On 04/19/2012 02:09 PM, Stefan Berger wrote:
This patch improves on the previously added virAtomicInt operations
by testing for the compiler and if GCC >= 4.1 (not found in docs prior to
that) is used on Linux and has the appropriate processor (that I have
access to) then use the implementation based on the gcc-builtins.
I also did not look at other systems (cygwin, win32) that do not need
access to virAtomic right now.
---
src/util/viratomic.h | 88
+++++++++++++++++++++++++++++++++++++++++----------
1 file changed, 72 insertions(+), 16 deletions(-)
Index: libvirt-acl/src/util/viratomic.h
===================================================================
--- libvirt-acl.orig/src/util/viratomic.h
+++ libvirt-acl/src/util/viratomic.h
@@ -30,6 +30,22 @@
typedef struct _virAtomicInt virAtomicInt;
typedef virAtomicInt *virAtomicIntPtr;
+# define __VIR_ATOMIC_USES_LOCK
+
+# if defined(__GNUC__)
+# if ((__GNUC__ == 4) && (__GNUC_MINOR__ >= 1)) || (__GNUC__ > 4)
+# if defined(__linux__)
+# if defined(__i386__) || defined(__x86_64__) || \
+ defined(__powerpc64__) || defined(__powerpc__)
Overkill. If you are compiling with gcc, you have access to the
compiler builtins on ALL platforms. It is true that some platforms are
able to map the builtins to single assembly instructions, while other
platforms end up being calls into entry points provided by libgcc, but
the compiler takes care of all of that under the hood, without you
having to think about it. There's no need to filter on OS or on
architecture; filtering on compiler is good enough.
--
Eric Blake eblake(a)redhat.com +1-919-301-3266
Libvirt virtualization library
http://libvirt.org