一种单精度浮点对数运算的硬件实现

The Hardware Implement of Logarithm in Single Precision Floating-point

超越函数计算在高性能微处理器、DSP以及GPU的设计中均会涉及,对数是其中比较重要的一种运算。目前的做法一般是采用多项式展开或者查表法：采用多项式展开法达到要求的精度,需要计算的乘法和加法非常多;若采用直接查表法,所需要的ROM资源将非常多。该文提出了一种高效的单精度浮点对数运算的实现方法,采用展开式和查表相结合的方法,可以实现对任意底数的对数运算,同时在运算器数量和ROM资源数量间达到平衡。

Transcendental function computing is used in high-performance microprocessors, DSP and GPU, the logarithm is one of the important operations. Most implementation is generally used polynomial expansion or look-up table method： achieving high precision accuracy needs a lot of multiplication and addition operations while using polynomial expansion method; the ROM resources will be very much while using direct look-up table method. This paper presents an efficient single-precision floating-point logarithm algorithm, using the polynomial expansion and look-up table combination, which can achieve any base of the logarithm, while the resources of ALU and ROM are balanced.