JPH0625965B2 - Rounding specification method - Google Patents

Description

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to type conversion in a system that performs numerical operations and control of rounding that occurs during operations with different data types.

[Conventional technology]

As the types of rounding in the numerical operation, for example, I _EEE has four types of rounding modes (toward + as, toward-as, toward 0, ne.
arest) is specified. In the conventional numerical operation system, by designating one of these four types of rounding, one type of rounding is executed regardless of the type of play. For example, according to reference (1), the rounding type can be selected by the RC (Round-Control) field of the control word, and if the user specifies one, that type of rounding will be used unless the specification is changed. Rounding is applied and performed for every operation.

Reference (1) "IAPX86 Family User's Manual"
Chapter 3 Intel Japan (1981) [Problems to be Solved by the Invention] In the above-described conventional rounding designation method, once the rounding type is designated, only one type is not rounded. Therefore C,
By executing an assignment statement, as in the FORTRAN language. In cases where automatic type conversion and rounding associated with it are often performed, and there are several types of rounding methods that correspond to the type of type conversion, each time a different type conversion occurs if only one type is specified. There is a drawback in that the rounding type must be re-specified and overhead occurs.

For example, the type conversion in C language will be described using the program example of FIG. 3 and the operation result. Program example
line10,17,24,31 is type conversion from real number to integer, line11,18,25,32 is type conversion from double precision of real number to single precision, line2,5,8,11 of operation result is real number to integer Result of type conversion to line3,6,9,1
Reference numeral 2 is the calculation result of the type conversion from the double precision of the real number to the single precision. As shown in these figures, the rounding when converting from a real number to an integer is rounded down, and the rounding when converting from a double precision of a real number to single precision is done by rounding off.

Therefore, when another rounding mode is specified during another operation, the rounding type must be specified again each time a type conversion occurs.

[Means for solving problems]

The rounding designation method of the present invention includes a first designating means for designating a rounding type, an identifying means for identifying a specific command, a second designating means for designating a rounding type according to an instruction of the identifying means, In the case where there is no instruction by the instructing means, there is an instructing means for instructing selection of either the first or the second instructing means, and a rounding means for performing rounding in accordance with the instruction of the instructing means.
When the rounding is performed according to the designation of the first designating means, and there is an instruction by the designating means, and the specific instruction is identified by the identifying means of the instruction, the designation of the second designating means is given priority, In this case, the rounding is performed, and in other cases, the rounding is performed by giving priority to the designation of the first designating unit.

〔Example〕

 Next, the present invention will be described with reference to the drawings.

FIG. 1 is a schematic configuration diagram of an embodiment of the claims of the present invention,
FIG. 2 is an explanatory diagram of the priority circuit, and FIG. 4 is a correspondence table according to instructions and rounding types.

In FIG. 1, 1 is a 4-bit flip-flop, 2 is an instruction decoder, 3 is a PLA, 4 is a 1-bit flip-flop, 5 is a priority circuit, and 6 is a rounding execution circuit.

The information 1 is set by a special write command of the CPU. Each bit corresponds to its rounding mode, 4
Set the value so that only one of the bits has the value 1. When the value of 4 is 0, this value is output to 6 according to the priority circuit of 5. When the value of 4 is 1, the specific instruction is decoded by 2, and when the corresponding one of the signal lines input to the PLA of 3 becomes 1, it is programmed by the PLA according to Table 1. Output line becomes 1 and 5
It is output from 3 to 6 according to the priority circuit of. 5 is a priority circuit, and when the output of 4 is 0, the output from 1 is given priority, and 6
When one of the outputs of 3 is 1, when the output of 4 is 1, the output is prioritized, otherwise the output from 1 is given, and the output is supplied to 6.

Next, a specific example of the priority circuit 5 will be described with reference to FIG. In FIG. 2, 11 to 16 are AND circuits, 17 to 19
Is an OR circuit, 20 is a NAND circuit, 21 to 24 are output signals from 1, 25 and 26 are output signals from 3, 27 is a 1-bit flip-flop 28 to 31 is an output signal to 6. For example, suppose the signal 24 from 1 was 1.

At this time, if the signal from 27 is 0, the signal from 1 is 24
Is prioritized and 30 is output to 6.

Next, it is assumed that the signal 24 from 1 is 1 and the signal of 27 is 1. At this time, when the signal 25 from 3 is 1, the signal from 3 is prioritized and 28 is output to 6. When the signals 25 and 26 from 3 are both 0, the signal from 1 is prioritized and 30 is output to 6.

As described above, either the signal designated by 1 or the signal designated by 3 is given priority by the output from 4, and rounding is performed.

〔The invention's effect〕

As described above, the present invention identifies an instruction in addition to the first specifying means for specifying the rounding type. By providing the second designating means for designating the type of rounding by the identifying means, the designating means for designating either the first designating means or the second designating means, and executing the rounding according to the instruction of the designating means, For a specific instruction whose rounding type is predetermined, it is not necessary to re-specify the rounding each time the instruction changes, and there is an effect that overhead does not occur.

[Brief description of drawings]

FIG. 1 is a schematic configuration diagram of an embodiment of the present invention, FIG. 2 is an explanatory diagram of a priority circuit, FIG. 3 is a program example and a diagram showing a calculation result, and FIG. 4 is a correspondence between an instruction and a rounding type. FIG. 1 ... 4-bit flip-flop, 2 ... instruction decoder, 3 ... PLA, 4 ... 1-bit flip-flop, 5 ... priority circuit, 6 ... rounding circuit, 11-16 ...
AND circuit, 17-19 ... OR circuit, 20 ... NAND circuit, 21-24 ... Output signal from 1 ... 25-26 ...
3 output signal, 27 1-bit flip-flop, 28 to 31 output signal to 6.

Claims (1)

[Claims] 1. In a system that handles a plurality of data types and executes a numerical operation, a first designating unit that designates a type of rounding, an identifying unit that identifies a specific instruction, and a rounding by an instruction of the identifying unit. And a rounding means for performing rounding in accordance with the instruction of the instructing means, the instruction means for instructing selection of either the first or the second instructing means. When there is no instruction by the means, the rounding means is activated according to the designation of the first designating means to perform rounding, and there is an instruction by the instructing means, and the specific instruction is identified by the identifying means. In this case, the rounding designation is performed by giving priority to the designation by the second designating means and by executing the rounding in the other cases by giving priority to the designation by the first designating means. method.