Floating-point storage method and floating-point arithmetic device

• Main Authors: ISHIKAWA TOMOICHI, MIYASAKI SHUJI
• Format: Patent
• Language: Chinese; English
• Published: 28.01.2004
• Edition: 7
• Subjects: ACOUSTICS; CALCULATING; COMPUTING; COUNTING; ELECTRIC DIGITAL DATA PROCESSING; MUSICAL INSTRUMENTS; PHYSICS; SPEECH ANALYSIS OR SYNTHESIS; SPEECH OR AUDIO CODING OR DECODING; SPEECH OR VOICE PROCESSING; SPEECH RECOGNITION

本发明提供采用定点处理机可简单地处理浮点数值的浮点数值的存储方法等，在将实数x表示为a*(2 n)时的a作为尾数部分并以n作为指数部分时，用定点数值将尾数部分存储到N位(N≥(U+L))的位字段上位侧的U位，用整数将指数部分存储到下位侧的L位。在对这种格式的2个实数进行乘法运算的场合下，以两个实数作为定点数值进行乘法运算，并只以所得到的结果的上位有效位作为尾数部分，另一方面，以两个实数作为整数进行加法运算，并只以所得到的结果的下位有效位作为指数部分，这样就能够用浮点格式得到乘法运算结果。 In order to provide a method or the like for storing floating point numbers to make it easier to manage the floating point numbers using a fixed point processor, when a real number x is represented by a*(2^n) where a mantissa is a and an exponent is n, the mantissa is stored as a fixed point number in the upper U bits of N-bit field (N>=(U+L)) and the exponent is stored as an integer in the lower L bits. For the multiplication of two real numbers represented in such a format, these two real numbers are multiplied as fixed point numbers so as to make only the upper significant bits of the multiplication result a mantissa, while these two real numbers are added as integers so as to make only the lower significant bits of the addition result an exponent. As a result, the multiplication result can be obtained in a floating point format.