许多现代浮点单元的一项主要体系结构功能，是能够将一个乘法与后面紧跟的加法作为单个运算执 行，且没有中间舍入误差。例如，Intel 的 Itanium 体系结构提供了一些指令，将三元运算 (a*b+c) 、(a*b-c) 和 (c-a*b) 中的每一个都组合为单个浮点指令（分别为 fma、fms 和 fnma）。这些单个 指令都比执行独立的乘法和加法指令快，并且因为没有中间乘法舍入，所以更为精确。该优化可以显 著提高那些含有多个交错乘法和加法运算的函数的速度-many modern floating point unit of a major system structure and function is to be a follow closely behind multiplication and addition of implementation as a single operator, with no intermediate rounding error. For example, Intel's Itanium architecture provides a number of directives to the ternary operator (a * b, c), (a * b-c) and (c-a * b) each have a portfolio to a single floating point instructions (Fullmetal respectively, and fms fnma) . These individual instruction than the implementation of an independent and additive multiplication instructions quickly, and because no middle multiplication into homes, so more precise. The optimization can significantly increase those containing more than staggered multiplication and addition operations function of the speed