JPS6297031A - Matching system for bit number of mantissa part - Google Patents

Abstract

PURPOSE:To eliminate an extreme delay compared with a case where a mantissa part has the same number of bits by obtaining the same number of bits at a compiling stage after validating an automatic accuracy increasing function and also cutting the lower bits of the mantissa part. CONSTITUTION:In its own system an automatic accuracy increasing function is validated by a validating means of a complier when it is recognized that the number of bits of the mantissa part of another system is larger than that of its own system. As a result, the data of the single accuracy is turned into the data of the double accuracy and furthermore the data of the double accuracy is turned into the data of the quadruple accuracy respectively When a source program is supplied, an object program is produced while the lower bits of the matissa part are cut (the lower bits of the mantissa part are set O as necessary). In such a way, the result of execution to secure the same number of bits without a delay extremely reduced compared with a case where the mantissa part has the same number of bits is obtained.

Description

DETAILED DESCRIPTION OF THE INVENTION (Field of Industrial Application) The present invention provides a method for determining the number of bits of the mantissa when executing a program for another computer system having a different number of bits of the mantissa in floating point representation. Regarding matching method.

(Prior Art) In floating point arithmetic, when expressing the same number, the execution results of a program for a computer system may differ significantly depending on the number of bits in the mantissa part. Therefore, if you run a program created to run on another computer system with a different number of bits in the mantissa and the output results are different, the cause may be the difference in the number of bits in the mantissa, or It becomes difficult to determine whether there are others.

For this reason, conventionally, when executing a program for another computer system, each instruction is simulated to match the number of bits in the mantissa.

(Problems to be Solved by the Invention) In such a conventional method, it is necessary to perform bit manipulation for each instruction, so the means for doing so is large-scale, and the execution time for one process is extremely long. There is a problem with storing it away.

(Means for Solving the Problems) The method of the present invention is a method for matching the number of mantissa bits when executing a program for another computer system in which the number of bits of the mantissa in floating point representation is different.
The compiler of the computer system has a means for recognizing the difference in the number of bits of the input mantissa, and as a result of this recognition, if the number of bits of the mantissa of another computer system is greater than that of the own computer system, the precision means for activating an automatic extension function; and means for converting a constant of a program for another computer system into an internal value, and then setting the number of bits on the right side of the mantissa to 0 until the number of effective bits of the mantissa becomes the same as that of the other computer system. , means for inserting into the instruction string of the target program an instruction that sets the bits on the right side of the mantissa to 0 until the number of effective bits of the operation result becomes the same as that of another computer system after execution of the floating point operation instruction. It is characterized by having been established.

(Example) Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a flowchart showing the processing steps o-f in one embodiment of the present invention.

When trying to run a program created to run on another computer system (hereinafter simply referred to as a system) that has a different number of bits in the mantissa in floating point representation, first check the number of bits in the mantissa in the other system. , for example, input parameters from a workstation or the like. The system that is currently trying to execute the program (referred to as the local system) naturally recognizes the number of bits in its own mantissa, so the number of bits in the mantissa of another system recognized by parameter input (step ■ in Figure 1) It is possible to compare the size with the number of bits (step ■).

According to the result, if the number of bits in the mantissa part of the other system is not greater than that of the own system, the process moves from step (2) to step (2), but if it is greater, the process moves from step (2) to step (3). In step (2), an automatic accuracy extension function, which is generally provided in computer systems having a floating-point arithmetic function, is enabled. As is well-known, the automatic precision extension function is a function that automatically changes the data format to one step higher in precision.As a result, for example, single precision data becomes double precision data, and double precision The data will be quadruple precision data.

Next, when the source program is input (step ■),
The object program is generated while the lower bits of the mantissa are truncated (step 2). The process of truncating the lower bits is the process of reducing the lower bits (1) of the mantissa to 0 as many times as necessary.

For example, the number of bits in the mantissa of another system is 2 in single precision.
1 bit, double precision is 50 bits, and if the number of bits of the mantissa part of the local system is 24 bits for single precision and 56 bits for double precision, move from step ■ to step ■, bypassing step ■. However, in the process of truncating the lower bits in step (■), 3 (-24-21) bits are truncated for single precision, and 6 (=56-50) bits are truncated for double precision, so that they are the same as the number of bits in other systems. It will be done.

Also, if the number of bits in the mantissa of the other system is 27 bits for single precision and 63 bits for double precision, and the number of bits for the mantissa of the own system is 24 bits for single precision and 56 bits for double precision, then , Since the number of bits of the other system is larger, the process moves from step ■ to step ■, and the automatic precision expansion function is enabled. Thereafter, when proceeding to step (2) via step (2), the number of bits in the mantissa of the own system is first changed to 56 bits for single precision and 112 bits for double precision due to the precision automatic extension function. After that, the lower bits are truncated, and single precision is truncated by 29 (-56-27) bits, and double precision is truncated by 49 (= 112-63) bits, so that the number of bits is the same as that of other systems. .

The bit number truncation processing as described above is performed in the following processing steps of the compiler and in the instruction sequence generated by the compiler in the target program.

First, real constants and complex constants in floating point representation are converted to internal values of the own system at the compiler processing stage, and then converted to internal values.

Also, for real number type or complex number type assignment statements, when the value on the right side is calculated and the result of the calculation is assigned to the variable or array element on the left side, the calculation is performed on the result of the calculation. For example, in the assignment statement A=BsC*D, first input the value of B to a predetermined register R, then calculate register R*Ci and input the calculation result to register R, then register R*Di The calculation result is input into register R, and finally, when register R is set to a person, the number of bits is rounded down.

However, when a real value or a complex value is input by an input sentence, the input real value or complex value is processed immediately after the input.

(Effects of the Invention) According to the present invention, as described above, when attempting to execute a program for another system in which the number of bits of the mantissa in floating point representation is different, the number of bits is equalized at the compilation stage. In order to generate instructions that perform This has the advantage that it can be obtained without excessive delay compared to the case where there is no difference in the number of bits of the mantissa and only by adding large-scale means.

[Brief explanation of drawings]

FIG. 1 is a diagram showing a processing flow of an embodiment of the present invention. Agent Patent Attorney Shinhan Uchihara 1 Figure

Claims (1)

[Claims] In a method for matching the number of mantissa bits when a program for another computer system having a different number of bits of the mantissa in floating point representation is to be executed, the above input to the compiler of the computer system means for recognizing a difference in the number of bits of the mantissa part; means for operating an automatic accuracy extension function when the number of bits of the mantissa part of the other computer system is greater than that of the own computer system according to the result of the recognition; means for converting a constant of a program for another computer system into an internal value, and then setting the bits on the right side of the mantissa to 0 until the number of effective bits of the mantissa becomes the same as that of the other computer system; and after executing a floating point arithmetic instruction. The present invention is characterized by comprising means for inserting into the instruction sequence of the target program an instruction that sets the bits on the right side of the mantissa to 0 until the number of effective bits of the operation result becomes the same as that of the other computer system. Mantissa bit number matching method.