JPH03127125A - Automatic change system for multiple branch processing by learning - Google Patents

Abstract

PURPOSE:To improve the processing speed by changing automatically to an optimum processing sequence at that time point in accordance with a variation of an operation environment of a program. CONSTITUTION:In all comparison processings for deciding 'equal' or 'unequal' in a branch processing of a program, at the time point of jumping to the processing of each branch destination of all branch conditions except two branch conditions of a branch condition under which a comparison processing executed in the beginning is formed and a branch condition under which constants of branch conditions of all deciding objects are not formed, the processing for changing the comparison processing before it is concerned and the concerned comparison processing, and a processing jump destination pointer after the branch condition before by one before it is concerned is formed and a processing jump destination pointer after the branch condition concerned is formed are changed. That is, in the case data to be decided which goes into a multiple branch processing by an operation environment of the program is a prescribed value with high probability, the value of its high probability is decided preferentially. In such a way, a total dynamic step can always be reduced.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to a multiple branch processing method for comparing constant values in a programming language, and in particular, a method for changing a processing sequence every time a program performs processing, This article relates to a processing method for improving performance.

[Conventional technology]

As described in Japanese Unexamined Patent Application Publication No. 63-221427, the conventional technology optimizes the size value judgment when calculating constant values to be compared in multiple branch processing that compares constant values, and eliminates unnecessary objects. A method is being considered to abolish it.

However, the prior art does not have a method for changing the sequence of constant value comparisons in multiple branch processing by learning the program itself.

[Problem to be solved by the invention]

In the prior art, no consideration is given to optimizing the processing sequence using multiple branch processing that compares constant values. As shown in the example in Figure 1, the multiple branching process involves more or less comparison processing for each constant depending on the data to be determined.

Among the values of the data to be determined, it is desirable to perform less comparison processing if there is a constant value with a high probability, and to perform more comparison processing if it is a constant value with low probability, from the viewpoint of processing moderation.

The present invention improves the performance of multiple branch processing by allowing the program itself to modify the processing sequence of comparison processing little by little as the processing is repeated.

An object of the present invention is to reduce the total step amount in multiple branch processing.

[Means to solve the problem]

FIG. 1 is a basic processing diagram of the present invention, showing the pre-learning C initial state and initial processing. The branching process is excluded from the application target force of the present invention, and the judgment mark of the comparison process that branches to each process is based on the data to be determined X and the constant value at.
The present invention is applied to multi-branch processing in which by ct d is limited to comparison processing using constant values such as = (equal) or ≠ (not equal).

FIG. 1 shows, as an example, a multi-branch process in which the process is branched into five processes depending on the value of X of the data to be determined.

In this state, it is assumed that the data to be determined X enters this multiple branching process with a value of C. Since a
Comparison of :a and X:b1. In comparison 2, the judgment is ≠,
Comparison 3 of x:c results in a determination of =.

From this comparison 3, the process moves to process 8. In process 8, the value of the constant value C to be determined is exchanged, and the jump destination pointer d after the branch processing in comparison 3 and the jump destination pointer Pa after the branch processing in comparison 2 are exchanged.

In process 9, a jump is made to the pointer Pa exchanged in process 8. Pointer Pa is the pointer p before being exchanged in process 8
Since the value is c, the process moves to C processing corresponding to C.

At this stage, the processes of Comparison 2 and Comparison 3 have been swapped, and the processing sequence has been changed. Again, the data to be determined X is C
If the value is , then after comparing ≠ in comparison 1 of

The process moves to process 6, where the values of a and C (old value) are exchanged,
Jump pointer Pc (old Pa value) after branch processing in comparison 2 and jump pointer P after branch processing in comparison 1
Since the pointer P^ is the pointer Pc (old PR value) before the exchange in process 6, C
The process will then proceed to C processing corresponding to .

At this stage, the processes of Comparison 1 and Comparison 2 have been swapped, and the processing sequence has been changed.

[Effect]

FIG. 2 shows changes in learning in multiple branch processing. State 1 in FIG. 2 shows the processing sequence in the initial state shown in FIG. It becomes !2. If the value of x=c is entered again, the processing sequence of the comparison process is changed again and becomes state 3. Once again, x=c
Since the x:c comparison process is the first process, the processing sequence is not changed and state 3. State 4 remains the same. In this way, it can be seen from FIG. 2 that the order of comparison processing of
Among the values of , constant values with high probability require less comparison processing each time they are processed. Further, for a constant value with a high probability, the number of comparison processes increases each time the process is performed.

If the data to be determined that enters multiple branch processing in the operating environment of the program has a constant value with a high probability, the multiple branch processing of the present invention attempts to give priority to the determination of the value with that high probability.

The order of comparison processing of the values becomes earlier in the processing sequence, and the comparison processing at the beginning becomes the last one.

Further, even if the operating environment of the program changes at some point during the program and the data to be determined that enters the multi-branch processing with a high probability changes, the multi-branch processing of the present invention will respond to the change in the operating environment and The processing sequence is changed again in an attempt to prioritize the value.

Therefore, the multi-branch processing of the present invention changes the processing sequence by itself in accordance with changes in the operating environment, and can always reduce the total dynamic steps.

〔Example〕

The present invention can be implemented by directly changing the text portion executed by the program, or by converting comparison constants for judgment and process jump destination pointers into array data without changing the program execution portion, and converting the comparison constants and process jump destination pointers into array data. One possible method is to change the . In this text, the latter will be explained as an example.

3 and 4 show an embodiment of the present invention, A, B, and C of the processing jump destination pointer 26 corresponding to the array number O''max shown in the array 24 of FIG. 4.

..., E, Z processing destination pointers are assumed. Therefore, it becomes a multi-branch process that branches into (wax + 1) processes.

Now, assume that the data to be determined X has a value of d. At this time, it is compared with the fourth array C3 (=d) of the comparison constant 25 in FIG. (=D
) Jump to the pointer and branch. This process will now be explained using FIG.

When executing the process 13 in FIG. 3 with the data to be determined as d, X and Co are compared and since Go is a, the determination is ≠, and the process goes to process 14. In process 14, a counter n is initialized to 1. Process 15 is X to the empty area of Cm a x.
Assign the value (=d).

In process 16, C1 is compared with C1, and since C1 is b, the determination is KI, and the process goes to process 17. Process 17 is counter n
Add +1 to Here, a loop process returns to process 16.

When n counts up to a value of 3, since the value of CS is d, X and C8 match in process 16, and a determination of = is made, and the process goes to process 19.

Since n has a value of 3, the determination is ≠. In processes 20 and 21, the comparison constant 25 and the fourth and third array data of the process jump destination pointer 26 in FIG. 4 are exchanged.

Therefore, Ca (=d), Cz (=c), Pa (=D)
and Pg (=C) are exchanged. After this, in process 22, Pg
Jump to pointer (=D).

In this state, after the determined data X is processed with the value of d twice in a row, Co becomes d.

PG is converted to the value of D. In this state, again.

When the data to be determined X is processed with the value of d, in process 13, Co
is d, so it is judged as =, and in process 18 Pa(=D)
Jump to the pointer.

Above, the processing when the determined data X is the value d has been described, but the processing is similar when the data X is the value of other at be ct . . . e.

The data to be determined X is at by C, ci,”・t e
If it does not match any value of
&X match, resulting in a determination of =. After this, in process 19
Since is the maximum, it is judged as =, and the process goes to step 23, where P
Jump to wax pointer.

According to this embodiment, when the data to be determined is constant with a high probability, the dynamic step amount is gradually reduced in order to set the processing sequence to prioritize the determination of the value with the high probability.

Furthermore, even if the value of data with a high probability changes, the processing sequence changes to give priority to the determination of that value. In this way, even if the value of the target data with high probability changes, the processing sequence is automatically corrected to a constant processing speed by changing the processing sequence, so there is no variation in performance, and furthermore, it is possible to manually correct the program processing sequence. This also has the effect of making it unnecessary.

〔Effect of the invention〕

Since the present invention is configured with the processing sequence described above, it produces the effects described below.

If the data to be determined that enters multiple branch processing in the program operating environment has a constant value with a high probability, the total dynamic step amount of the program is It shrinks each time it is processed.

Further, even if the operating environment of the program changes and the value of the determined data that has a high probability of entering multiple branch processing changes, the program itself changes the processing sequence in accordance with the value. In this way, in response to changes in the operating environment of the program, the processing sequence is automatically changed to the optimum one at that time, thereby increasing the processing speed.

[Brief explanation of the drawing]

FIG. 1 is a flowchart of the program execution unit according to an embodiment of the present invention, FIG. 2 is an initial state and process explanation diagram of multiple branch learning processing, FIG. 3 is a change diagram of learning multiple branch processing, and FIG. 4 is a diagram showing changes in learning of multiple branch processing. 3 shows a data array section of the present invention. 1.2, 3.4... Comparison process between the data to be determined and constant value, 6, 8.10... Exchange process of constant value and jump destination pointer after branch processing, 5, 7, 9, 11゜Figure

Claims (1)

[Claims] 1. In the branch processing of the program, there are three or more processing destinations to branch to, and the condition for branching is a comparison process that determines whether the compared data and the constant to be determined are "equal" or "not equal." In addition, in program processing where the comparison process is multi-branched and consists of two or more, the comparison process is limited to cases where "equal" holds true, and "equal" or "not equal"
Of all the comparison processes that perform this judgment, all branches except for two branch conditions: the branch condition for which the first comparison process was satisfied, the constant for all judgment targets, and the branch condition for which the branch condition was not satisfied. At the time of jumping to the process of each branch destination of the condition, the process that replaced the previous comparison process with the corresponding comparison process, the process after the previous branch condition was met, the jump destination pointer, and the corresponding branch condition being met. An automatic change method for multiple branch processing using learning, characterized in that a process for exchanging a pointer to a subsequent processing jump destination is provided.