Subj : Re: full-blown experiential smr-based reference counted pointer impl...
To   : comp.programming.threads
From : Joe Seigh
Date : Fri Apr 01 2005 01:28 pm

On Fri, 01 Apr 2005 12:16:04 +0200, Alexander Terekhov <terekhov@web.de> wrote:

>
> Chris Thomasson wrote:
> [...]
>> // Acquire Hazard
>>  load target
>> loop:
>>  store target into hazard
>>    ( store/load )
>>  load target
> /*
>>    ( store/load + store/store )
> */
>>  compare target with hazard, goto loop if non-equal
>      (load/load + load/store) // cc/dd hints aside
>
> in probably better. You don't need to have that second fence inside
> loop.
>
>>
>> // inside gc region wrt target
>>
>> // Release Hazard
>>    ( load/store + store/store )
>>  store null into hazard
>>
>> Is that about it?
>
> You also need "acquire" on cleanup/object destruction.
>
That would be in the polling thread.  You need "acquire"
on any object destruction if it's not provided by any
synchronization primatives being used.  Naturally since
this is one of the things I already know about, I seem to
have managed to leave it out of RCU.  :)

Also you need release semantics on any stores into shared
pointers being guarded by SMR.  Did we leave anything out?

-- 
Joe Seigh

.