On Tue, Apr 24, 2012 at 10:09:10AM -0400, Stefan Berger wrote:
On 04/20/2012 11:56 AM, Laine Stump wrote:
On 04/20/2012 06:44 AM, Stefan Berger wrote:
This patch improves the previously added virAtomicInt implementation by using gcc-builtins if possible. The needed builtins are available since GCC>= 4.1. At least the 4.0 docs don't mention them.
--- src/util/viratomic.h | 95 ++++++++++++++++++++++++++++++++++++++++++--------- 1 file changed, 79 insertions(+), 16 deletions(-) ACK.
Now we can start thinking about eliminating locking for things that just need a refcount :-) Of which there aren't too many yet... I'll push this one later today. Thanks.
We can't eliminate locking for the objects in general, but we can eliminate locking in places where we only touch the ref-count, while keeping the locks for aall other API calls on the object. This would make like easier for us in a number of places, by avoiding potential lock aqcuisition order inversion problems we've hit in the past - particularly in the async callbacks/event dispatch Regards, Daniel -- |: http://berrange.com -o- http://www.flickr.com/photos/dberrange/ :| |: http://libvirt.org -o- http://virt-manager.org :| |: http://autobuild.org -o- http://search.cpan.org/~danberr/ :| |: http://entangle-photo.org -o- http://live.gnome.org/gtk-vnc :|