JP2004029982A - Parallelization conversion system, parallelization conversion method, program, and compiler - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To discriminate a parallelization directive sentence described by a user from a parallelization directive sentence automatically generated by a processing system at the time of converting a sequential program into a parallel program. <P>SOLUTION: In the parallelization compiler 11, a directive sentence generating part 16 generates the result of automatic parallelization analysis performed by a parallelization analyzing part 13 as a parallelization directive sentence in a source file 18. In this case, a unitary character string("hash value") obtained by transmitting the character string of the parallelization directive sentence through uni-directional functions("hash functions") is added as a comment so as to be discriminated from the parallelization directive sentence generated/changed/inserted by a user. <P>COPYRIGHT: (C)2004,JPO

Description

[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a computer program conversion technique, and more particularly to a computer processing technique for converting a program written to perform processing sequentially into a program that performs parallel processing automatically or according to a programmer's instruction. It is.
[0002]
[Prior art]
Conventionally, as a technique for creating a program for performing parallel processing, a programmer inserts a directive for instructing parallelization into a program written to perform processing sequentially, and the computer uses a parallelizing compiler to issue the instruction using a parallelizing compiler. There is a program that analyzes a sentence and converts it into a program that performs parallel processing.
[0003]
For example, in "OpenMP Fortran Application Program Interface Version 2.0, November 2000, http://www.openmp.org/", a programmer loop is defined, which defines OpenMP which is a parallel language for a shared memory type parallel computer. Instruction is inserted into the program, and the compiler performs parallel conversion according to the instruction.
[0004]
However, it is burdensome for a programmer to issue all the parallelization instructions. As a technique for solving this problem, there is a technique in which a computer automatically analyzes the parallelism of a program using a compiler or a parallelization support tool, and generates a result as a parallelization directive in the program.
[0005]
A conventional technique for automatically generating a parallelization directive in a program by such a computer is described in, for example, "Haoqiang Jin, Michael Frumkin, Jerry Yan", "Automatic Generation of OpenMP Direction Appointments and Recommendations. 1940, Issue, pp 440-Lecture Notes in Computer Science.
[0006]
The parallelization directive automatically generated by the computer in this way needs to be a versatile directive such as, for example, an OpenMP directive in order to ensure the portability of the program.
[0007]
However, the prior art has the following problems.
[0008]
That is, a source program to which a directive has been once added by the processing system (computer) may be corrected by the programmer and input to the processing system again. At this time, if the programmer corrects the executable statement, the instruction sent by the processing system in the past may become incorrect. The processing system needs to regenerate such directives so as to match the program modified by the programmer.
[0009]
On the other hand, similar parallelization directives may be manually added by a programmer. Since the processing system must follow the instructions of the programmer, it is necessary to keep such directives unchanged.
[0010]
Here, in order to maintain portability, when the processing system adds a versatile directive such as an OpenMP directive, for example, it is determined whether the directive described in the program is the one described by the programmer. The problem arises that it is not possible to determine whether the system has been automatically added before.
[0011]
For this reason, the processing system cannot correct the instruction sentence generated in the past by the processing system. Alternatively, there is a possibility that the instruction sentence added by the programmer is also modified.
[0012]
[Problems to be solved by the invention]
The problem to be solved is that, in the conventional technique, the instruction to instruct parallelization in a program for performing parallel processing is written by a programmer or automatically added by a processing system (computer) before. Is not determined by the processing system.
[0013]
SUMMARY OF THE INVENTION An object of the present invention is to solve the problems of the prior art, and to improve efficiency of automatic conversion by a computer of a source program written to perform sequential processing into an object program that performs parallel processing. It is to be.
[0014]
[Means for Solving the Problems]
In order to achieve the above object, according to the present invention, when a parallel processing system (computer) converts an input program described so as to perform processing sequentially into a program that performs processing in parallel, it performs automatic parallelization analysis. The parallelization directive obtained as a result is generated in the original source program, and a unique character string obtained by passing the character string of the parallelization directive through a one-way function is used as a comment statement. Generated on the same line as the parallelization directive, and when the program is input again and parallelized, a unique character string obtained by passing the character string of the parallelization directive through the same one-way function as above, By checking the character string of the parallelization directive described in the program with a unique character string obtained by passing the parallelization directive through a one-way function, the directive was automatically inserted by the processing system. What is And to detect that the user has not changed. Then, after deleting the directive and the character string which are inserted by the processing system and not changed by the user from the parallel processing program, the parallel processing program is re-written including the parallelization directive and the automatic generation of the character string. Perform parallel processing.
[0015]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings.
[0016]
FIG. 1 is a block diagram showing a configuration example of a computer device that executes a compiler provided with a parallelization system according to the present invention. FIG. 2 is a block diagram showing a parallelization process according to the present invention at the time of compiling operation by the computer device of FIG. 3 is an explanatory diagram showing a configuration example of a parallel processing program generated by the computer device in FIG. 1, FIG. 4 is a flowchart showing a processing operation example of the intermediate language generation unit in FIG. 1, FIG. 5 is a flowchart illustrating an example of a processing operation of the directive generating unit in FIG. 1.
[0017]
1 includes a CPU (Central Processing Unit), a main memory, a display device, an input device, an external storage device, and the like. For example, programs and data recorded on a CD-ROM via an optical disk drive are stored in an external storage device. After being installed in the apparatus, the main memory is read from the external storage device and processed by the CPU, thereby executing each function as the parallelizing compiler 11 having the function as the parallelizing system according to the present invention.
[0018]
In FIG. 1, reference numeral 10 denotes an input program (source file) described so as to perform processing sequentially (described as "input source" in the figure), 11 denotes a parallelizing compiler, 17 denotes a parallelizing compiler 11 A generated object program for parallel processing in the form of a computer-executable code (described as “object code” in the figure), and a source file 18 for the object program 17 for parallel processing (described as “output source” in the figure) .
[0019]
The parallelizing compiler 11 is generated by the intermediate language generation unit 12, the parallelization analysis unit 13, the parallelization conversion unit 14, the object code generation unit 15, the directive generation unit 16, and the intermediate language generation unit 12, and Intermediate language 19 that is referred to and updated by the conversion analysis unit 13, the parallelization conversion unit 14, the object code generation unit 15, and the directive sentence generation unit 16, and information that is generated and referred to at the time of compile processing in each of the processing units 12 to 16. Is stored in the information table 20.
[0020]
The intermediate language generation unit 12 reads the input program 10 and performs lexical analysis, syntax analysis, semantic analysis, and the like to generate an intermediate language 19. The intermediate language 19 also includes information on a parallel instruction by a directive.
[0021]
The parallelization analysis unit 13 refers to the intermediate language 19 and the various information tables 20 generated and registered by the intermediate language generation unit 12 to analyze the control structure of the program and the reference relation of the data, and to determine the loop that can be parallelized. The communication required for the determination and the parallelization is determined, and the intermediate language 19 and the various information tables 20 are updated.
[0022]
The parallelization conversion unit 14 performs conversion of the control structure of the program, generation of a communication message, and the like based on the result of the parallelization analysis by the parallelization analysis unit 13 with reference to the intermediate language 19 and the various information tables 20. The word 19 and the various information tables 20 are updated.
[0023]
After optimizing the intermediate language 19, the target code generation unit 15 outputs the parallel processing object program 17.
[0024]
Then, the directive generation unit 16 reads the input program 10, generates and inserts a parallelization directive into the input program 10 based on the parallelization information obtained by the parallelization analysis unit 13. The character string of the directive is passed through a one-way function to generate a unique (unrivaled) character string, and this character string is inserted as a comment on the same line as the parallelization directive and output as the source file 18 I do.
[0025]
With such a configuration, the computer device performs the parallel processing operation according to the present invention by the directive generating unit 16 as shown in FIG.
[0026]
That is, the instruction sentence generation unit 16 reads the input program 10 (step 201), generates a parallelization instruction sentence based on the parallelization information obtained by the parallelization analysis unit 13, and inserts it into the input program 10. At the same time (step 202), a unique (unrivaled) character string is generated by passing the character string of the parallelization directive through a one-way function, and the character string is placed on the same line as the parallelization directive. It is inserted as a comment (Step 203) and output as the source file 18 (Step 204).
[0027]
FIG. 3 shows an example of the source file 18 output from the instruction generating unit 16 in such a procedure.
[0028]
Since the parallelizing compiler 11 knows that the “DO j loop” can be parallelized in the program shown in FIG. 3, the directive generating unit 16 indicates the PARALLEL DO directive 31 of “OpenMP” (“! OMP PARALLEL DO PRIVATE (I, J) ") is generated and inserted.
[0029]
Furthermore, the directive generation unit 16 generates a unique character string 32 (here, the hash value “012345789 fedcba”) obtained by passing the character string of the parallelization directive through the one-way function on the same line as the inserted PARALLEL DO directive 31. )) Is inserted in the form of a comment (“! HASH012234789feedcba”).
[0030]
By adding the character string 32 in this manner, when compiling the source file 18, the directive generation unit 16 performs parallel processing using the same hash function as when the previous parallelization directive was inserted. The parallelizing compiler 11 automatically calculates a hash value from the character face of the conversion instruction sentence, and compares and compares the hash value with the hash value in the source file 18 in, for example, the intermediate language generation unit 12. It is possible to automatically detect and check whether the user has changed the parallelized instruction sentence.
[0031]
An appropriate prefix is added to the comment describing the hash value so that the user does not confuse the comment with the user description or accidentally delete the comment. In the example shown in FIG. 3, a character string “HASH” indicating that the hash value is inserted by the compiler is added after the comment symbol (“!”). The hash function used at this time is selected so that the value range is sufficiently large and the generation of the same hash value for different character faces does not practically occur.
[0032]
By doing so, when the user wants to reinsert the directive into the output source, only the parallelization directive automatically generated by the parallelizing compiler 11 is updated, and the parallelization directive inserted by the user is updated. , And the parallelization directive automatically generated by the parallelizing compiler 11 and changed by the user can be stored without any modification.
[0033]
Next, among the processing operations according to the present invention when the parallelization directive and the character string are generated by the directive generation unit 16 and the source file 18 inserted into the input program is reconverted, the above-described intermediate language generation The process of reading the parallelization instruction sentence by the unit 12 will be described with reference to FIG.
[0034]
After reading the parallelization directive in the processing of step 401, the intermediate language generation unit 12 of FIG. 1 performs processing in step 402 depending on whether a directive generation option is specified for the parallelization directive. Divide.
[0035]
If the directive generation option is not specified, the process moves to step 405 to compile according to all the parallelization directives, and unconditionally executes the parallelization directive information as in the related art. Is registered in the various information tables 20.
[0036]
If the directive generation option is specified, the parallelization directive automatically generated by the parallelizing compiler 11 itself is re-analyzed and re-inserted, and the parallelization directive issued by the user (operator, programmer). To read only the data, the processing of steps 403 and 304 is performed.
[0037]
That is, in step 403, the hash value obtained from the character face of the parallelization instruction sentence is compared with the hash value embedded as a comment in the source file 18 to check whether or not they match. If the hash value is not embedded in the source file 18 as a comment, it is determined that the hash values do not match.
[0038]
If it is determined in step 404 that the hash values match and the parallelization directive is automatically generated by the parallelizing compiler 11 itself, the information of the directive is discarded without being registered in the table (step 406). If the hash values do not match, the instruction is regarded as a user instruction, the processing in step 405 is executed, and the information on the parallel instruction is fetched into the various information tables 20.
[0039]
Next, the processing operation of the directive generating unit 16 in FIG. 1 will be described with reference to FIG.
[0040]
The instruction generating unit 16 in FIG. 1 first checks whether or not an instruction generating option is specified in the input program 10 in the processing in step 501, and terminates without doing anything if it is not specified.
[0041]
If the directive generation option is specified, all the parallelization directives in the input program 10 are deleted (step 502). Thereafter, the processes in steps 503 to 506 are repeated, and the parallelization instruction sentence for the input program 10 is generated and inserted based on the information (various information tables 20) of the parallelization analysis result by the parallelization analysis unit 13 in FIG. .
[0042]
That is, whether or not the parallelization directive sentence generated by the intermediate language generation unit 12 of FIG. 1 and registered in the various information tables 20 by the processing of FIG. 4 is a parallelization directive sentence automatically generated by the parallelization compiler 11 itself. (Step 503), and for a parallelization directive automatically generated by the parallelizing compiler 11 itself, a hash value is generated from the face of the parallelization directive and the hash value is generated on the same line as the parallelization directive. In addition, for the parallelization directive sent by the user (step 504), only the parallelization directive is inserted into the input program 10 (step 505). (Step 506).
[0043]
If the parallelization directive is already specified for the statement or effective area targeted by the parallelization directive, the parallelization directive is not inserted and the existing directive is saved. . It is possible to regenerate the instruction while retaining the user instruction.
[0044]
As described above with reference to FIGS. 1 to 5, in the parallelizing compiler 11 of the present example, the computer device converts an input program described to perform processing sequentially into a program that performs processing in parallel. At the same time, the parallelization directive obtained as a result of the automatic parallelization analysis is generated in the original source program, and the character string of the parallelization directive is passed through a one-way function (hash function). The generated unique character string (hash value) is generated and inserted as a comment statement on the same line as the parallelization instruction statement.
[0045]
With this comment statement, the user can easily specify the parallelization directive sentence automatically generated by the compiler in the source file of the program converted for parallel processing.
[0046]
Furthermore, in this example, when the program converted for parallel processing is input again and parallelized, a unique character string obtained by passing the character string of the parallelization directive through the same one-way function as above By checking the character string of the parallelization directive described in the program with a unique character string obtained by passing the character string of the parallelization directive through a one-way function, the directive is processed by a processing system (computer). It can be automatically detected that the data has been automatically inserted and has not been changed by the user.
[0047]
Then, after deleting a directive and a character string that the processing system (computer) has inserted and the user has not changed from the parallel processing program, automatic generation of a parallelization directive and a character string for this parallel processing program is performed. Is performed.
[0048]
As described above, when the source program (the source file 18) containing the parallelization directive automatically generated by the parallelization compiler 11 itself is recompiled, the parallelization directive written by the programmer and the parallelization compiler 11 itself are automatically compiled. By distinguishing the generated parallelization directive from the generated parallelization directive and erasing and regenerating only the parallelization directive automatically generated by the parallelizing compiler 11, the parallel processing program can be efficiently re-converted.
[0049]
The present invention is not limited to the examples described with reference to FIGS. 1 to 5 and can be variously modified without departing from the gist thereof. For example, in this example, the parallelization conversion system of the present invention is described as a function of the parallelizing compiler 11, but the present invention is realized as a tool for inputting a sequential program and outputting a source containing a parallelization directive. You may.
[0050]
Further, in this example, OpenMP which is a parallelization instruction specification for a shared memory type parallel computer is used as a directive statement language. However, the present invention may be implemented using a parallelization language for a distributed memory type computer. It is possible. For example, it is possible to input a sequential program, perform parallel analysis for a distributed memory type parallel computer, and insert the result into the input source as a parallelization directive of HPF which is a parallel language for the distributed memory type computer. it can.
[0051]
Further, in this example, a parallel processing program in which the character string (hash value) generated by the directive generation unit 16 is inserted as a comment on the same line as the parallelization directive is output as a source file 18 as a separate file. However, the original input program 10 may be replaced.
[0052]
Further, the computer may be configured without a keyboard or an optical disk drive. In this example, the optical disk is used as the recording medium, but an FD (Flexible Disk) or the like may be used as the recording medium. As for the installation of the program, the program may be downloaded and installed via a network via a communication device.
[0053]
【The invention's effect】
According to the present invention, whether a directive (parallelization directive) instructing parallelization in a program for performing parallel processing is written by a programmer or automatically added before by a processing system (computer). It is possible for the processing system to automatically determine whether or not there is, and it is possible to improve the efficiency of automatic conversion by a computer from a source program written to perform processing sequentially to an object program that performs processing in parallel. It is possible.
[Brief description of the drawings]
FIG. 1 is a block diagram illustrating a configuration example of a computer device that executes a compiler provided with a parallelization system according to the present invention.
FIG. 2 is a flowchart illustrating an example of a parallel processing operation according to the present invention when a compiling operation is performed by the computer device of FIG. 1;
FIG. 3 is an explanatory diagram illustrating a configuration example of a parallel processing program generated by a computer device in FIG. 1;
FIG. 4 is a flowchart illustrating an example of a processing operation of an intermediate language generation unit in FIG. 1;
FIG. 5 is a flowchart illustrating an example of a processing operation of a directive generation unit in FIG. 1;
[Explanation of symbols]
10: Input program ("input source"), 11: Parallelizing compiler, 12: Intermediate language generating unit, 13: Parallelizing analyzing unit, 14: Parallelizing converting unit, 15: Object code generating unit, 16: Directive statement generating Part, 17: object program (“object code”), 18: source file (“output source”), 19: intermediate language, 20: various information tables, 31: PARALLEL DO directive, 32: character string (“hash value )).

Claims (7)

First means for analyzing the parallelism of an input program written so as to perform processing sequentially, generating a parallelization directive, and inserting the generated instruction into the input program, and processing the input program in parallel A parallel conversion system for converting into a parallel processing program,
A second means for generating a unique character string from the parallelization directive generated by the first means, and inserting the character string together with the parallelization directive;
If the character string in the input parallel processing program is matched with the character string generated by the second means from the parallelization instruction sent with the character string inserted, if the character strings match, the parallelization instruction Is generated by the first means and is determined to have not been changed by the user. The parallelization conversion system according to claim 1, wherein
The second means generates the character string by passing the character string of the parallelization directive through a one-way function, and inserts the character string as a comment on the same line as the parallelization directive. Characterized parallel conversion system. 3. The parallelization conversion system according to claim 1, wherein a source file of a program in which the parallelization directive and the character string are inserted in the input program is stored in a storage device. A parallel conversion system comprising: A program for causing a computer to function as each unit in the parallelization conversion system according to claim 1. A first procedure of analyzing the parallelism of an input program described to be sequentially processed, generating a parallelization directive, and inserting it into the input program, and performing the input program in parallel A parallelization conversion method for a system for converting into a parallel processing program,
A second procedure of generating a unique character string from the parallelization directive generated in the first procedure and inserting the same with the parallelization directive;
A third procedure of comparing the input character string in the parallel processing program with a character string generated by the second means from the parallelization instruction sentence in which the character string is inserted;
If the two character strings match as a result of the collation, a fourth procedure for determining that the parallelization instruction sentence has been generated in the first procedure and has not been changed by the user;
A fifth procedure for deleting the character string and the specific directive from the parallel processing program,
The first and second procedures are executed on the parallel processing program from which the character string and the specific directive have been deleted in the fifth procedure, and the parallel processing program is re-converted. Parallelization conversion method. The parallelization conversion method according to claim 5, wherein
The second procedure includes generating the character string by passing the character string of the parallelization directive through a one-way function, and inserting the character string as a comment on the same line as the parallelization directive. Characteristic parallelization conversion method. A compiler for converting a source program written to perform sequential processing to a computer into an object program that performs parallel processing,
In the above computer,
A first procedure for analyzing the parallelism of the source program and generating and inserting a parallelization directive;
A second procedure of generating a unique character string from the parallelization directive and inserting it together with the parallelization directive;
A third procedure of storing a source file of the parallel processing program into which the parallel processing instruction and the character string are inserted, in a storage device;
If the character string input in the parallel processing program for reconversion matches the character string generated by the second procedure from the parallelization instruction sentence in which the character string is inserted, the parallelization instruction A fourth procedure for determining that the sentence is a specific instruction sentence generated by the first procedure and not changed by the user;
A fifth procedure for deleting the character string and the specific directive from the parallel processing program;
A compiler which executes a compile procedure including the first to third procedures for the parallel processing program from which the character string and the specific directive have been deleted.

Images