Subj : Re: stl & threads
To   : comp.programming.threads
From : David Schwartz
Date : Tue May 24 2005 11:14 pm


"Uenal Mutlu" <520001085531-0001@t-online.de> wrote in message 
news:d70ptf$nqj$05$1@news.t-online.com...

>>     Suppose I write code like this:
>>
>>  while(IsLockedByAny() && !IsLockedByMe()) i++;
>>
>>     What guarantees that the compiler doesn't optimize away the repeated
>> reads of 'cLocked' and spin forever? I admit that it's dumb to do 
>> something
>> like that, but it's also bad for a mutex class to contain errors.

> Yeah, a good argument.
> The solution is to make them volatile:
>  volatile size_t           cLocked;
>  volatile DWORD            dwThrId;

    That is not *guaranteed* to work, although it happens to (at least on 
every possible incantation of a platform that has Win32 critical sections 
that exists today). The 'volatile' keyword does not force any memory 
barriers or any other type of inter-CPU synchronization.

    DS

.