On Mon, Mar 14, 2011 at 4:42 PM, Marshall Lochbaum
<[email protected]> wrote:
> If J ever implements parallel architecture, the first thing to do would be
> to parallelize operations of rank less than or equal to _1 . At this point,
> any code that relies on side-effects and implicit loops will be in trouble.

Yes.

But that does not mean that they would occur out of order.  I believe
that "out-of-order sequential parallelism" is a quirk introduced by
time slicing implementations, and I believe that supporting it on true
parallel hardware would be a mistake.

It would be quite reasonable for the implementation to forbid them entirely.

It would also be plausible for the implementation to impose rank on
the side effect(s).

-- 
Raul
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm

Reply via email to