The default version (interpolate8x8_ia64.s) of the interpolate functionssuffer from some rounding errors, but are fast. I our oppinion thereis no need to be such exact here. Just in the case we provide anotherset of function with exact rounding: interpolate8x8_ia64_exact.s