JPH04308936A - Text optimizing system - Google Patents

Abstract

PURPOSE:To improve the application ratio of the optimization of the same value relation by rearranging an intermediate text at the time of optimizing the text. CONSTITUTION:This system is equipped with an input text block dividing means 1 which inputs the intermediate text for one basic block from an input text storing file 8, in-block text depending relation analyzing means 2 which analyzes a depending relation between the texts from the definition referring relation of the data in the text, text rearranging means 3 which rearranges the text so that an interval between the texts having the depending relation can be the minimum, and text outputting means 4 which outputs the rearranged text to an output text storing file 9. The application ratio of the optimization of the same value relation is improved in order to improve an object efficiency.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a text optimization method in a compiler system.

0002

[Prior Art] Conventionally, optimization methods in compiler systems mainly include optimization for intermediate text and optimization for machine language, and each has been performed in a wide variety of ways. Compared to each type of optimization, methods for further improving optimization efficiency are less common. "Optimization of equivalence relations" is a process that analyzes when generating machine language that refers to certain data, and analyzes that the values stored in registers when the same data was previously defined or referenced remain unchanged. Instead of loading the data from memory and using it, the register is used directly, but conventionally, whether or not equivalence relation optimization was applied was determined by the state at the time the data was referenced.

0003

[Problem to be Solved by the Invention] In the conventional optimization of equivalence relations as described above, when generating machine language for data reference, it is first analyzed whether the value of the data is stored in a register or not. So to speak, I was just waiting for a chance opportunity. Therefore, there is a drawback that the effect of optimizing equivalence relations is deteriorated due to the arrangement of intermediate texts.

0004

[Means for Solving the Problems] The text optimization method of the present invention inputs intermediate text within a range where control is transferred in a straight line from an input text storage file to a branch text/label text in a compiler system. ,
From the input text block dividing means that expands on memory and chains text in the input order using a text table, and the defined reference relationship of data in the expanded text,
An intra-block text dependency analysis means that analyzes text dependencies and expresses the dependencies and text output intervals using a text dependency table, and minimizes the sum of the output intervals between dependent texts within a block. The present invention includes a text rearranging means for rearranging the output order of texts, and a text output means for sequentially outputting the rearranged texts on the memory to an output intermediate text storage file.

[0005]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Next, the present invention will be explained in detail with reference to the drawings.

FIG. 1 is a diagram showing an embodiment of the present invention. Referring to FIG. 1, an embodiment of the text optimization method of the present invention includes an input text block dividing means 1, an intra-block text dependency analysis means 2, a text rearranging means 3,
It is composed of a text output means 4, an expanded text 5, a text table 6, a text dependency relationship table 7, an input text storage file 8, and an output text storage file 9.

The input text block dividing means 1 inputs text one text at a time from the input text storage file 8, and when a branch text is input, up to that text,
After inputting the label text, the text up to the immediately preceding text is expanded in memory (Developed text 5). At the same time, one text table 6 as shown in FIG. 2 is created for each text, and the addresses of the expanded text 5 in memory and the text for the text before and after the input order (ultimately the output order) are created. The address of table 6 and the value of the text number (text number) are set in advance. It also assumes virtual text at the end of the block,
On the other hand, a text table 6 is generated in the same manner as above (however, the expanded text address does not need to be set).

The intra-block text dependency relationship analysis means 2 analyzes text dependency relationships for the developed text 5 based on defined reference relationships of variable data appearing in the text. There are two types of dependencies between text T and previous text P:

■Data D accessed by text P,
Referenced by text T.

■Data D accessed by text P,
Text T defines. This dependency relationship is expressed by chaining the text tables 6 in a text dependency relationship table 7 as shown in FIG. The chaining method is as follows for each of the above dependence relationships.

■Using the text dependency table 7,
The text table 6 of text T and text P is chained. This chain consists of text P and text T
This is to make the output interval as small as possible. At this time, the dependency relationships are weighted according to the type of data D, and the weights are set in the dependency relationship table 7.

(Memory data) < (memory, non-temporary data) < (temporary data) At the same time, the difference in text numbers (text distance) between the two chained text tables 6 is also set in the text dependency table 7. put. (Hereafter, in two chained texts, the text that follows the previous text is called the "successful dependent text" and the text that precedes the text that follows is called the "preceding dependent text.") ■Text dependency relationship Text table 6 is chained from text P to text T using table 7. This chain only represents that text P must be output before text T. Therefore, there is no need to set the text distance as in ■. (Hereafter, in two chained texts, the text that follows the previous text is called the "required subsequent text" and the text that precedes the text that follows is called the "required preceding text.") The chaining process is , following the texts in the block in the output order, find the text T and the data D that the text T is accessing, the nearest text P before the text T that is accessing the text T.
It is done between. After the above processing has been completed for all the texts in the block, between the text table 6 for the last virtual text and the text table 6 for all the texts in the block that have no subsequent dependent or required text. , perform the chaining process described in ■■ above.

[0013] The text rearranging means 3 finally
The purpose is to rearrange texts so that the sum of text distances set in the text dependency table 7 is minimized. First, the text is traced in the reverse order of the output order chain from the text table 6 (text table 6 for the last virtual text) indicated by the text output order end address, and the following process is repeated while updating the target text T. . Process 3.1: Target text T is treated as processed. Process 3.2: ■ If there is one preceding dependent text P of target text T, ■ All subsequent required texts and subsequent dependent texts of text P have been processed, and ■ The text distance between text T and text P is 1. do nothing. Process 3.3: ■ There is one preceding dependent text P of target text T, ■ All subsequent required texts and subsequent dependent texts of text P have been processed, and ■ The text distance between text T and text P is not 1. For example, set the text number of text P to the text number of text T - 1, trace the output order text chain in reverse from text table 6 of text T, and set the text number of each table up to text table 6 immediately before text P by 1. countdown, and finally text T and the previous text T in the output order
' so that the text P is output between the text T,
The output order text chain of each text table 6 of T' and P is updated. Processing 3.4: ■ If there are two or more preceding dependent texts P of target text T, and ■ All subsequent required texts and subsequent dependent texts of text P have been processed, then text P, (1) Successor to text P If (2) and (1) are the same in descending order of the sum of weights of the dependent texts, set Pi (i=1, 2...) in order of descending sum of weights of preceding dependent texts of text P, and text T and text P
The same thing as Process 3.3 is performed so that the text distance of i becomes i. If (1) and (2) are the same, then (2)
) in ascending order until they are no longer the same,
Order Pi. Processing 3.5 ■ Dependent preceding text P of target text T (number is arbitrary)
If there is unprocessed text among the subsequent essential text and subsequent dependent text, the same process as in process 3.4 is performed only on the processed text of text P.

[0014] When all output order chains are followed and the above processing is performed, the sum of the text distances of the intra-block tables is minimized within the possible range.

The text output means 4 sequentially traces the text tables 6 from the text output order starting point pointer and sequentially outputs the expanded text 5 formed by the expanded text address of each table to the output text storage file 9.

FIG. 3 shows an example of the text rearrangement means 3 of source program→intermediate text→dependency diagram→optimized dependency diagram→optimized intermediate text according to the present text optimization method.

[0017]

As explained above, in the compiler system, when optimizing intermediate text, the definition reference relationship of data accessed by text is analyzed, and the intervals between definition-reference and reference-reference text for the same data are reduced. By minimizing, the application rate of optimization of equivalence relations during machine language generation is improved and object efficiency is improved.

[Brief explanation of the drawing]

FIG. 1 is a diagram showing an embodiment of the present invention;

FIG. 2 is a diagram showing data relationships in one embodiment of FIG. 1;

FIG. 3 is a diagram illustrating an example of text rearrangement means for source program→intermediate text→dependency diagram→after-optimization dependency diagram→after-optimization intermediate text as a result of applying the present invention.

[Explanation of symbols]

1...Input text block dividing means, 2...Intra-block text dependency analysis means 3...Text rearrangement means, 4...Text output means, 5...Development text, 6...Text table, 7...Text dependency relationship table, 8...Input text Storage file, 9...Output text storage file.

Claims (1)

[Claims] Claim 1: In a compiler system, intermediate text within a range where control is transferred in a straight line between branch text/label text is input from an input text storage file, expanded in memory, and written in the order of input using a text table. Analyze the text dependencies from the defined reference relationships of the data in the expanded text and the intra-block text dependencies that represent the dependencies and output intervals of the text using a text dependency table. analysis means,
A text sorting means that rearranges the output order of texts so that the sum of output intervals between dependent texts in a block is minimized, and texts that have been sorted in memory are sequentially output to an output intermediate text storage file. A text optimization method comprising a text output means for outputting text.