JPH0439703B2 - - Google Patents

Description

[Detailed Description of the Invention] [Summary] Based on the source program and its execution analysis information,
A processing method that generates control lines for optimization and automatically creates a new source program with the optimization control lines inserted, and a processing method that collects and displays vectorization information for a specific program unit by inputting a command. be.

[Industrial Application Field] The present invention relates to a source program creation processing method, and particularly to optimization processing using program execution analysis information.

When data processing is performed using a vector processor, the created source program is required to be optimized to use the vector processor most effectively and to shorten processing time.

[Prior art] When a program is actually run with practical data input, how many repetitions will be executed by the DO statement, and what is the probability that the IF statement will be true? Execution analysis information is estimated by analyzing the above.

Currently, at the stage of compiling a source program, it is possible to use this execution analysis information to optimize the vector processor so that it operates most efficiently.

Therefore, control information including execution analysis information is inserted into a predetermined location of the source program, and the program is compiled based on this information.

Currently, there is no means to automatically generate and automatically insert control lines for this purpose through batch processing.
Therefore, there was a problem in that the control line had to be inserted manually.

Furthermore, it has not been possible to obtain vectorization information for converting a source program into one suitable for processing by a vector processor by simply specifying a command or the like for each specific program unit (subroutine).

[Problems to be Solved by the Invention] The present invention aims to provide a novel source program optimization processing method that solves these conventional problems.

[Means for Solving the Problems] FIG. 1 shows a principle block diagram of the original program optimization processing method of the present invention.

In Figure 1, 1 is the DO from the source program.
This is a repetition count extraction means that extracts a statement and extracts the number of repetitions of a DO statement from execution analysis information.

Reference numeral 2 denotes a control line generation means 1 that generates an optimization control line in which the number of repetitions of the DO statement is set and inserts it into the new source program data set.

3 is a true rate extraction means that extracts the IF statement from the source program and extracts the true rate of the IF statement from the execution analysis information, and 4 generates an optimization control line in which the true rate of the IF statement is set, and generates the new source program. This is a control line generating means 2 to be inserted into the data set.

By inputting a source program and its execution analysis information and sequentially executing the above means 1, 2, 3, and 4, a new source program with optimal control lines inserted at predetermined locations is generated.

Reference numeral 5 denotes a partitioned data set creation means for creating and displaying a partitioned data set having program unit names as members from the execution analysis information of the source program.

Reference numeral 6 denotes vectorization designation information creation means for creating a data set from which unnecessary parts are deleted based on the analysis information of the designated program unit name, and outputting the data set to the vector compiler.

Reference numeral 7 denotes an analysis information display means for editing and displaying vectorization information output from the vector compiler using data from the vectorization designation information creation means 6.

By means of the means 5, 6, and 7, it is possible to specify vectorization information for each routine from the execution analysis information of the source program and display the analysis information.

[Operation] With the above configuration, a new source program with optimization control lines inserted at predetermined locations can be automatically generated from the source program and execution analysis data, and vectorization information can be generated by specifying a desired routine. It becomes possible to display.

[Example] The present invention will be described in more detail below with reference to Examples shown in FIGS. 2 to 4.

FIG. 2 is a block diagram showing an example of processing for inserting optimization control lines according to an embodiment of the present invention.

In FIG. 2, A indicates an input source program, B indicates execution analysis information of the source program, C indicates processing according to the present invention, and D indicates a newly generated source program.

The input source program example has the following meaning.

REAL〓8TF1 (500), TF2 (500), ... is a table TF1 with a maximum of 500 real 8-byte numbers.
and TF2 array declaration. The same applies to TF3 and TF4.

DO 10 I=1,N 10 A(I)=I means execute statement number 10, that is, A(I)=I from 1 to N in I, and This means that you should enter the value of .

DO 20 J = 1, M IF (J.GT.II) = THEN TF1 (J) = DSIN (TF2 (J)) =... ENDIF 20 CONTINUE is from J 1 to M, if J is greater than 11,
This means that you should repeatedly execute TF1 (J) = DSIN (TF2 (J) =...). DSIN is a double precision SIN.

DO 80 J=1,M IF(TF1(J).GT.0.0E0)THEN TF3(J)=DSIN(TF2(J))+… 80 CONTINUE is from 1 of J to M, if TF1(J ) is greater than 0.0 (single precision), it means that TF3 (J) = DSIN (TF2 (J)) + ... should be executed repeatedly.

B is an example of the result of giving the execution analysis processing system a source program and practical data and having it perform an execution analysis.

This execution analysis information includes the number of repeated executions corresponding to STATEMENT of the source program in the EXEC column, and the TRUE column of the source program.
The true rate corresponding to STATEMENT is listed.

C shows the process of generating and inserting optimization control lines using the above two data sets.

FIG. 3 is a flowchart showing optimization control row insertion processing according to an embodiment of the present invention.

While copying the input source program to a new source program data set, the following processing is performed depending on the type of executable statement of the source program.

The record retrieved from the source program is
Determine whether the statement is a DO statement and the number of repetitions is unknown, and if so, proceed to step.

The number of executions of the DO statement and the number of executions of the executable statement that immediately follows the DO statement are extracted from the program execution analysis data set, the number of executions is divided by the number of times the DO statement is executed, and the value of the quotient is the number of repetitions (loop (number of times).

Generates an optimization control line that sets the number of repetitions of the DO statement and inserts it into the new source program data set.

The record retrieved from the source program is
Determine if it is an IF statement, and if so, proceed to.

Exists in the program execution analysis dataset
Extract the true rate of the IF statement (values below the decimal point are rounded down).

Generates an optimization control line that sets the true rate of the IF statement and inserts it into the new source program data set.

D in FIG. 2 shows the source program newly generated in this way.

The automatically generated and inserted optimization control lines are written as VOCL LOOP REPEAT (150) for the number of repetitions, VOCL STMT.IF (50) for the true rate, and are placed next to the executable statement. inserted into.

VOCL stands for Vectorize Optimize Control Line
Although it is defined as an abbreviation, it can be in any form as long as it can be interpreted by the compiler.

FIG. 4 is a flowchart illustrating a process for collecting and displaying vectorized information according to an embodiment of the present invention.

In FIG. 4, B is the same execution analysis information as shown in FIG.

E is a means (means 5) for creating and displaying a partitioned dataset whose members are program unit names from the execution analysis information of the source program.

F is the source program data set (sequential organization)
and the segmented data set compiled from this.

G is an analysis information display screen for specifying vectorization.

In G, V-COST indicates the relative processing time when processed by a vector processor, and S-COST
COST indicates the relative processing time when processed by the scalar processor, and V-RATE indicates the vectorization rate for each routine. V-
COST and V-RATE are "0" because vectorization has not been specified yet.

H indicates a screen where a routine from which vectorization information is desired is specified by the command VCHECK.

I creates a dataset by removing unnecessary parts from the analysis information of the specified program unit name,
This is means (means 6) for returning the data to sequential organization and outputting it to a vector (VP) compiler.

J indicates the data set created by means 6.

K is VP based on the data created by means 6.
A means (means 7) for editing and displaying vectorization information output from a compiler is shown.

L is an analysis information display screen after vectorization information edited by means 7 is collected.

As shown in L, routine name SUB001 and
Analysis data by the VP compiler is displayed in SUB003 for the relative processing time V-COST and vectorization rate V-RATE when processed by the vector processor.

[Effects of the Invention] As described above, according to the present invention, optimization for processing by a vector processor can be automatically performed, and the effect of contributing to improvement of processing efficiency is significant.

[Brief explanation of drawings]

Fig. 1 is a block diagram of the principle of the present invention, Fig. 2 is a block diagram showing an example of optimization control line insertion processing according to an embodiment of the invention, and Fig. 3 is an optimization control line insertion processing according to an embodiment of the invention. FIG. 4 is a flowchart illustrating the process of collecting and displaying vectorized information according to an embodiment of the present invention. In the figure, 1 is a repetition number extraction means, 2 is a control line generation means 1, 3 is a true rate extraction means, 4 is a control line generation means 2, 5 is a segmented data set generation means, and 6 is a vectorization specification information generation means. , 7 indicate analysis information display means, respectively.

Claims (1)

[Claims] 1. A first means (1) for extracting a DO statement from a source program and extracting the number of repetitions of the DO statement from execution analysis information, and generating an optimization control line in which the number of repetitions of the DO statement is set. and the second means of inserting it into the source program (2)
A third means (3) of extracting the IF statement from the source program and extracting the true rate of the IF statement from the execution analysis information, and generating a control line that sets the true rate of the IF statement and inserting it into the source program. A source program optimization processing method, comprising the fourth means (4), and configured to generate a source program into which optimization control information is inserted from the source program and execution analysis information. 2. Fifth means (5) of creating and displaying a partitioned data set whose members are program unit names from the execution analysis information of the source program, and a data set in which unnecessary parts are deleted from the analysis information of the specified program unit name. and a seventh means (7) for editing and displaying vectorization information of the vector compiler, and a seventh means (7) for editing and displaying vectorization information of the vector compiler. A method for optimizing a source program, characterized in that it is configured to perform collection and display.