> That surprises me—I would have thought a quick bit-test on the status register and conditional jump wouldn't be all that bad.

20%-30% performance loss at least, IIRC.

> Are you thinking something like a software interrupt on integer overflow?

Right.