Re: RFC: Multithreading design constraints

From: Miki Jovanovic (miki_at_[hidden])
Date: 2000-06-08 09:40:06

--- In boost_at_[hidden], John Maddock <John_Maddock_at_c...> wrote:
> >No, you need an atomic operation to do "lock.data = 0", so you've
> >again got a function call (or inline assembly -- completely off the
> >rails of defined behaviour).
> 
> At least on intel, all aligned writes are atomic, there is no need 
for
> assember in this case.

If we follow boost strategy, the atomic operations should be 
templetes, with specialisations for system specific fast routines. In 
the example above, who says that lock.data is not an instance with 
implemented assignment operator? We should provide read/write atomic 
functions even for this case.

Cheers,

Miki.