JPH09190335A - Cpu - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide a CPU which has a calculation precision calculating function with which a user can easily recognize a part of a program where the precision may possibly be deteriorated during the execution of calculation and also in the secondary processing that is carried out after the calculation. SOLUTION: An existing double precision floating point data format is extended, and the effective digit number (VD) and the effective precision (E) of the variable value of an arithmetic result can be stored in the extended format. That is, an ALU 11 calculates the numbers (VN) and precision (E) of arithmetic variables stored in the operand register parts 12 and 13 in addition to the numerical value of those variables. Then the ALU 11 holds the calculated numbers (VN) of arithmetic variables at a register part (effective digit number part) 14c and holds the precision (E) of the variables at a register part (effective precision part) 14d respectively.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a CPU applied to an electronic computer, and more particularly to a C having a calculation accuracy calculation function.
Regarding PU.

[0002]

2. Description of the Related Art Conventionally, a CPU mounted on an electronic computer stores numerical data of arithmetic variables, but the number of significant digits and the precision of a variable after arithmetic operation are unique to that variable. It was not saved as data.
FIG. 2 is a schematic explanatory view of a conventional technique, and here, FIG.
The flow of the calculation is shown in FIG. 5, and the format example of the variable is shown in FIG.

When an arithmetic operation is performed by the conventional electronic computer as described above, there is a single-precision / double-precision calculation precision management function for arithmetic variables, but "precision reduction due to rounding" and "digit loss" occur when performing the arithmetic operation. However, the user cannot directly know the location where the accuracy has deteriorated.

Therefore, it is not possible for the user to directly know at which point in the program and how much accuracy is reduced. In order to avoid this decrease in accuracy, conventionally, the user has taken various special measures and carefully designed the software.

For this reason, in the past, in the arithmetic processing when ultra-high accuracy is required, the reliability of the accuracy of the calculation result depends on the special treatment for avoiding the deterioration of accuracy. There was a problem in terms of reliability.

[0006]

As described above, conventionally, when an arithmetic operation is carried out by an electronic computer, there is a single-precision / double-precision calculation accuracy management function for arithmetic variables, but " In the case where "precision decrease due to rounding" or "precision decrease due to digit loss" occurs, it is not possible for the user to directly know the location of the precision decrease.

Therefore, conventionally, it is not possible for the user to directly know where in the program and how much the accuracy is reduced. Therefore, in the past, in order to avoid this decrease in accuracy, the user has made various special measures and carefully designed the software. However, since the accuracy depends on this measure, the accuracy of the calculation result is reduced. There was a problem in terms of reliability.

The present invention has been made in view of the above circumstances,
An object of the present invention is to provide a CPU having a calculation accuracy calculation function, by which a user can easily recognize a position where accuracy may decrease in a program by secondary processing during and after calculation. .

[0009]

SUMMARY OF THE INVENTION The present invention extends an existing double precision floating point data format so that it can be stored in a format in which the number of significant digits of a calculation result and the effective precision are extended. That is, specifically, ALU (arithmetic logic unit)
In this case, not only the arithmetic operation of the variable value but also the number of significant digits and effective precision of each variable can be calculated. For each variable at this time, enter the numerical value, the number of significant digits, and the precision in advance, calculate the number of significant digits and the precision of the calculation result in each calculation, and use this information as the information attached to the arithmetic variable data. Save (overwrite) together. FIG. 2 is a schematic explanatory diagram relating to the present invention at this time. FIG.
In the above, 14c is an arithmetic register section (effective digit section) which holds the effective digit number (VN) of the arithmetic variable, and 14d is an arithmetic register section (effective precision section) which holds the effective accuracy (E) of the same variable. is there.

[0010]

The ALU (arithmetic logic unit) is not limited to the numerical values of arithmetic variables, but the number of significant digits (VN) and effective precision (E) of the variables.
Then, the effective digit number (VN) and the effective precision (E) of the arithmetic variable are stored in a specified place in the existing extended format of the multiple floating point. The number of effective digits (VN) and the effective precision (E) are updated every time the calculation is performed, like the numerical value.

[0011]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a schematic explanatory view of the present invention.
Reference numeral 11 is an ALU (arithmetic logic unit). This ALU11
Calculates not only the numerical value of an arithmetic variable, but also the significant digits (VN) and effective precision (E) of the same variable, and places the significant digits ( VN), effective precision (E) are saved.

Numerals 12 and 13 are operand register sections for storing respective operation data of operation values and operands to be inputted to the ALU 11, and 14 is an operation result register section for storing operation result data.

14c is the number of significant digits (V
N) is a register unit (effective digit number part) that holds 14
d is a register unit (effective precision unit) that holds the effective precision (E) of the variable.

The ALU 11 is an operand register unit 1
Not only the numerical values of the arithmetic variables stored in Nos. 2 and 13, but also the number of significant digits (VN) and the effective precision (E) of the variables are calculated.
Then, as shown in FIG. 7B, the effective digit number (VN) and the effective precision (E) of the arithmetic variable are stored in a specified position in the existing multiplex floating point extended format. That is, the register part (the significant digit number portion) 14c has a significant digit number (V
N) is held, and the effective precision (E) of the variable is held in the register unit (effective precision unit) 14d. This number of significant digits (V
N) and the effective precision (E) are updated every time the calculation is performed, like the numerical value.

An embodiment according to the present invention will be shown below.
This will be described with an example.

Embodiment 1 In an OS (operating system) of a computer, a command for outputting a numerical value of a variable to a screen or to a file specifies that the effective digit number and the effective precision of the variable are also output at the same time. When an option is newly added (using a command in C language as an example): Program example printf (“Z =% VN / n”, Z) (% VN: number of significant digits in numerical value output command, Option to specify effective precision output) Output example Z = 311345, VN = 6, E = 0.00
It becomes 001. In this case, the user can directly visually confirm the effective digit number (VN) and the effective precision (E) of the variable after performing the arithmetic operation on the screen. Further, by outputting the above-mentioned effective digit number (VN) and effective accuracy (E) to a file and performing statistical processing of the data, it is possible to recognize in which part of the program the user designed the accuracy deterioration easily occurs. Is possible.

Embodiment 2 In the OS (operating system) of a computer, when a new command for returning the effective digit number of a variable and effective precision is added as a return value (a command in C language is used as an example), -Program example: If the number of significant digits of the variable is less than 3, an error message is displayed.

If (VN (variable name) <3) printf ("E
rror. Check Here! )); VN (variable name): Command that returns the number of significant digits of the variable ・ Program example: If the number of significant digits of the variable is less than 3, it automatically switches to another arithmetic logic that does not reduce the precision. .

If (VN (variable name) <3) AnotherLog
ic (); AnotherLogic (): This is a function that performs processing for avoiding accuracy deterioration. In this case, it is possible to detect the decrease in accuracy after the arithmetic operation and display the error message.

As a result, the user can directly recognize the location where the accuracy is lowered in the program, and it is possible to obtain the same effect as that of the first embodiment. In addition to that, it becomes possible to automatically switch to the arithmetic logic in the electronic computer that avoids the deterioration of accuracy.

When a function for automatically switching the arithmetic logic is added, if a function for performing processing for avoiding a decrease in accuracy is already prepared as a library function in the electronic computer, the user can execute the arithmetic operation after the arithmetic operation. It is possible to improve the reliability of the calculation result by using the arithmetic logic that automatically avoids the accuracy deterioration without specially programming the measures for avoiding the effective accuracy deterioration of the variable.

[0023]

As described above in detail, according to the present invention, the user can easily recognize the location in the program where the accuracy may deteriorate by the secondary processing during and after the calculation. CP with calculation accuracy calculation function
U can provide.

The specific effects of the present invention are listed below.

(1) By detecting the number of significant digits and automatically switching the arithmetic logic, it is possible to avoid a decrease in calculation accuracy due to arithmetic operation as much as possible, and to obtain a more reliable calculation result. become.

(2) It is possible to calculate the number of significant digits and the precision of each calculation and save the contents as data.

(3) After performing the numerical calculation, by statistically analyzing the number of significant digits, the accumulated error, etc., the user can easily recognize the variables and the arithmetic calculation formulas which are likely to cause the calculation error due to the deterioration of accuracy. Become. Further, by feeding back the analysis result to the design, it becomes possible to carry out the optimum program design satisfying the purpose of use of the user.

[Brief description of the drawings]

FIG. 1 is a schematic explanatory diagram of the present invention.

FIG. 2 is a schematic explanatory diagram of a conventional technique.

[Explanation of symbols]

11 ... ALU (arithmetic logic part), 12, 13 ... Operand register part, 14 ... Operation result register part, 14c ... Register part (effective digit number part) for storing effective digit number (VN) of arithmetic variable, 14d ... Arithmetic variable Register (effective precision part) that stores the effective precision (E) of VN, effective number of digits of VN ... arithmetic variable, E ... effective precision of arithmetic variable.

Claims (1)

[Claims] 1. A CPU comprising: an arithmetic operation means; and means for storing data representing calculation accuracy of a variable value generated when the arithmetic operation is performed by the means, as unique data of the variable.