[gclist] Finalizers & Reference counting.

Lassi A. Tuura lassi.tuura@cern.ch
Thu, 29 Aug 2002 12:35:59 +0200

> As far as I can tell, nobody in this thread was able to produce real
> numbers on cache performance of GCC on x86 (despite the existence of
> CPU performance counters and tools to use them, which should allow
> that).  Someone produced numbers for GCC on AIX/Power4, which weren't
> great but which weren't appalling either:

Not for x86, but there was several for PPC (both AIX and Mac OS X I
think) and Alpha.  See for example the threads:

Do note that the discussion is still in an early stage.  People are
still gathering data with various tools, and more information will
come.  The numbers so far seem to indicate that there is a difference
between GCC 2.95.x and 3.x (GC was introduced between the two).  I
certainly do not have the competence to judge how significant the
differences are, or how good or bad the numbers are to begin with. 
Pointers to info on how to analyse this kind of information would be
most welcome.

There was also discussion on how GCC's memory allocator was spreading
objects into different pages (by size buckets I believe).  In some cases
this was causing closely-related objects to be allocated on different
pages, which of course also affects memory access patterns.  That is
obviously an issue with that particular implementation of GC; I have no
idea how hard-wired a design choice it is.

Horner's Five Thumb Postulate: experience varies directly
with equipment ruined.