rtld_total_time change is bogus...

[Alfred M. Szmidt]

   > The following change is bogus, rtld_total_time is defined if and
   > only if HP_TIMING_NONAVAIL is not define, but it tries to use it
   > even if it is not defined.  Patch attached that reverts the
   > behaviour as it was before.

   Try again, this time without removing the optimization.

I didn't remove any optimisation.  From [libc]/elf/rtld.c
(_dl_start_final):

...
#ifndef HP_TIMING_NONAVAIL
  hp_timing_t rtld_total_time;
...
#endif

  if (__builtin_expect (GL(dl_debug_mask) & DL_DEBUG_STATISTICS, 0))
    print_statistics (&rtld_total_time);

  return start_addr;
}

I don't see how that would be solved by any kind of optimisation...