基本函數浮點運算器之設計

Abstract

本論文主要是專注於浮點基本函數計算的研究。我們嘗試三種方法來做浮點基本函數的計算，亦即Table-driven演算法、CORDIC演算法、以及CCM演算法。主要的目標精確度是IEEE Single and Double Extended Format。其中Table-driven演算法是承襲前人的架構，在實作上加以改進;而後兩者則是從Fix-point演算法加以推導設計出高效率並適用於浮點數的架構。在這篇論文中對三種方法的演算法發展、軟體模擬驗證、實現上的問題及硬體設計上的考量都會完整的涵蓋，並且在結尾處我們討論三種方法間以及和現存設計表現上的比較。文中我們也對精確度上的一些難題做探討，並做出詳細的模擬結果。

This thesis is concentrated on the evaluation of floating-point elementary functions. Our main destination is to achieve IEEE single and double-extended precision. Three techniques have been supposed to take the job. That is, Table-driven method, CORDIC algorithm, and CCM algorithm. For Table-driven method we follow the construction of my predecessors and improve it on implementation, whereas for the later two methods we deduce their algorithms from fixed-point to floating-point ones which can be efficiently realized on hardware. In the thesis the algorithm development, simulation verification with software, problems and considerations on hardware realization of the methods will be fully covered. After all comparison of the methods with each others and the existing works is discussed. A short conclusion summarizing our work and future task end the thesis.