:6:24: warning: integer constant is so large that it is unsigned [enabled by default]
$

(This is with a home-compiled GCC 4.7.1 on Mac OS X 10.7.4.)

Change the constant to 0x12345678900987654321 and the compiler says:

xxx.c: In function ‘main’:
xxx.c:6:24: warning: integer constant is too large for its type [enabled by default]

So, it isn't easy manipulating these creatures. The outputs with the decimal constant and hex constants are:

ab54a98cdc6770b1
5678900987654321

For printing in decimal, your best bet is to see if the value is larger than UINT64_MAX; if it is, then you divide by the largest power of 10 that is smaller than UINT64_MAX, print that number (and you might need to repeat the process a second time), then print the residue modulo the largest power of 10 that is smaller than UINT64_MAX, remembering to pad with leading zeroes.

This leads to something like:

#include <stdio.h>
#include <inttypes.h>

/*
** Using documented GCC type unsigned __int128 instead of undocumented
** obsolescent typedef name __uint128_t.  Works with GCC 4.7.1 but not
** GCC 4.1.2 (but __uint128_t works with GCC 4.1.2) on Mac OS X 10.7.4.
*/
typedef unsigned __int128 uint128_t;

/*      UINT64_MAX 18446744073709551615ULL */
#define P10_UINT64 10000000000000000000ULL   /* 19 zeroes */
#define E10_UINT64 19

#define STRINGIZER(x)   # x
#define TO_STRING(x)    STRINGIZER(x)

static int print_u128_u(uint128_t u128)
{
    int rc;
    if (u128 > UINT64_MAX)
    {
        uint128_t leading  = u128 / P10_UINT64;
        uint64_t  trailing = u128 % P10_UINT64;
        rc = print_u128_u(leading);
        rc += printf("%." TO_STRING(E10_UINT64) PRIu64, trailing);
    }
    else
    {
        uint64_t u64 = u128;
        rc = printf("%" PRIu64, u64);
    }
    return rc;
}

int main(void)
{
    uint128_t u128a = ((uint128_t)UINT64_MAX + 1) * 0x1234567890ABCDEFULL +
                      0xFEDCBA9876543210ULL;
    uint128_t u128b = ((uint128_t)UINT64_MAX + 1) * 0xF234567890ABCDEFULL +
                      0x1EDCBA987654320FULL;
    int ndigits = print_u128_u(u128a);
    printf("\n%d digits\n", ndigits);
    ndigits = print_u128_u(u128b);
    printf("\n%d digits\n", ndigits);
    return(0);
}

The output from that is:

24197857200151252746022455506638221840
38 digits
321944928255972408260334335944939549199
39 digits

We can verify using bc:

$ bc
bc 1.06
Copyright 1991-1994, 1997, 1998, 2000 Free Software Foundation, Inc.
This is free software with ABSOLUTELY NO WARRANTY.
For details type `warranty'. 
ibase = 16
1234567890ABCDEFFEDCBA9876543210
24197857200151252746022455506638221840
F234567890ABCDEF1EDCBA987654320F
321944928255972408260334335944939549199
quit
$

Clearly, for hex, the process is simpler; you can shift and mask and print in just two operations. For octal, since 64 is not a multiple of 3, you have to go through analogous steps to the decimal operation.

The print_u128_u() interface is not ideal, but it does at least return the number of characters printed, just as printf() does. Adapting the code to format the result into a string buffer is a not wholly trivial exercise in programming, but not dreadfully difficult.