JPH03282827A - Support system for parallel conversion of program - Google Patents

Abstract

PURPOSE:To reduce the work of parallel conversion by automatically dividing the structure of a program operated by s single task which a user executes by the manual work. CONSTITUTION:This system consists of a program analyzing part 2, a block structure analysis list output part 3, a parallel conversion analyzing part 4, a source program update part 5, and a command input part 6. The user inputs a command from a terminal equipment 8 and inputs a source program 7 to generate a block structure file 9 as intermediate data and outputs a block structure analysis list to a SYSOUT file 10 or the terminal equipment 8 and finally outputs a parallel conversion source program 11. Thus, the time require to decompose all source programs for parallel conversion of a large-scale source program to block structures by the manual work is shortened.

Description

[Detailed description of the invention] [Industrial application field] The present invention relates to a program parallelization support system.

[Conventional technology]

Conventionally, in order to convert a user program that operates on a single task to one that operates on multiple tasks in parallel, the user manually decomposes the user's program into a block structure, and each decomposed block After analyzing the input/output data of the block, it is investigated whether it can be operated in parallel with other blocks, and if it is possible to operate in parallel, an instruction statement for the compiler is inserted into the corresponding block in the source program.

[Problem to be solved by the invention]

In the conventional method described above, when parallelizing a user program that runs on a single task, the user has to manually break down the program into a block structure and investigate whether it is possible to run it in parallel with other blocks. Therefore, the user must investigate whether a block references/updates data updated in another block. Therefore, if there is a lack of consideration or an incorrect decision, the execution result of a parallelized program may be different from that obtained when a single task is executed.

Additionally, the programs targeted for parallelization are often large-scale source programs, and it would take a lot of time to manually decompose all the source programs into block structures in order to parallelize them. There is a problem with that.

[Means to solve the problem]

The program parallelization support method of the present invention includes a source program analysis section that inputs a source program that operates in a single task, analyzes the contents of the source program, and outputs a block structure file; and a source program analysis section that inputs the block structure file. a block structure list output unit that outputs a block structure analysis list; and a block structure list output unit that inputs the block structure file to verify whether a specified block group can operate in parallel, and if parallelization is possible, creates parallelization information. a parallelization analysis section; a source program update section that inputs the parallelization information and the source program that operates on a single task to create a parallelized source program that operates on multiple tasks; and a source program update section that instructs operations from a terminal device. and a command input section for inputting commands.

〔Example〕

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

FIG. 1 is a block diagram showing an embodiment of the present invention. In the figure, the program parallelization support method 1 of the present invention is such that the user inputs commands from the terminal device 8, inputs the source program 7, creates a block structure file 9 as intermediate data, and creates a 5YSOUT file. 10 and the terminal device 8, and finally a parallelized source program 11 is output.

When the program parallelization support method 1 operates, the command input unit 6 prompts the command to the terminal device 8. From the terminal device 8, the user first instructs the storage location and program name of the source program 7 that runs in a single task. do. Thereafter, the command input section 6 transfers control to the program analysis section 2.

The program analysis section 2 inputs the source program 7, decomposes it into a block structure, outputs it to the block structure file 9, and returns control to the command input section 2. The block structure analysis list output section 3 is given control by the command input by the command input section 2 from the terminal device 8, inputs the block structure file 9, and edits and outputs the contents of the 5YSOUT file 10 and the block structure file 9 to the terminal device 8. do.

The user then refers to this block structure analysis list and instructs the command input unit 8 about the name of the block to be parallelized.

The parallelization analysis unit 4 receives control from a command person, Yamabe 8, according to instructions from the user. The parallelization analysis unit 4 is given the name of the block group to be parallelized from the command input unit 8,
Extracts the input/output data of the specified block from the block structure analysis file 9 and verifies whether it is possible to operate in parallel. If parallelization is not possible, determines which access changes are causing the inability to parallelize. The reason why parallelization is not possible, such as whether parallelization is possible, is output to the terminal device 8. If parallelization is possible, control is transferred to the source program update unit 5 along with information for parallelization.

Based on the information from the parallelization analysis unit 4, the source program update unit 5 inputs the source program 7, searches for the source line of the block to be parallelized, adds instructions to the compiler, and creates a parallelized source program. generate.

FIG. 2 is an explanatory diagram showing an example of a source program and a block structure file. In FIG. 2(a), the content of the source program consists of block A consisting of a series of simple processes, block B consisting of repeated processes, and block C consisting of a series of processes within block B. Then, the block structure analysis file shown in Figure 2(b) contains BLOCKOI, BLOCKO2, BLOCKO.
Three blocks of 3 are generated, block A is BLOC
KOl, block B to block 02.

Block C corresponds to BLOCKO3. Each block in the block structure analysis file stores the block name, processing details of the block, first and last line numbers in the block's source program, and names of input data and output data.

〔Effect of the invention〕

As described above, the present invention has the effect of reducing parallelization work by automatically dividing the structure of a program that operates in a single task, which conventionally had to be done manually by the user.

In addition, it is possible to verify the correctness of the parallelization work and find out the reason why parallelization is not possible, so it is possible to avoid cases where the user has made a mistake in parallelization and the execution result of the program is not parallelized. It is possible to prevent the occurrence of different problems depending on the situation.

[Brief explanation of drawings]

FIG. 1 is a configuration diagram showing one embodiment of the present invention, and FIG. 2 is an explanatory diagram showing an example of a source program and a block structure analysis file. 1... Program parallelization support method, 2... Program analysis section, 3... Block structure analysis list output section,
4... Parallelization analysis section, 5... Source program update section, 6... Command input section, 7... Source program, 9... Block structure analysis file, 11... Parallelization source program .

Claims (1)

[Claims] a source program analysis unit that inputs a source program that operates in a single task, analyzes the contents of the source program, and outputs a block structure file; and a block structure that inputs the block structure file and outputs a block structure analysis list. a list output section, a parallelization analysis section that inputs the block structure file, verifies whether or not a specified block group can operate in parallel, and creates parallelization information if parallelization is possible; A source program update section that inputs the source program that operates on a single task and creates a parallelized source program that operates on multiple tasks, and a command input section that inputs commands to instruct operations from a terminal device. A program parallelization support method characterized by: