Re: Non-deterministic encoding

The only time I've heard of issues like this has to do with the floating
point unit using 16 extra bits for accuracy which aren't preserved across
context switches.  Also, don't the SSE2 extensions alias on the floating
point registers?  Does that have an impact?

As a test in the off-chance that this is your problem (which I doubt), can
you run a test on a relatively unloaded dual-core machine and set processor
affinity to one core?  If that gives different results than you are seeing,
then that should be telling.

--Keith