JPH04252330A - Processing system for floating point addition/subtraction - Google Patents

Abstract

PURPOSE:To process the floating point addition/subtraction at a high speed by omitting the useless shift operation time and computing time. CONSTITUTION:In the first invention, an exponent deciding means (circuit) 5 decides whether the exponents are equal to each other among input date. If so, the addition/subtraction processing is immediately applied to the mantissas without carrying out the processing of a shift control data output circuit 1. In the second invention, a scale-over deciding part 5a decides whether the difference of the exponents exceeds a digit matching scale or not. If so, the mantissa of the input data on a larger exponent is outputted as the processing result without performing any shift operation nor addition/subtraction processing. Thus the addition/subtraction result of floating point is outputted at a high speed.

Description

[Detailed description of the invention]

0001

[Industrial Application Field] This invention relates to addition and subtraction processing of floating point data, and is designed to omit unnecessary operations when the exponent parts are equal and when the exponent difference overflows the digit alignment. The results are output at high speed.

0002

2. Description of the Related Art FIG. 2 is a block diagram showing an example of a circuit related to conventional floating-point mantissa addition/subtraction operations disclosed in Japanese Patent Publication No. 1-60857.
is a shift control data output circuit, 2 and 3 are shift circuits, and 4 is a shift control data output circuit.
is an adder.

Next, the operation will be explained. The exponent parts of the first and second input data are compared by the shift control data output circuit 1 to calculate shift control data. Next, in order to align the digits of the mantissa part of the first input data, a shift operation is performed in a shift circuit 2 based on the shift control data generated by the shift control data output circuit 1. Similarly,
In order to align the digits of the mantissa part of the second input data, the shift control data generated by the shift control data output circuit 1 is
The shift circuit 3 performs a shift operation based on the data. Then, the shifted and digit-aligned mantissa parts are added by an adder 4.

0004

[Problem to be Solved by the Invention] Since the conventional floating-point addition/subtraction device is configured as described above, it is necessary to always shift the mantissa even for input data that does not require digit alignment. In addition, addition operations must always be performed even for input data with exponent differences that are so large that digit alignment cannot be performed, resulting in problems such as the inability to quickly output results due to wasted operations. .

[0005]

[Means for solving the problem] This first invention is based on the first invention.
, an exponent part determination means (exponent part determination circuit 5) for determining whether or not the exponent parts in each of the second input data are equal;
When the exponent parts are determined to be equal by the exponent part determining means, the mantissa part is immediately subjected to addition/subtraction processing without executing the processing by the shift control data output circuit 1.

[0006] The second invention includes a scale over determination section 5a that determines whether or not the difference in exponent parts exceeds the scale of digit alignment. When it is determined that the input data is a bar, the mantissa part of the input data having the larger exponent part is output as a processing result without performing a shift operation and addition/subtraction processing.

[0007]

[Operation] It is determined whether the exponent parts of the first input data and the second input data are equal, and when it is determined that they are equal, the shift control data output circuit and the shift The circuit is passed and the mantissa part is immediately transferred to addition/subtraction processing. Also, the difference between the exponent parts changes the scale of digit alignment.
If the input data is large enough to cause a bar, the mantissa part of the larger input data is output as the processing result.

[0008]

DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. In FIG. 1, 1 is a shift control data output circuit;
2 and 3 are shift circuits; 4 is an adder; 5 is an exponent part determination circuit as an exponent part determination means for determining whether or not the exponent parts are equal; and 5a is a circuit in which the difference in the exponent parts controls the scale of digit alignment. -bar-
scale-over determination unit, 6, 7 for determining whether or not to
8 is a 2-line selection circuit, and 8 is a 3-line selection circuit.

Next, the operation will be explained. A shift control data output circuit 1 compares the exponent parts of the first input data and the second input data to calculate shift control data. Next, the shift circuits 2 and 3 perform digit alignment of the mantissa part using the shift control data. Then, the adder 4 adds the mantissa parts whose digits have been aligned. This operation is the same as the conventional one.

On the other hand, when the exponent part determination circuit 5 determines that the exponent parts are equal to each other, the selection circuits 6 and 7
Both input data of the mantissa part become the input data of the adder 4 as they are.

On the other hand, if the exponent parts are not equal, the output of the result of the shift operation becomes the input data of the adder 4.

Further, when the scale over determination section 5a determines that the difference in the exponent part is a scale over and is greater than the value that allows digit alignment, the selection circuit 8 selects the difference in the exponent part. The larger mantissa input data is selected and output as a result without any shift processing or arithmetic processing.

On the other hand, if digit alignment is possible, the selection circuit 8 selects the calculation result of the adder.

Note that when the input data is m×10n, m is the mantissa part and n is the exponent part.

For example, the first input data is 3.15×1
0-5 and the second input data is 2.14 x 10-4, the exponent parts are not the same, so the decimal point of the mantissa part of the second input data is shifted to 21.40. ×10-5
changed as .

According to the present invention, the first input data is
The input data is 3.15×10-5 and the second input data is 2.1
If it is 4×10 −5 , the exponent parts are the same, so the shift operation is omitted.

Further, according to the present invention, for example, the first input data is 3.15×10 −5 and the second input data is 2.
．． In the case of 14×105, the exponents are too different,
Since the first input data falls within the error range, digit alignment is meaningless, and 2.14 x 105 is directly output to the outside as the calculation result.

[0018]

Effects of the Invention The first invention is configured such that when it is determined that the exponent parts are equal, the mantissa part is immediately subjected to addition and subtraction processing without executing the processing by the shift control data output circuit, thereby eliminating unnecessary shifts. The effect is that time can be shortened and floating point addition/subtraction processing can be performed at high speed.

[0019] In the second invention, when it is determined that the difference in the exponent part exceeds the digit alignment scale, the input data with the larger exponent part is changed without performing shift operation and addition/subtraction processing. Since the structure is configured to output the mantissa part as the processing result, unnecessary shift time and calculation processing time can be reduced.
The effect is that floating point addition/subtraction processing can be performed at high speed.

[0020]

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a floating-point addition/subtraction processing method according to an embodiment of the present invention.

FIG. 2 is a block diagram of a conventional floating-point addition/subtraction device.

[Explanation of symbols] 1 Shift control data output circuit 2, 3 Shift circuit 4 Adder 5 Exponent part judgment circuit 5a Scale over determination section 6, 7, 8 Selection circuit

Claims (2)

[Claims] [Claim 1] A device that performs addition/subtraction processing on first input data and second input data divided into a mantissa part and an exponent part, wherein each of the above-mentioned input data is processed by a shift control data output circuit. The exponent part of the data is compared and calculated, and shift control data is calculated based on the result of this calculation.The shift circuit shifts the decimal point position of the mantissa part of each input data based on the shift control data. In the floating-point addition/subtraction processing method, the digits are aligned using
The present invention is characterized in that an exponent part determining means is provided for determining whether or not the exponent parts in the data are equal, and when the exponent part determining means determines that the exponent parts are equal, the mantissa part is immediately subjected to addition/subtraction processing. Floating point addition/subtraction processing method. [Claim 2] The first input data and the second input data, which are divided into a mantissa part and an exponent part, are subjected to addition and subtraction processing, and each of the above-mentioned input data is processed by a shift control data output circuit. The exponent part of the data is compared and calculated, and shift control data is calculated based on the result of this calculation.The shift circuit shifts the decimal point position of the mantissa part of each input data based on the shift control data. In a floating-point adder/subtractor that performs digit alignment, adds and subtracts the digit-aligned mantissa, and outputs it, the difference between the exponent parts of each input data is determined, and this difference is used as the digit-aligned mantissa. Skate
A scale over determination section is provided to determine whether or not the scale is over, and when it is determined that the scale is over, the exponent part can be adjusted without performing a shift operation or addition/subtraction processing. A floating point addition/subtraction processing method characterized in that the mantissa part of the larger input data is output as a processing result.