Re:This isn't nearly as bad as the division bug (Score 1)

Just because you have two systems which conform to IEEE 754 doesn't guarantee you'll get the same answer for every floating point operation. You're easily able to produce different answers depending on the width of the registers in the FPU. Take a single-precision device which has a 64-bit internal floating point register vs one with a 32-bit one, both conform to 754, however due to rounding the one with a 32-bit register can provide a subtly different answer. The book Accuracy and Stability in Numerical Algorithms is pretty good on this.