摘要
针对基础数学函数在高性能计算中可靠性较差的问题,本文提出了一种核心运算隔离分段式处理的数学函数可靠性优化方法。首先采用寄存器运算对浮点异常情况进行编码优化,然后在浮点函数运算过程中,以集中运算的方式对核心运算进行隔离,并对运算异常阶段进行分段式处理,最后结合基础数学函数算法特点,以低误差的幂运算和求和运算为主体实现一种适用于基础数学函数的低误差计算方法。仿真实验结果表明,本文提出的方法对于以计算密集型存在的众多高性能计算,能够降低其计算误差。
Aiming at the problem that the reliability of the basic mathematical function is poor in the high performance calculation, this paper presents a mathematical function reliability optimization method for the core computing isolation segmentation processing. Firstly, the algorithm is used to optimize the floating point anomaly, and then the core operation is quarantined in the form of centralized operation during the operation of the floating point function, and the operation exception stage is segmented.Finally, the basic mathematical function algorithm Characteristics, low error of the power operation and summation operation as the main body to achieve a basic mathematical function of the low error calculation method. The simulation results show that the proposed method can reduce the computational error for many high performance calculations with computationally intensive.
出处
《科技通报》
北大核心
2017年第11期20-23,共4页
Bulletin of Science and Technology
关键词
数学函数
可靠性优化
核心运算隔离
分段式处理量法
函数误差优化
mathematical function
reliability optimization
core computing isolation
piecewise processing method
function error optimization
