[cpp-threads] A question about N2153

[Peter Dimov]

Alexander Terekhov wrote:
> On 1/17/07, Peter Dimov <pdimov at mmltd.net> wrote:

>> if( fetchadd_release( &ref_count, -1 ) == 1 ) // old value
>> {
>>    acquire_fence();
>>    destroy_object();
>> }

[...]

>> this is fine on x86/SPARC. On PowerPC, however, the most efficient
>
> As you surely recall, the most efficient on PowerPC is the following:

[...]

Right. I was talking about the most efficient PPC rendition of the above 
"portable" example in order to understand the semantics of N2153 
acquire_fence. Imagine for the sake of discussion that the programmer 
doesn't maintain platform-specific versions and just uses this code 
everywhere.

The state machine in the N2153 notation might be expressed as something 
along the lines of

bool r = true;
long v = load_raw( &count );

do
{

    if( r && v != 1 )
    {
        release_fence(); // lwsync
        r = false;
    }

} while( !compare_and_exchange_raw( &count, &v, v - 1 ) );

if( v == 1 )
{
    acquire_fence(); // need isync here for max efficiency
    destroy_object();
}

but this is platform-specific since it won't be optimal on IA-whatever. It's 
still an interesting case to analyze on the four architectures, of course, 
but let's proceed one thing at a time. :-)