JP2751977B2 - Dynamic analysis information prediction calculation method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

The present invention relates to a method for predicting and calculating dynamic analysis information of a program created in a high-level programming language.

[0002]

2. Description of the Related Art In dynamic characteristic analysis, a program is actually run to collect execution information of a place where a user wants to measure. Based on this information, the user can select a program unit or DO loop for which the program is to be accelerated,
It is used to confirm that the speedup is as expected.

[0003]

In such a conventional system, in order to detect a portion of a program requiring a long execution time, a program unit or a D (Digital Unit) is required at the time of dynamic characteristic analysis.
It is necessary to collect information for each measurement point such as an O loop. In such a case, if the measurement point is in the repeated processing, the measurement is performed many times, the overhead time for collecting information increases, and a huge execution time must be devoted to obtain the dynamic analysis information. The problem that it does not become occurs.

Further, in order to reduce the execution time required for dynamic analysis, the number of input data in a program is reduced,
In a matrix operation or the like, when the data of a program is reduced, such as by reducing the number of dimensions of an array, the execution time obtained by the data reduction and the execution time ratio in the entire program are larger than when the program is executed with the original data. May be different. In such a case, there is a problem in that even if speeding up a portion requiring an execution time is performed in accordance with the reduced dynamic analysis execution information, the speed is not increased with the original data.

SUMMARY OF THE INVENTION The present invention has been made to solve the above problems, and automatically predicts and calculates the dynamic characteristics of a program that changes by changing variable data in the program. An object of the present invention is to provide a method for predicting and calculating dynamic analysis information that can reduce the execution time at the time of analysis.

[0006]

According to a first aspect of the present invention, there is provided a method for predicting and calculating dynamic analysis information, comprising the steps of: If there is measurement location information indicating a program unit or a DO loop for which execution information is to be collected, the measurement location information is input, and the location and control constituting the loop structure for each program unit are moved to a statement other than immediately after. Detects control branch points and measurement points, and determines the loop structure, including the start and end points of the loop, the type of loop, the end condition of the loop, and the data used to control the loop and whether the control data value is fixed or variable. If the information or value is variable, collect information about the statement or variable that defines the value, and for the control branch point, the branch location, branch type, branch condition, At the control branch point where the control is transferred, information on the name of the program unit to be called and information on the call location are collected, and information on the measurement location and structure, such as the measurement location is in a loop structure. After the processing of each program unit is completed, the calling relationship of each program unit is analyzed based on the above information, and each loop structure changes the program execution conditions such as the control data value and the number of input data Then, it is determined whether or not the execution status is a variable loop that may change. If the static analysis information includes measurement location information, the measurement is performed in the variable loop and the execution information such as the execution time and the number of loops is affected. A program analysis process that searches for locations and outputs the collected information and the detected variable loops and static analysis information on the measurement locations affected by them. Means, input the static analysis information output by the program analysis processing means, display the characteristic analysis information for each variable loop affected by the same execution conditions, and if there is information on the measurement location, information on the variable loop Information display means for displaying characteristic analysis information as information relating measurement points affected by the measurement.

According to a second aspect of the present invention, there is provided a method for predicting and calculating dynamic analysis information according to the first aspect, wherein the static analysis information obtained by the program analysis processing means and the source are calculated. If a program is entered and information on the measurement location is specified, it is entered, and if a variable loop is detected in the source program, a measurement statement that collects the execution information of the variable loop is inserted and the measurement location is detected. Then, source program generation processing means for generating a measurement program by inserting a measurement statement for collecting execution information of the measurement location into the source program, and a measurement program generated by the source program generation processing means, translation·
Execution information collection processing means including processing for calculating execution information such as the number of loops and execution time by combining and executing, and changing the number of input data of the source program input by the execution information collection processing means, the declaration of an array, and the like. Input change information on execution conditions and their assumed values, static analysis information created by program analysis processing means, and execution information calculated by execution information collection processing means, and the effect of execution conditions that are changed based on program calling relationships Is detected, the number of executions of the loop is searched from the execution information, the execution information of the assumption value is calculated from the ratio to the assumption value specified in the change information, and if there is a measurement point in the loop, the The execution information of the assumed value at the measurement point is calculated, and this process is repeated for all the affected loops. Based on the execution information of the assumed value, the entire program or the program is executed. Each ram unit, is characterized in that comprises an execution information prediction processing means for calculating a predicted execution information of each loop.

According to a third aspect of the present invention, there is provided the dynamic analysis information predictive calculation method according to the second aspect, wherein the source program generation processing means inputs the source program, and executes the measurement If information is specified, enter it, and if change information is specified regarding the execution conditions such as the number of input data to be changed and the When a location specified by the change information is detected in the source program, a measurement statement that collects execution information of the variable loop is inserted, and when a measurement location is detected, a measurement statement that collects execution information of the measurement location is inserted. It is characterized in that a measurement program is generated by inserting it into a source program.

[0009]

In the method for predicting and calculating dynamic analysis information according to the first aspect, the program analysis processing means receives a source program and obtains measurement unit information indicating a program unit or a DO loop for collecting execution information. For example, input measurement point information and detect the points that make up the loop structure for each program unit, control branch points where control moves to statements other than immediately after, and measurement points. For loop structures, start and end points of the loop , Loop type, loop end condition,
If the data controlling the loop and the value of the control data is fixed or variable, and the value is variable,
Collects information about statements and variables that define the value, and for the control branch point, information about the branch location, branch type, branch condition, control destination when branching, and the control branch where the program unit call is performed At the point, information on the name of the program unit to be called and the call location are collected, and information on the measurement location and structure such as the measurement location within the loop structure is collected.After the processing of each program unit is completed, the above information is used. Analyze the calling relationship in program units, and determine whether each loop structure is a variable loop whose execution status may change if the execution condition of the program such as the value of control data or the number of input data is changed, If there is measurement location information in the static analysis information, search for measurement locations in the variable loop that are affected by execution information such as execution time and loop count. Collected information and the detected variable loop and outputs a static analysis information about the measurement locations affected. The information display means inputs the static analysis information output from the program analysis processing means, displays the characteristic analysis information for each variable loop affected by the same execution condition, and changes the information if there is information on the measurement location. The characteristic analysis information is displayed as information relating the information on the loop and the measurement location affected by the loop.

Further, in the method for predicting and calculating dynamic analysis information according to the second aspect, the static analysis information and the source program obtained by the program analysis processing means are input by the source program generation processing means and the source program is generated. If the information is specified, enter it, and if a variable loop is detected in the source program, insert a measurement statement that collects the execution information of the variable loop. The measurement statement to be collected is inserted into the source program to generate a measurement program. The execution information collection processing means inputs the measurement program generated by the source program generation processing means, and translates, combines, and executes the measurement program to calculate execution information such as the number of loops and execution time. Further, the execution information prediction processing means changes the execution data for changing the number of input data of the source program input by the execution information collection processing means, the declaration of the array, and the like, and the change information on the assumed values, Analysis information, execution information calculated by the execution information collection processing means is input, a loop affected by the execution condition to be changed based on the calling relationship of the program is detected, and the number of executions of the loop is searched from the execution information, The execution information of the assumption value is calculated from the ratio to the assumption value specified in the change information, and if there is a measurement point in the loop, the execution information of the assumption value of the measurement point is calculated. This process is repeated, and the predicted execution information for the entire program, for each program unit, and for each loop is calculated based on the execution information of the assumed values.

Further, in the method for predicting and calculating dynamic analysis information according to claim 3, the source program generation processing means inputs a source program and, if information on a measurement location is specified, inputs the information. If change information on the execution conditions such as the number of input data or array to be changed in the source program and the location affected by it is specified, enter it and specify the location specified in the source program with the change information. Is detected, a measurement statement for collecting execution information of the variable loop is inserted. When a measurement point is detected, a measurement statement for collecting execution information of the measurement point is inserted into the source program to generate a measurement program.

[0012]

Next, an embodiment of the present invention will be described with reference to the drawings. FIG. 1 is a block diagram showing an embodiment including the first and second aspects. The characteristic analysis mechanism 10 includes a source program 1, measurement location information 3 for collecting execution information (program units, DO loops, etc.), and program change information 4 (REA
Input number of D statements, number of dimensions of array, etc.) and execute information 1
4a and the predicted execution information 5 of the program calculated from the static analysis information 11a. The program analysis processing unit 11, the information display unit 12, the source program generation processing unit 13, the execution information collection processing unit 14, and the execution And information prediction processing means 15.

The program analysis processing means 11 receives the source program 1 and outputs static characteristic information 11a which is information on a loop structure, a control branch point, a program calling relation, a variable loop, and the like for each program unit.
If the measurement point information 3 is specified, the measurement point information is also input, and information on the measurement point such as whether or not the measurement point belongs to the variable loop is output in accordance with the static analysis information 11a. The information display means 12 inputs and processes the static characteristic information 11a and displays them as characteristic analysis information 2 on a terminal or the like.

The source program generation processing means 13 inputs the measurement location information 3 (such as measuring execution information for each program unit) specified by the user, the source program 1, and the static characteristic information 11a. 1 to search for a measurement point designated by the measurement point information 3 and insert a measurement sentence for measuring the execution information. Also, for a variable loop (a loop whose execution time changes by modifying a program) stored in the static characteristic information 11a, all the program units in which measurement statements are inserted in order to collect execution information of the loop are collected. The same process is performed to generate the measurement program 13a.

The execution information collection processing means 14 inputs the measurement program 13a, performs translation, combination, and execution,
It outputs execution information 14a for each measurement location. Further, the execution information prediction processing unit 15 receives the execution information 14a, the static analysis information 11a, and the program change information 4, and calculates the amount of execution per variable loop as (total amount of execution / number of loop executions). .

That is, in accordance with the calling relationship of the program, the loops specified by the program change information input by the user are searched in order from the deeply nested loops. Execution information 1
The execution information in the case of the assumed value is calculated from the ratio with the actually measured value obtained from 4a. After performing this process for all loops according to the calling relationship, if necessary,
The prediction execution information 5 is calculated for each measurement point, such as the entire program, each program unit, and each loop.

FIG. 2 shows an embodiment in which the source program generation processing means 13 of FIG. 1 corresponds to claim 3.

In the source program generation processing means 13, in claim 2, all the determination loops determined to be variable loops in the static analysis information 11a are to be measured as execution information. If the user sets the change information 4 instead of inputting the information 11a, the specified portion (variable loop) is set as the measurement target. Therefore, there is an advantage that the number of measurement points for collecting execution information can be reduced as compared with the second aspect.

However, when the program change information 4 is not input, the location to be measured in the measurement program is the location specified by the measurement location information 3 (or the location to be sampled by default). In this case, if there is a measurement location where the execution information changes under the influence of the loop specified in the change information 4 input by the execution information prediction processing means 15, the prediction execution information 5 of the measurement location is calculated. If the measurement location includes the variable loop specified by the change information 4, but the variable loop has no execution information (not measured), the calculation formula set by the characteristic analysis mechanism 10 Alternatively, the execution information of the variable loop can be calculated according to the calculation formula defined by the user, and subsequently, the prediction execution information 5 of the measurement location can be predicted.

FIG. 3 is an example of a source program described in the FORTRAN language.

The program unit 100 includes a variable loop and sections 101 and 102. Section 101 is a double loop ending at statement number 10, and is a variable loop whose number of loops is controlled by a constant name N. Also,
Section 102 is a variable loop that controls the number of loops based on the number of input records of the READ statement. The section 102 includes a CALL statement that calls the program unit 200, and passes the constant name N as a parameter.

The program unit 110 includes the variable loop 11
Contains one. The section 111 is a variable loop in which the number of loops is controlled by the formal argument M. Here, it is assumed that the measurement of all CALL sentences is specified in the measurement location information 3.

The program analysis processing means 11 shown in FIG.
When the source program 100 is input, the loop outputs static analysis information 200 shown in FIG. 4 such as information on loops, whether or not each of them is a variable loop, branch information, measurement points, and the like. In the case of the program unit 110, the static analysis information 2 shown in FIG.
10 are created. When the static analysis information is generated for all the program units, a variable loop affected by the same execution condition change is searched based on the information based on the calling relation of the program units.

In this example, the sections 101 and 111 are
It can be seen that the variable loop is affected by the change of the constant name N defined in the source program 100.
In addition, it can be seen that the section 102 is a variable loop that is affected by a change in the number of input records. Further, when the search is performed based on the calling relationship in the unit of a program, it can be seen that the measurement location 103 and the measurement location 104 include the section 101 and are included in the section 102. These pieces of information are generated as static analysis information.

FIG. 5 is a display example of characteristic analysis information created based on the static analysis information 11a output by the program analysis processing means 11.

Information on the variable loop and the measurement location affected by the constant N and information on the variable loop and the measurement location affected by the number of input records are displayed.

FIG. 6 shows an example of execution information collected by the execution information collection processing means 14. In the execution information 300, the execution time (the time from the start to the end of the own measurement point, excluding the execution time of other measurement points included from the start to the end of the own measurement point) for the measurement location and the variable loop. Information and execution count information are collected. In this example, information is collected using a double loop as one measurement point.

The processing of the execution information prediction processing means 15 will be described using two information examples, change information 400 and change information 401 shown in FIG. Execution information prediction processing means 15
The user inputs the static analysis information 200 and 210, the execution information 300, and the change information 400. Change information 400
In this example, the number of input records is assumed to be 200. When the execution number information of the section 102 in which the number of input records is used as the number of loops is input from the execution information 300, it is found that the number of input records is 20.

Therefore, the measurement locations affected by the number of input records and the execution information of the variable loop are all ten times (200/200)
20). First, in the nesting relationship between the measurement point in the section 102 and the variable loop, the execution information of the section 111, which is the deepest nest, is calculated.
04, execution information is calculated in the order of the section 102. In the case of this change condition, since only the section 102 is a direct variable loop, the measurement points, the variable loop,
By updating the execution information of the measurement points including the section 102,
The prediction execution information 500 shown in FIG. 8 can be calculated.

By using the prediction execution information 500, information such as information for each program unit can be re-edited. In the case of the change information 401, the constant name N is changed from 10 to 100. In this case, the same processing as in the case of the change information 400 is performed, and the prediction execution information 501 is created. Actual execution information 400, prediction change information 500, prediction change information 501
Are compared, the execution information 400 indicates that M
The AIN (the part excluding the section 102) is a place where the execution ratio decreases as the number of input data and the array dimension to be processed increase. If the absolute value of the execution time is not large, the AIN is a target for speeding up. No need to do. Rather, it is clear that increasing the speed of the section 111 increases the effect when the execution conditions such as the number of input data are expanded.

[0031]

According to the first aspect of the present invention, a user can reduce the execution time during dynamic analysis by displaying information on a variable loop and its control variable name from static characteristic information of a program. There is an advantage that a changed part of the program (the number of array dimensions, etc.) can be easily detected. Further, when execution information has already been collected, it is possible to detect a portion where the execution information changes due to the number of dimensions of the array in the program or the number of input data.

According to a second aspect of the present invention, the first aspect includes inputting execution information to be changed from a portion affected by a change in an execution condition based on static information and dynamic information such as a variable loop. Then, the prediction execution information is automatically calculated for each change information. Therefore, once static analysis and dynamic analysis are performed, execution information can be automatically calculated for each specified execution condition without re-analysis every time various execution conditions in the program are changed. is there. Also, in a program or the like that handles large-scale data, it is possible to predict execution information when actual data is used based on information on dynamic analysis performed by reducing the number of dimensions of an array or reducing input data. Therefore, there is an advantage that the execution time during dynamic analysis can be reduced.

Further, according to the third aspect of the present invention, when the user designates a place to change the number of loops and the like and performs dynamic analysis using the place as a measurement place, variable loops other than the designated place are measured. Since there is no location, there is an advantage that the overhead time for collecting information at the time of dynamic analysis is reduced and the execution time is reduced. Another advantage is that a loop that is not recognized as a variable loop by the program analysis processing means can be specified.

[0034] Further, when claim 1 and claim 3 are specified so that the user can change the information, the change information is interactively specified while displaying the static analysis information. The operation for setting the change information is facilitated. Also,
Even when a dynamic analysis is performed by designating only a measured location without designating a location to be changed, there is an advantage that execution information that can be predicted within the range of the information is calculated.

[Brief description of the drawings]

FIG. 1 is a configuration diagram of an embodiment of the present invention including claims 1 and 2;

FIG. 2 is a configuration diagram of an embodiment of the present invention including claims 1 and 3;

FIG. 3 is an explanatory diagram showing an example of a source program.

FIG. 4 is an explanatory diagram of a storage example of static analysis information created by the program analysis processing means 11 when the source program shown in FIG. 3 is input.

FIG. 5 is an explanatory diagram of a display example of characteristic analysis information output by an information display means 12;

FIG. 6 is an explanatory diagram of execution information collected by execution information collection processing means 14;

FIG. 7 is an explanatory diagram of information specified by change information.

8 shows the static analysis information of FIG. 4, the execution information of FIG. 6, and FIG.
It is explanatory drawing of the prediction execution information calculated based on the change information of.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Source program 2 Characteristic analysis information 3 Measurement location information 4 Change information 5 Prediction execution information 10 Characteristic analysis mechanism 11 Program analysis processing means 12 Information display means 13 Source program generation processing means 14 Execution information collection processing means 15 Execution information prediction processing means 11a Static analysis information 13a Measurement program 14a Execution information 100,110 Program unit 101,102,111 Section 103,104 Measurement location 200,210 Static analysis information 300 Execution information 400,410 Change information 500,510 Prediction execution information

Claims (3)

(57) [Claims] In a characteristic analysis mechanism for analyzing characteristics of a program, a source program is input, and if there is measurement point information indicating a program unit, a DO loop, or the like from which execution information is collected, the measurement point information is input. Detects the points that make up the loop structure and the control branch points and measurement points where control moves to statements other than immediately after each program unit. For the loop structure, the start and end points of the loop, the type of loop, and the conditions for ending the loop , The data that controls the loop and whether the value of the control data is fixed or variable, and if the value is variable, collects information about the statement or variable that defines that value, and Is called at the control branch point where the branch location, branch type, branch condition, control transfer destination when branching, and program unit call are performed Collects information on the program unit name and call location, collects information on the measurement location and structure such as the measurement location is in a loop structure, and, after the processing of each program unit is completed, based on the above information, Analyzing the calling relationship of each loop, and determining whether or not each loop structure is a variable loop whose execution status may change if the execution conditions of the program such as the value of control data and the number of input data are changed, and static analysis If there is measurement location information in the information, search for measurement locations in the variable loop that are affected by execution information such as execution time and number of loops, and retrieve the collected information and the detected variable loops and measurement locations affected by the information. Program analysis processing means that outputs static analysis information, and static analysis information output by the program analysis processing means are input and affected by the same execution conditions. And information display means for displaying characteristic analysis information as information relating the variable loop and the measurement points affected by the variable loop if there is information related to the measurement points. A prediction calculation method for dynamic analysis information that is a feature. 2. The method of predicting and calculating dynamic analysis information according to claim 1, wherein the static analysis information and the source program obtained by the program analysis processing means are input, and information on a measurement location is specified. When a variable loop is detected in the source program by inputting it, a measurement statement for collecting execution information of the variable loop is inserted. When a measurement point is detected, a measurement statement for collecting execution information of the measurement point is written in the source program. Source program generation processing means for generating a measurement program by inserting it into the program, and the measurement program generated by this source program generation processing means are input, translated, combined, and executed to execute the number of loops, execution time, etc. Execution information collection processing means including processing for calculating information, and declaration of the number of input data and array of source program input by this execution information collection processing means Change information based on the calling conditions of the program by inputting the execution information for changing the execution condition and its assumed value, the static analysis information created by the program analysis processing means, and the execution information calculated by the execution information collection processing means A loop affected by the execution condition is detected, the number of times the loop is executed is searched from the execution information, and the execution information of the assumed value is calculated from the ratio to the assumed value specified in the change information. If so, calculate the assumed value execution information of the measurement location, repeat this process for all the affected loops, and predict the entire program, each program unit, and each loop based on the assumed value execution information. A prediction calculation method for dynamic analysis information, comprising execution information prediction processing means for calculating execution information. 3. The dynamic analysis information predictive calculation method according to claim 2, wherein the source program generation processing means inputs a source program, and inputs information on a measurement location if the information is specified. If the execution conditions such as the number of input data to be changed or the array declaration to be changed in the program and the change information related to the location affected by it are specified, enter them and specify the location specified by the change information in the source program. When detected, insert a measurement statement to collect execution information of the variable loop, and when a measurement location is detected,
A prediction calculation method of dynamic analysis information, wherein a measurement program is generated by inserting a measurement statement for collecting execution information of a measurement location into a source program.