sync_add_and_fetch problem

[Michael Meissner]

On Thu, Dec 04, 2008 at 02:38:08PM -0500, Charles Crisler wrote:
> OK, I have gotten it worked out now. Thank you for all of the help.
> 
> How can someone check what the default architecture setting is for their
> gcc? This seems to fit the evidence but I would really like to verify
> it.

Gcc defines different macros, based on what the -march=xxxx option is:

The trouble is the processor varients are exclusive (i.e. if you say
-march=i586, it defines __i586__ and not __i486__).  This means you might need
to add a bunch of #if's to capture all of the names.  At this writing, the
names defined are (I'm eliminating multiple names defined for the same target):

      __i486__
      __i586__
      __i686__
      __geode__
      __k6__
      __athlon__
      __k8__
      __amdfam10__
      __pentium4__
      __nocona__
      __core2__

The problem is if somebody in GCC 4.5 adds a new processor name.

If you can live with a more modern system as the base, you could use the
feature test macros (testing for __SSE__ for instance if your minimum target is
-march=pentium3).

Note, the one-laptop-per-child project is shipping a processor that does not
support the SSE instruction set, so by doing so, you might restrict where your
programs can run.

If you only anticipate running your program on your current machine and use a
recent compiler, you can use -march=native, and the compiler will figure out
what machine you are using and use all of the instructions that the machine
provides.

-- 
Michael Meissner, IBM
4 Technology Place Drive, MS 2203A, Westford, MA, 01886, USA
meissner@linux.vnet.ibm.com