| Message ID | 1373610771-11819-1-git-send-email-pingfank@linux.vnet.ibm.com |
|---|---|
| State | New |
| Headers | show |
On 07/11/2013 11:32 PM, Liu Ping Fan wrote: > Refcnt's atomic inc/dec ops are frequent and its idiom need no seq_cst > order. So to get better performance, it worth to adopt _relaxed > other than _seq_cst memory model on them. You'd need to update the documentation then. As it stands, what you've written looks like a bug. > +#ifndef _GLIBCXX_ATOMIC_BUILTINS This will never be defined. It's private to the libstdc++ implementation. See how we've defined things using __atomic elsewhere in the file, looking at one of the __ATOMIC defines. And in either case, it's better form to use positive tests than negative ones. I.e. #ifdef rather than #ifndef > #define atomic_fetch_inc(ptr) __sync_fetch_and_add(ptr, 1) > #define atomic_fetch_dec(ptr) __sync_fetch_and_add(ptr, -1) I'd prefer atomic_fetch_inc_relaxed, as that's more self-documenting. But I'll re-iterate the necessity of documentation in this area. r~
On Sat, Jul 13, 2013 at 12:24 AM, Richard Henderson <rth@twiddle.net> wrote: > On 07/11/2013 11:32 PM, Liu Ping Fan wrote: >> Refcnt's atomic inc/dec ops are frequent and its idiom need no seq_cst >> order. So to get better performance, it worth to adopt _relaxed >> other than _seq_cst memory model on them. > > You'd need to update the documentation then. As it stands, what you've written > looks like a bug. > Ok, will update atomic.txt >> +#ifndef _GLIBCXX_ATOMIC_BUILTINS > > This will never be defined. It's private to the libstdc++ implementation. See > how we've defined things using __atomic elsewhere in the file, looking at one > of the __ATOMIC defines. > Oh, I misunderstood the description of _GLIBCXX_ATOMIC_BUILTINS. Got it, thanks. > And in either case, it's better form to use positive tests than negative ones. > I.e. #ifdef rather than #ifndef > Will fix. >> #define atomic_fetch_inc(ptr) __sync_fetch_and_add(ptr, 1) >> #define atomic_fetch_dec(ptr) __sync_fetch_and_add(ptr, -1) > > I'd prefer atomic_fetch_inc_relaxed, as that's more self-documenting. > But if _relaxed not supported, it will fall back on _seq_cst, then atomic_fetch_inc_relaxed will be wrong named? Thanks and regards, Pingfan > But I'll re-iterate the necessity of documentation in this area. > > > r~
diff --git a/include/qemu/atomic.h b/include/qemu/atomic.h index 0aa8913..1f474b7 100644 --- a/include/qemu/atomic.h +++ b/include/qemu/atomic.h @@ -183,8 +183,15 @@ #endif /* Provide shorter names for GCC atomic builtins. */ +#ifndef _GLIBCXX_ATOMIC_BUILTINS +/* close to C11 memory_order_seq_cst */ #define atomic_fetch_inc(ptr) __sync_fetch_and_add(ptr, 1) #define atomic_fetch_dec(ptr) __sync_fetch_and_add(ptr, -1) +#else +/* C11 memory_order_relaxed */ +#define atomic_fetch_inc(ptr) __atomic_fetch_add(ptr, 1, __ATOMIC_RELAXED) +#define atomic_fetch_dec(ptr) __atomic_fetch_add(ptr, -1, __ATOMIC_RELAXED) +#endif #define atomic_fetch_add __sync_fetch_and_add #define atomic_fetch_sub __sync_fetch_and_sub #define atomic_fetch_and __sync_fetch_and_and @@ -192,8 +199,15 @@ #define atomic_cmpxchg __sync_val_compare_and_swap /* And even shorter names that return void. */ +#ifndef _GLIBCXX_ATOMIC_BUILTINS +/* close to C11 memory_order_seq_cst */ #define atomic_inc(ptr) ((void) __sync_fetch_and_add(ptr, 1)) #define atomic_dec(ptr) ((void) __sync_fetch_and_add(ptr, -1)) +#else +/* C11 memory_order_relaxed */ +#define atomic_inc(ptr) ((void) __atomic_fetch_add(ptr, 1, __ATOMIC_RELAXED)) +#define atomic_dec(ptr) ((void) __atomic_fetch_add(ptr, -1, __ATOMIC_RELAXED)) +#endif #define atomic_add(ptr, n) ((void) __sync_fetch_and_add(ptr, n)) #define atomic_sub(ptr, n) ((void) __sync_fetch_and_sub(ptr, n)) #define atomic_and(ptr, n) ((void) __sync_fetch_and_and(ptr, n))
Refcnt's atomic inc/dec ops are frequent and its idiom need no seq_cst order. So to get better performance, it worth to adopt _relaxed other than _seq_cst memory model on them. We resort to gcc builtins. If gcc supports C11 memory model, __atomic_* buitlins is used, otherwise __sync_* builtins. Signed-off-by: Liu Ping Fan <pingfank@linux.vnet.ibm.com> --- include/qemu/atomic.h | 14 ++++++++++++++ 1 file changed, 14 insertions(+)