35d
Subj : Re: full-blown experiential smr-based reference counted pointer impl...
To   : comp.programming.threads
From : Joe Seigh
Date : Fri Apr 01 2005 09:24 am

On Fri, 01 Apr 2005 14:49:37 +0200, Alexander Terekhov <terekhov@web.de> wrote:

> Joe Seigh wrote:
> [...]
>> > // Acquire Hazard
>> >  load target
>> > loop:
>> >  store target into hazard
>>
>>       ( store/load + store/store)
>
> There's no need store/store here.
>        ( store/load )
>
>> >  load target
>>
>> >  compare target with hazard, goto loop if non-equal
>
>         ( load/load + load/store )
>
> Acquire must be done with respect to the successful validation
> load. ABA, y'know.
>

I knew about the ABA problem, but you're right, the 2nd load
should have acquire semantics, a dependent load if the architecture
supports it.

-- 
Joe Seigh

.

0