cbrt() of Linux libm seems buggy

cbrttest.c

#include <dlfcn.h>
#include <math.h>
#include <stdlib.h>
#include <stdio.h>
#define DBL_ULPOFONE 0x1p-52

void *load_cbrt(void **dlhp, char *path){
  *dlhp = dlopen(path, RTLD_LAZY);
  if (*dlhp == NULL) {
    fprintf(stderr, "%s\n", dlerror());
    exit(-1);
  }
  void *fptr = dlsym(*dlhp, "cbrt");
  if (fptr == 0) {
    fprintf(stderr, "%s\n", dlerror());
    exit(-1);
  }
  return fptr;
}
// typedef unsigned long int uint;

int main(int argc, char *argv[]){
  if (argc < 3) {
      fprintf(stderr, "usage:a.out /path/to/libm.so double\n");
      exit(-1);
  }
  void *dlh;
  double (*my_cbrt)(const double) = load_cbrt(&dlh, argv[1]);
  double d = atof(argv[2]);
  printf("my_cbrt(%.17g) = %.17g (@%p)\n", d,   my_cbrt(d), my_cbrt);
  printf("   cbrt(%.17g) = %.17g (@%p)\n", d,      cbrt(d),    cbrt);
  return dlclose(dlh);
}

cbrttest_simple.c

#include <math.h>
#include <stdlib.h>
#include <stdio.h>
int main(int argc, char *argv[]){
  if (argc < 2) {
    fprintf(stderr, "usage:a.out double\n");
    exit(-1);
  }
  double d = atof(argv[1]);
  return printf("cbrt(%.17g) = %.17g\n", d, cbrt(d));
}

cbrttest_simple.c is simple and portable that compiles with:

$ cc -Wall -O2 cbrttest_simple.c -lm

correct platforms should print:

$ ./a.out 0x1.ffffffffffffep-1
cbrt(0.99999999999999978) = 0.99999999999999989

but Linux prints:

$ ./a.out 0x1.ffffffffffffep-1
cbrt(0.99999999999999978) = 1.0000000000000002