This is extremely useful, since the most common action after a failed
compare-and-swap is to loop around, trying again with the current
value as found in memory.
Code currently written as:
do {
Type value = atomic.load();
...
} while (!atomic.testAndSetRelaxed(value, desired));
Becomes:
Type value = atomic.load();
do {
...
} while (!atomic.testAndSetRelaxed(value, desired, value));
In most CPU architectures, the value that was found in memory is known
to the compare-and-swap code, so this is more efficient than the
previous code. In architectures where the value is not known, the new
code is no worse than before.
The implementation sometimes modified an existing function, sometimes
it added a new one, depending on whether more registers were needed in
the assembly (like ARMv6-7), the code became more complex (ARMv5), the
optimizer failed (C++11), or it was just plain equivalent (MIPS).
Change-Id: I7d6d200ea9746ec8978a0c1e1969dbc3580b9285
Reviewed-by: Oswald Buddenhagen <oswald.buddenhagen@digia.com>
Reviewed-by: Lars Knoll <lars.knoll@digia.com>
Reviewed-by: Thiago Macieira <thiago.macieira@intel.com>
634f82f1f1