Subj : Re: Memory Barriers, Compiler Optimizations, etc.
To   : comp.programming.threads
From : Joseph Seigh
Date : Thu Feb 03 2005 09:03 pm

On Thu, 3 Feb 2005 17:10:02 -0800, SenderX <xxx@xxx.com> wrote:

>

>
>> If
>> the former, how are programmers in languages like C/C++ expected to make
>> the association between reads/writes and memory barriers?
>
>
> Just to clarify and answer your question directly:
>
>
> loads are usually associated with "consumer side" memory barriers.
> hoist-load-xxx, acquire, ect...
>
> stores are usually associated with "producer side" memory barriers.
> sink-store-xxx, release, ect...
>
>
>
>
I'm leaning toward the way
http://www.hpl.hp.com/research/linux/qprof/README_atomic_ops.txt
does it where the "memory barrier" is a qualifier to the operation
it's tacked on to.  Which means a load_acquire would have a
different membar than a store_acquire.  For sparc, load_acquire
would have "#LoadLoad | #LoadStore" and store_acquire would have
"#StoreLoad | #StoreStore".  For ia32 it would be MFENCE for both.


-- 
Joe Seigh

.