Re: Patch proposal: use __builtin_expect

[Andrew Haley <aph at pasanda dot cygnus dot co dot uk>]

> Date: 22 Apr 2000 18:00:30 -0000
> From: Andrew Haley <aph@cygnus.co.uk>
> 
> > Date: Sat, 22 Apr 2000 10:57:18 -0700
> > From: Anthony Green <green@cygnus.com>
> > 
> > Andrew wrote:
> > > Hmm.  So __builtin_expect means that we expect the result *not* to be
> > > true; it actually means "expect false". 
> > 
> > Not quite.  You always provide the expected result.  I always gave `0',
> > so it does mean expect false in those cases.
> > 
> >     __builtin_expect (x, 5)
> > 
> > ...means we expect x to be 5.  It seems safest/easiest to just expect 0.
> 
> Ah, so it's easy to replace those examples with 
> 
> __builtin_expect (x, false)
> 
> To say we expect the result to be false.  Better, no?

Or, come to think of it

#define if_unusually(x) __builtin_expect ((x), false)

jint
_Jv_MonitorEnter (jobject obj)
{
#ifndef HANDLE_SEGV
  if_unusually(! obj)
    JvThrow (new java::lang::NullPointerException);
#endif
  if_unusually(INIT_NEEDED (obj))
    obj->sync_init ();
  _Jv_SyncInfo *si = (_Jv_SyncInfo *) obj->sync_info;
  return _Jv_MutexLock (&si->mutex);
}

I'm not sure about the name "if_unusually"; maybe "if_exceptionally"
or just "UNUSUAL" might be better.  But I am sure that the syntax
should be sugared somehow...

Andrew.