JPH07200308A - Optimum instruction string selection execution system - Google Patents

Abstract

PURPOSE:To output an object so that an object with a highest efficiency is most frequently executed by outputting plural objects whose execution efficiency is unknown. CONSTITUTION:The system is made up of an optimization section 1 comprising a plural-instruction generation enable intermediate code detection means 11 detecting an intermediate code able to generate plural objects whose execution efficiency is not known without execution and an intermediate code conversion means 12 executing plural objects respectively with respect to the detected intermediate code, recording arm object with a highest execution efficiency and executing only the object with highest efficiency after that. Thus, when plural objects whose execution efficiently is not known beforehand are generated, the objects with the highest efficiency are executed as many as possible.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an optimum instruction sequence selecting and executing system for generating intermediate code in a language processing system of an electronic computer program language.

[0002]

2. Description of the Related Art For some parts of a source program,
If a plurality of different instruction sequences can be generated by a plurality of algorithms and it is not possible to determine in advance which instruction sequence will have the shortest execution time, refer to Japanese Patent Laid-Open No. 63-198130 for the applicable technique. In this publication, a plurality of different instruction sequences are generated, and the value of a parameter that seems to influence the performance of each instruction sequence is examined at the time of execution. For example, it may be more efficient if the parameter is smaller than a certain value or smaller. A technique for generating an object that selects and executes one of a plurality of instruction sequences, such as executing one instruction sequence, is shown.

[0003]

In the above-mentioned prior art,
If the judgment based on the parameters cannot be set properly, the optimum instruction sequence cannot always be executed.

An object of the present invention is to measure the execution speeds of all instruction sequences at the time of execution and reflect the results even if the determination by parameters cannot be set well.
An object of the present invention is to provide an optimum instruction string selection execution system capable of executing an optimum instruction string.

[0005]

An optimum instruction sequence selection execution system of the present invention is a compiler for generating a target program from a given source program and supplying the target program to a computer system. It includes a parsing unit that interprets and expands into intermediate code, an optimizing unit that performs optimization for effectively using the processor at the level of the intermediate code, and a target program generating unit, and further, the optimizing unit. Can convert a series of intermediate code in a loop corresponding to a part of the source program into another intermediate code that obtains the same calculation result, and the target program with any intermediate code can calculate the result quickly. It is possible to detect and record a series of intermediate code that cannot be judged without executing the target program of that part. A plurality of instruction string generateable intermediate code detecting means and a series of intermediate codes detected by the plurality of instruction string generateable intermediate code detecting means are converted into intermediate codes that obtain a plurality of the same calculation results, and each of them is converted. When it finds the intermediate code that executes sequentially and finishes the fastest, intermediate information that records that information, and the subsequent execution of that part converts the fastest ending intermediate code into the intermediate code that executes It is characterized in that it comprises a conversion means.

[0006]

An embodiment of the FORTRAN compiler of the present invention will be described in detail with reference to the drawings.

Referring to FIG. 2, which shows the structure of the entire compiler to which the present invention is applied, the syntax analysis unit 22 uses FORTR.
The source program 21, which is an AN program, is converted into the intermediate code 13. The optimizing unit 1 inputs the converted intermediate code 13, optimizes it, and converts it into an intermediate code 14. The code generation unit 23 generates the object code 24 by the converted intermediate code 14. The present invention relates to the optimizing unit 1 and increases the execution efficiency of the object code 24. FIG. 1 shows the configuration of a portion related to the present invention in FIG.

Referring to FIG. 1, an optimizing section 1 which is a characteristic part of an embodiment of the present invention comprises an intermediate code detecting means 11 and an intermediate code converting means 12 capable of generating plural instruction sequences.
including.

Referring to FIG. 1, an intermediate code detecting means 11 capable of generating a plurality of instruction sequences converts a series of intermediate codes in a loop corresponding to a part of a source program into another intermediate code which obtains the same calculation result. It is possible to detect a series of intermediate codes which cannot be judged without executing the target program of that part.

The intermediate code converting means 12 adds the intermediate code which obtains the same calculation result to the series of intermediate codes detected by the intermediate code detecting means 11 capable of generating a plurality of instruction sequences, executes them, and executes the fastest one of them. When the intermediate code that terminates execution is found, that information is recorded, and the subsequent execution of that portion is a means for converting only the intermediate code that terminates execution fastest into intermediate code.

The intermediate codes 13 and 14 are intermediate codes in the process of being converted from the source program 21 shown in FIG.

In the following description, the source code detected by the intermediate code detection means 11 capable of generating a plurality of instruction strings
A series of intermediate code in a loop corresponding to a part of the program can be converted into another intermediate code that obtains the same calculation result, and which part of the intermediate code can calculate the result faster A series of intermediate codes that cannot be determined without executing the target program is called intermediate code capable of generating multiple instruction sequences.

The operation of one embodiment of the present invention will be described with reference to FIGS. 3, 4 and 5.

Referring to FIG. 3, the intermediate code 31 is an example of the intermediate code 13 input to the intermediate code converting means 12 and is an intermediate code for sorting the first dimension of a two-dimensional array along the second dimension. is there. The execution time measuring block 4 in FIG. 4 and the optimum intermediate code execution block 5 in FIG. 5 are examples of the intermediate code 14 that is output as a result of the intermediate code converting means 12 inputting the intermediate code 31.

In this example, the multi-instruction string-generating intermediate code detecting means 11 in FIG. 1 can convert the sort function in the loop as a multi-instruction string-generating intermediate code into quick sort, bubble sort, and heap sort. Shall be detected. Since the optimum algorithm for sorting can be changed depending on the number of elements and the arrangement of the elements, it is not always possible to select the optimum algorithm by simply judging the number of elements as a parameter.

Referring to FIG. 3, the intermediate code 3 is the intermediate code 31 capable of generating a plurality of instructions and the repeat control intermediate code 3.
2, the optimization unit 1 uses the conventional optimization technique to repeatedly control the intermediate code 32 to the intermediate code 3.
Is identified as a loop, and the plural instruction string generateable intermediate code 31 is detected as the plural instruction string generateable intermediate code by the plural instruction string generateable intermediate code detecting means 11.

The intermediate code converting means 12 uses the intermediate code 3
The execution time measurement block 4 shown in FIG. 4 for collecting the execution time data for inputting the intermediate code 31 capable of generating a plurality of instruction sequences based on the record detected in 1, and executing this execution code Based on the data measured by the time measurement block 4, the optimum intermediate code execution block 5 shown in FIG.

Referring to FIG. 4, the execution time measuring block 4 includes an execution time measuring intermediate code 41, a converted intermediate code 42, and a repeat control intermediate code 43.

The execution time measurement intermediate code measures the execution time of each changed intermediate code 42 and stores it in a variable generated by the compiler.

The post-conversion intermediate code 42 is each post-conversion intermediate code that can be generated from the multi-instruction sequence-generating intermediate code 31.

The repeat control intermediate code 43 is an intermediate code excluding the forward branch that repeats a series of processes from the repeat control intermediate code 32 shown in FIG. 3 that guarantees the number of times each converted intermediate code 42 is executed. The reason why the forward branch is excluded is to measure the execution time of all the converted intermediate codes 42.

The execution time measuring block 4 measures the execution time of each post-conversion intermediate code.

Referring to FIG. 5, the optimum intermediate code execution block 5 comprises an optimum judgment intermediate code 51, a converted intermediate code 52, and a repeat control intermediate code 53.

The optimum determination intermediate code 51 is based on the execution time measured by the execution time measuring block 4 shown in FIG. 4, which intermediate code of each converted intermediate code 52 performs all remaining iterations. Is an intermediate code for determining.

The post-conversion intermediate code 52 is the same as the post-conversion intermediate code 42 shown in FIG. 4, and is each post-conversion intermediate code that can be generated from the multi-instruction string generation intermediate code 31.

The repeat control intermediate code 53 is a copied intermediate code of the repeat control intermediate code 32 shown in FIG. 3 which guarantees the number of times of execution of the converted intermediate code 52. Since the optimum post-conversion code 52 can be selected, in order to execute all the rest of the number of iterations with the optimum post-conversion intermediate code 52, a forward branch is included so as to form a loop. Different from code 43.

By converting the intermediate code in this way, even when the selection of the intermediate code by the optimum algorithm cannot be divided by the parameter or the like, it becomes possible to generate the object code which can be executed at the highest speed.

[0028]

According to the present invention, a plurality of different instruction sequences are generated for a certain portion of the source program, the execution time of each of them is measured, and the result is recorded, so that the subsequent portion of that portion can be recorded. At the time of execution, it is possible to use only the instruction sequence that is excellent as a result, and it is possible to reduce the execution time of the program.

[Brief description of drawings]

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

FIG. 2 is a diagram showing a configuration example of a compiler including the present invention.

FIG. 3 is a diagram showing an example of a FORTRAN program to which the present invention is applied.

4 is a diagram showing a first half portion of an intermediate code obtained as a result of applying the present invention to the FORTRAN program shown in FIG.

5 is a diagram showing a second half portion of an intermediate code obtained as a result of applying the present invention to the FORTRAN program shown in FIG.

[Explanation of symbols]

 1 Optimizer 11 Intermediate Code Detecting Means Which Can Generate Multiple Instruction Sequences 12 Intermediate Code Converting Means 13 and 14 Intermediate Code 21 Source Program 22 Parser 23 Code Generator 24 Object Code 3 Source Program 4 Execution Time Measurement Block 41 Execution time measurement intermediate code 42 Converted intermediate code 43 Repeat control intermediate code 5 Optimal intermediate code execution block 51 Optimal determination intermediate code 52 Converted intermediate code 53 Repeat control intermediate code

Claims (1)

[Claims] 1. An optimum instruction sequence selecting and executing system for a compiler, which generates and supplies an object program from a given source program to an electronic computer system, wherein the source program is interpreted to form an intermediate code. It has a syntactic analysis unit that expands to, an optimization unit that performs optimization for effectively using the processor at the intermediate code level, and a target program generation unit, and the optimization unit is a part of the source program. It is possible to convert a series of intermediate code in the corresponding loop to another intermediate code that obtains the same calculation result, and which intermediate code is the target program that can calculate the result quickly. An intermediate code that can generate multiple instructions that detects and records a series of intermediate codes that cannot be determined without executing the program. The detecting means and the series of intermediate codes detected by the multi-instruction sequence-generating intermediate code detecting means are converted into an intermediate code that obtains a plurality of the same calculation results, and each converted intermediate code is executed. When the intermediate code that finishes execution fastest is found, information that is found is recorded, and the execution of that part after the time of finding is the intermediate code conversion means that converts the intermediate code that finishes execution fastest to the intermediate code that executes. An optimum instruction sequence selection execution system characterized by including.