Re: g_hash_table_resize (#59026)

[Owen Taylor <otaylor redhat com>]

Daniel Egger <egger suse de> writes:

> Am 19 Aug 2001 14:00:13 -0400 schrieb Owen Taylor:
> 
> > But gcc-2.96, which generally has considerably better optimization
> > misses the fact that the two branches of the if are exclusive.
>  
> > And its dependent on processor ... when I tried timing multiple
> > versions there was very little consistency in ordering between
> > a PIII and a celeron. (With total variation being small --
> > a few percent even in my artificial micro-benchmark)
> 
> Which benchmark is this? I can contribute some data for PPC and various
> compilers as well as sparc64, sparc32, Alpha, Pentium IV, Athlon,
> Itanium and various others if wanted. :) But I'm interested on seeing
> how this performs in comparison on PPC with gcc 3.0.1.

It's the one that I posted in my first mail:

====
  GHashTable *ht = g_hash_table_new (g_direct_hash, NULL);
  int i, j;

  for (i=0; i < 1000; i++)
    {
      for (j = 0; j < 1000; j++)
        g_hash_table_insert (ht, GUINT_TO_POINTER(j), GUINT_TO_POINTER(j));
      for (j = 0; j < 1000; j++)
        g_hash_table_remove (ht, GUINT_TO_POINTER(j));
    }
=====

(Actually, with 10000 iterations instead 1000 to get timeable data.)

Regards,
                                        Owen