Boost logo

Boost :

From: Andy Glew (glew_at_[hidden])
Date: 1999-12-30 20:01:33


> I recall a convincing presentation to the committee
> that you can't trust library-based threading because
> optimizers might reorder operations around the library
> calls. That might be what Bjarne was referring to.

True enough. Optimizing compilers create a weakly ordered
memory model.

I *love* it. We hardware guys have just about given up on
weak ordering - instead, we reorder operations, and then
"snoop" them to detect ordering model violations.

Pity compilers don't do that.

Maybe we should disable reordering compiler optimizations,
just do it in hardware, so that multithreading works transparently.


Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk