Mere moments ago, quoth I:
On 15/10/2013 03:07, Quoth Andrey Semashev:
I don't think the commit is correct. [...] Probably the changes in interlocked.hpp should now also be wrapped in a test for BOOST_ATOMIC_X86_HAS_CMPXCHG8B, since you're right that this would be more correct.
But the intrinsic *does* exist on x86 >= Pentium, so defining it here means that the assembly code in platform_cmpxchg64_strong is not required. (Having said that, it would probably be harmless to completely omit the changes to interlocked.hpp.)
The changes in windows.hpp still seem relevant as noted above.
To rephrase this slightly: the patch and commit as current should improve behaviour while doing no harm (except possibly when compiling for < Pentium, but MSVC doesn't let you do that any more anyway). It probably could be further improved on, given other changes made to trunk in the meantime, however (such as closing that <Pentium loophole).