[M3devel] results of threadtest program on Windows7
hendrik at topoi.pooq.com
Wed Mar 2 04:39:45 CET 2011
On Tue, Mar 01, 2011 at 10:17:18AM -0800, Mika Nystrom wrote:
> Hi Jay,
> 2. Not sure how you could get negative timings. That would suggest that
> Time.Now() is decreasing---which seems contrary to its specification.
> Although the code is doing some nasty stuff with the timings, mainly
> reading and writing them without locks. But it's using INTEGER for
> the results so I don't think there can be an atomicity problem there.
> Even if there were coherence problems between processors (is this
> technically possible?) the "now" value has to be coherent since it's
> from the thread that prints the results, and the "then" values could
> only be getting older, not newer...
I've been on a machine before on which the clock occasinally ran
backwards -- notably a CDC Cyber machine. I don't doubt that OS and
hardware designers have found ways to replicate this behaviour on moderm
More information about the M3devel