static long sys_write(int fd, const char *buf, long len)
{
    long ret;
    __asm__ volatile (
        "syscall"
        : "=a"(ret)
        : "0"(1L),
          "D"((long)fd),
          "S"(buf),
          "d"(len)
        : "rcx", "r11", "memory"
    );
    return ret;
}

static void sys_exit(int code)
{
    __asm__ volatile (
        "syscall"
        :
        : "a"(60L),
          "D"((long)code)
        :
    );
    __builtin_unreachable();
}

static long strlen(const char *s)
{
    long n = 0;
    while (s[n]) n++;
    return n;
}

static void print_char(char c)
{
    sys_write(1, &c, 1);
}

static void str(const char *s)
{
    long n = strlen(s);
    if (n > 0) sys_write(1, s, n);
}

static void uint(unsigned long v)
{
    char buf[20];
    int  i = 19;
    buf[i] = '\0';
    if (v == 0) { print_char('0'); return; }
    while (v && i > 0) {
        buf[--i] = '0' + (v % 10);
        v /= 10;
    }
    str(buf + i);
}

static void print_int(long v)
{
    if (v < 0) { print_char('-'); uint((unsigned long)-v); }
    else         uint((unsigned long)v);
}

static void hex(unsigned long v)
{
    static const char *digits = "0123456789abcdef";
    char buf[17];
    int  i = 16;
    buf[i] = '\0';
    if (v == 0) { str("0x0"); return; }
    while (v && i > 0) {
        buf[--i] = digits[v & 0xf];
        v >>= 4;
    }
    buf[--i] = 'x';
    buf[--i] = '0';
    str(buf + i);
}

typedef __builtin_va_list va_list;
#define va_start(ap, last) __builtin_va_start(ap, last)
#define va_arg(ap, T)      __builtin_va_arg(ap, T)
#define va_end(ap)         __builtin_va_end(ap)

static void impl(const char *fmt, ...)
{
    va_list ap;
    va_start(ap, fmt);
    for (const char *p = fmt; *p; p++) {
        if (*p != '%') { print_char(*p); continue; }
        p++;
        switch (*p) {
            case 'd': print_int(va_arg(ap, int));                 break;
            case 'l': p++;  /* assume %ld */
                      print_int(va_arg(ap, long));                break;
            case 'u': uint(va_arg(ap, unsigned int));             break;
            case 'x': hex(va_arg(ap, unsigned long));             break;
            case 's': str(va_arg(ap, const char *));              break;
            case 'c': print_char((char)va_arg(ap, int));          break;
            case '%': print_char('%');                             break;
            default:  print_char('%'); print_char(*p);            break;
        }
    }
    va_end(ap);
}

#define print(fmt, ...) impl((fmt), ##__VA_ARGS__)

void _start(void)
{
    print("Hello, world!\n");
    print("Integer:  %d\n",  -42);
    print("Unsigned: %u\n",  1337u);
    print("Hex:      %x\n",  (unsigned long)0xDEADBEEF);
    print("String:   %s\n",  "no libc needed");
    print("Char:     %c\n",  'Z');
    print("Long:     %ld\n", (long)-9876543210L);
    print("Two args: %d + %d = %d\n", 10, 20, 30);
    sys_exit(0);
}