intel/compiler: set correct precision fields for 3-source float instructions