Re: [boost] scope_lock performance issue

From: Stefan Seefeld (seefeld_at_[hidden])
Date: 2003-10-18 11:33:42

Christophe Meessen wrote:

> It is an overkill for 32 bit values, but I just wanted to measure the 
> mutex protection not the assignement operations.

Are you alluding to assignment being an atomic operation for 32 values ?
Even if that is true, locking is not redundant (and thus 'overkill'), as
locks provide 'memory barriers', which are a means for the system to flush
caches, and thus assure that all CPUs see the same value for the given
variable. Without such a tool, there's just no way to assure that the
value is written down from the CPU's register through the various
intermediate caches to a place that is shared among all CPUs.

Of course, if you know your hardware, you can optimize for that. But
here we are talking about portable programming...

Regards,
                Stefan