The IEEE problems are actually specific to x87 and the fact that x87 has extended precision in registers and single/double precision in memory. So your result

The IEEE problems are actually specific to x87 and the fact that x87 has extended precision in registers and single/double precision in memory. So your result can depend on whether the compiler spilled intermediate calculations to memory or kept them in registers. Luckily SSE doesn't have this issue. You get single or double precision consistently. IEEE actually gives a consistent functional specification for arithmetic that fits purely functional programming pretty well.