. . to see if there is any consistency in where it will run vs. where it won't. I've seen divide by 0 errors in older programs apparently caused by both these factors.

Of course the Borland Pascal "over 1-GHz" problem is well known, but I don't remember if that reported a divide by 0. Fortunately it was easily fixed if you had the source, and a run-time fix is available if you don't.