[cpp-threads] C++ memory model

[Anthony Williams]

Mark Batty wrote:
> In the process of this formalization and proof some parts of the draft
> standard seemed difficult to understand or inconsistent - mostly
> relatively local points.  I've written a note describing the
> formalization and some specific suggestions for local clarifications
> and improvements to the standard.  There's a copy here:
> 
>  http://www.cl.cam.ac.uk/~mjb220/cpp/cpp.pdf

Re section 3.5, the intent of clause 29.3p8 of the working draft is to 
disallow "out of thin air" values. In the example given, the value 42 
can be stored into x by thread 2 iff it is stored into x by thread 2! 
The intent is that the execution (or not) of a particular operation 
cannot be conditional on its own execution.

Likewise, the intent of clause 29.3p9 is to ensure that code like the 
following works as expected --- values must eventually become visible to 
other threads:

std::atomic_bool flag(false);

Thread 1:
while(!flag.load()); // busy wait for flag

Thread 2:
flag.store(true); // wake thread 1

I do not think your suggest clarification helps here. The point is that 
compiler cannot optimize the loop to

bool const local_flag=flag.load();
if(!local_flag) while(true);

Anthony
-- 
Author of C++ Concurrency in Action | http://www.manning.com/williams
just::thread C++0x thread library   | http://www.stdthread.co.uk
Just Software Solutions Ltd         | http://www.justsoftwaresolutions.co.uk
15 Carrallack Mews, St Just, Cornwall, TR19 7UL, UK. Company No. 5478976