On Thu, Mar 25, 2010 at 08:46:06AM -0700, Richard Henderson wrote: > On 03/25/2010 06:39 AM, Nathan Froyd wrote: > > On Wed, Mar 24, 2010 at 05:11:43PM -0700, Richard Henderson wrote: > >> Use __sync_bool_compare_and_swap to yield correctly atomic results. > >> As yet, this assumes running on an strict-memory-ordering host (i.e. x86), > >> since we're still "implementing" the memory-barrier instructions as nops. > > > > Did the approach taken by other targets (arm/mips/ppc) not work on > > Alpha? > > Mips doesn't even pretend to be atomic.
It pretends just as much as ppc and arm. See translate.c:OP_ST_ATOMIC. > Powerpc and Arm -- if I've got this straight -- use some sort of > stop-the-world > mutex+condition and then perform the compare-and-exchange by hand. I can't > see how that's better than using an actual compare-and-exchange provided by > the host cpu. In fact, I'm mildly horrified by the prospect. Oh, I didn't say it was pretty. But it does work fairly well in practice--enough to pass most of glibc's NPTL testsuite, for instance. (The remaining cases are tricky things, like cross-process locks.) I think--though Paul would remember better than I--that the stop-the-world approach might have been taken due to a desire to continue compiling with gcc < 4.1. I don't know how much of a desdirata that still is. (stop-the-world is also somewhat less complex than the previous implementation, which involved page protection games.) Certainly using actual compare-and-exchange would be much faster. -Nathan
