Subj : Re: Memory Barriers, Compiler Optimizations, etc.
To   : comp.programming.threads
From : Alexander Terekhov
Date : Wed Feb 02 2005 03:49 pm


SenderX wrote:
[...]
> I an still wondering about intel stuff being able to reorder a write
> followed by a read to a different location...

That's because (think of op(X) as "store" and op(Y) as "load" to/from
different locations)

    op(X).release
    op(Y).acquire

doesn't prevent 

    op(Y).acquire
    op(X).release

reordering. 

In the case lock operations (now think IA64), reordering must not 
induce deadlocks (lock operation must not "suspend" preceding 
releases) but critical regions can "overlap" for better performance. 

regards,
alexander.

.