JP2003216440A - Program analyzing method, program therefor and program analyzing system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve a problem in a conventional method that in a case when a user checks whether an object variable can be made private or not, the user must check the information that 'a definition point of the variable is executed before the execution of a use point of the object variable in the same loop iteration, when the use point of the object variable is executed', because the information is not displayed. <P>SOLUTION: The loop-independent control flow from a loop inlet point to the definition point of a certain variable or a loop-independent control flow from the loop inlet point to a use point of the variable not through the definition point of the certain variable is detected and extracted with respect to the loop given by an analyzed object designating part 191, by a hierarchical control flow graph creating part 120, an analyzed object input part 130, an innermost branch analyzing part 140, an inclusive relation analyzing part 150, and a control flow analyzing part 160, and the program execution order or reverse order is displayed by using a sentence on a sentence information display part 192 or the program slice along the control flow by the sentence information display part 192. <P>COPYRIGHT: (C)2003,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention
The present invention relates to a technique for outputting analysis information for a designated sentence or reference point, and particularly to a technique suitable for improving the efficiency of tuning a parallel computer program.

[0002]

2. Description of the Related Art Tuning of a program is required to bring out the performance of a parallel computer, but this tuning requires manual work and takes time. For example, in tuning loop parallelization,
The privatization of arrays is important, but program analysis for this is especially difficult.

Examples of conventional techniques related to tuning a program for such a parallel computer include, for example, "Swift Explorer: Uninteractive and" written by Shiwei Riao, Ammer Daiwan, Robert Pea Bosch Jr. Interprocedural Parallelizer "(PpOpPp99, pages 37 to 48, 1999) (Shih
-Wei Liao, Amer Diwan, Robert P. Bosch Jr., Anwar
Ghuloum, Monica S. Lam, SUIF Explorer: An Interact
ive and Interprocedural Parallelizer, PPoPP '99, p
p. 37-48, 1999).

In this technique, the compiler cannot determine whether or not there is a data dependency between two array references in a loop on the program, and if loop parallelization cannot be performed, the system analyzes the target program, The statement that influences the value of those array subscripts ("program slice") is displayed, and the user is conducting the parallelism investigation of the above loop based on this.

In this case, however, the user has to check the necessity and possibility of privatization of the loop parallelization target variable while following the displayed program slice, and If there is a complicated control flow, there is a problem that the user's load is heavy, it takes time to investigate, and it is easy to make an error.

[0006]

The problem to be solved is that in the prior art, the system can display only program slices as information for investigating loop parallelization.

An object of the present invention is to solve these problems of the prior art and to reduce the time required for the user to make a privatization determination of a target variable.

[0008]

In order to achieve the above object, according to the present invention, when a program stored in a storage device is read and the program is analyzed and an analysis result is output, a loop designated as an analysis target is executed. The control flow that reaches the loop exit of the loop and any one of the variable definition point and the usage point is detected from the loop entrance of the loop and one of the definition point and the usage point of the variable defined in this loop. , The control flow that does not cross the loop repetition is extracted from the detected control flow, and the statement information on the extracted control flow is stored in a file in the program execution order or in the reverse order or in the source program on the display device screen. Output by highlighting. The information output in this way is
Whether or not privatization such as "the definition point of the target variable is executed" and "when the use point of the target variable is executed, the definition point of the variable is executed before the execution in the same loop iteration" It shows the judgment conditions,
The research load on the information of the user is reduced.

[0009]

BEST MODE FOR CARRYING OUT THE INVENTION Embodiments of the present invention will be described in detail below with reference to the drawings.

FIG. 1 is a block diagram showing a configuration example of a program analysis system according to the present invention, and FIG. 14 is an explanatory diagram showing an example of a designation operation of a program to be analyzed by the program analysis system in FIG. 20 is an explanatory diagram showing an example of an analysis result corresponding to the program of FIG. 14 of the program analysis system in FIG.

The program analysis system shown in FIG.
A computer system including a CPU (Central Processing Unit), a main memory, a display device, an input device, and an external storage device, and a CD-R via an optical disk drive device or the like.
After the programs and data recorded in the storage medium such as the OM are installed in the external storage device, the contents read from the external storage device into the main memory and processed by the CPU are realized as the functions of the respective processing units.

That is, the program analysis system of this example comprises an analysis processing unit 100, an analysis target designating unit 191, and an information display unit 192.
Furthermore, the syntax analysis unit 110, the hierarchical control flow graph creation unit 120, the analysis target input unit 130, the innermost branch analysis unit 1
40, an inclusion relation analysis unit 150 including an inclusion flag setting unit 151 and a passage flag setting unit 152, and a control flow analysis unit 160.

With such a configuration, in the program analysis system of this example, the input program display screen 1 shown in FIG.
In response to the loop program analysis instruction 401,
The program is analyzed, and the analysis result is shown in FIG.
As shown in, the predetermined sentence information is highlighted on the screen of the display device.

That is, in the program analysis system,
When performing analysis of a program for which analysis is specified regarding privatization and output of analysis results, enter the identification information for the loop to be analyzed and the definition points and points of use of variables in the loop, and then loop the loop to be analyzed. Detects a control flow that reaches either one of the definition point and the use point of the loop exit and the variable from any one of the definition point and the use point of the entry and the variable and does not cross the loop of the analysis target loop. -The extracted and detected / extracted sentence information on the control flow is output to a file or highlighted on the source program.

For example, the control flow from the loop entrance of the loop to be analyzed to the other without passing through one of the variable definition point and the use point, or from one of the variable definition point and the use point to the other The control flow that reaches the loop exit of the analysis target loop without passing is detected and extracted, and the statement information is displayed and output in the program execution order or the reverse order.

Program slices along the detected / extracted control flow for array subscripts or scalar variables at variable definition points or use points are used as statement information. Alternatively, the intersection of the program slice for the array subscript or the scalar variable at the definition point or the use point of the variable and the detected / extracted control flow is used as the sentence information.

In this way, in the program analysis system, the statement on the loop independent control flow from the loop entry point to the definition point of a certain variable, or the program slice along the control flow, or the loop entry point A statement on a loop-independent control flow from a variable to a point of use of the variable without passing through the definition point of the variable, or a program slice along the control flow is displayed in the program execution order or the reverse order.

In this display, the condition "the definition point of the target variable is executed" and "when the use point of the target variable is executed, in the same loop iteration, before the execution of the variable The definition point is executed ", and these conditions are used when the user investigates the necessity and privatization possibility of the privatization of the target variable. The system display eliminates the need for the user to investigate the condition by himself, thus reducing the load.

The details of the configuration and operation of the program analysis system in FIG. 1 will be described below.

In FIG. 1, the analysis target designating section 191 is
The input program 181 stored in the external storage device or the like is read and input, and the analysis target file 182 is output by allowing the user to specify the analysis target loop and the analysis target reference point in the program on the operation screen shown in FIG.

An intermediate word 171, a hierarchical control graph 172, an analysis target list 173, a control flow list 174, and a sentence information file 183, which will be described later, including the analysis target file 182, are stored in an external storage device.

In the analysis processing unit 100, the syntax analysis 11
0 inputs the input program 181, generates the intermediate word 171 and outputs it, and the hierarchical control flow graph creation unit 120
Inputs an intermediate word 171 to generate a hierarchical control flow graph 1
Create 72.

The analysis target input unit 130 inputs the analysis target file 182, writes the analysis target in the hierarchical control flow graph 172, and displays the analysis target list representing the calling procedure sequence from the analysis target loop to the analysis target reference point. 173 is generated and output.

The innermost branch analysis unit 140 receives the hierarchical control flow graph 172 and branches to a sentence including the analysis target reference point and a node branching from the sentence including the analysis target reference point. Edges corresponding to nodes and branches other than the inner loop entrance are detected, and flags are set so that the control flow does not pass through these edges.

Further, the inclusion relation analysis unit 150 inputs the hierarchical control flow graph 172, analyzes whether or not each node includes an analysis target reference point, and selects one of the analysis target reference points.
Flag a node that does not include the analysis target reference point that is specified not to pass.

Further, the control flow analysis unit 160 inputs the analysis target list 173 and the hierarchical control flow graph 172, analyzes the control flow connecting the analysis target reference points designated by the user, and controls the control flow. Listing 174
Further, the control flow list 174 is input to analyze the statement or program slice in the program on the control flow, and the analysis result is output to the statement information file 183.

Then, the sentence information display unit 192 displays the sentence information file 18 stored in the analysis processing unit 100 in this way.
3 and the input program 181 are read from the storage device and input, and the sentence information obtained by the analysis processing unit 100 is highlighted on the input program 181 on the screen of the display device.

In the inclusion relation analysis unit 150 in the analysis processing unit 100, the inclusion flag setting unit 151 inputs the hierarchical control flow graph 172 and sets a flag indicating whether or not each node includes the analysis target reference point, The passage flag setting unit 152 sets a flag in the analysis target reference points that do not include the analysis target reference points that are designated not to pass.

Next, the operation and display of the program analysis system having such a configuration will be described with reference to FIGS.
This will be specifically described.

In FIG. 14, the analysis target designating section 191 is provided.
Shows a specific operation example of the analysis target designating unit 1
91 displays an operation screen including an input program display screen 1401, an analysis target line number display unit 1402, an analysis target line display unit 1403, and a sentence number display unit 1404. A specific example of the input program 181 is displayed on the input program display screen 1401.

User input program display screen 1401
3 corresponding to the line numbers "11", "13", and "17" above
When one line is selected, the line number and the line (text content thereof) are displayed in the analysis target line number display unit 1402 and the analysis target line display unit 1.
403 is displayed.

When the user selects the analysis target reference point 1405 on the analysis target line display portion 1403, it is displayed ("x
(J) ”) is highlighted. In addition, the user passes the sentence indicated by "skip" from the sentence indicated by "start" by designating each sentence in "start", "skip", and "end" in the item column 1406. Without "end"
It is possible to specify the control flow leading to the sentence indicated by.

The sentence corresponding to "skip" can be omitted. The designation of the analysis target reference point (1405) is for calculating the program slice on the control flow. To display only the statement on the control flow, the designation of the analysis target reference point (1405) is required. It can be omitted.

FIG. 15 is an explanatory diagram showing an example of the contents of the analysis target file in FIG.

This example is the contents of the analysis target file 182 output by the analysis target designating unit 191 of FIG. 1 based on the contents shown in FIG. Before explaining the contents of FIG. 15, the tag structure and tag format of the file to be analyzed are shown in FIG.
It will be explained based on.

FIG. 9 is an explanatory diagram showing an example of the tag structure of the file to be analyzed.

In the example shown in FIG. 9, there are a total of 14 tags. Each tag begins with the symbol "#" and ends with a line break. Tags used in pairs are shown by connecting them with a curve on the left side in the figure.

For example, the tag 901 and the tag 914 are used as a pair. One of the tags used in pairs begins with the symbol "#" and the other begins with the symbol "# /". Although not shown in FIG. 9, there are also tags with arguments. Also,
Tags that start with the symbol "# /" have no arguments.

FIG. 10 is an explanatory diagram showing an example of the tag format of the analysis target file in FIG.

FIG. 10A shows a tag format 1001 for each tag, and the [column] of the argument of "#LINE" represents the column when the reference point is designated. Omitted if no reference point is specified.

In FIG. 10B, the tag "#K" in FIG.
IND ”903 shows an analysis method type 1002 that is an argument, and in FIG. 10C and FIG. 10D, a sentence type 1003 that is an argument of the tag“ #LINE ”905.9908.911 in FIG. The type 1004 is shown.

Returning to the explanation of FIG. 15, in FIG. 15, the line 1501 indicates that the procedure name including the loop to be analyzed is "te".
st ”, and the line 1502 indicates that
It indicates that the sentence number of the DO sentence is 10.

Line 1503 shows that the type of analysis method is from the sentence indicated by tag "ST" to the sentence indicated by tag "EN" without passing through the sentence indicated by tag "SK". Lines 1504 to 1506 indicate that the start sentence is a sentence other than the procedure call sentence or the loop with the line number “11” of the procedure name “test” and the sentence type is the analysis target.

Further, lines 1507 to 1510 show information on a sentence that the user has specified not to pass, and lines 1508 and 1509 among them indicate the upper layer when the sentence or procedure is regarded as a layer. It shows a sequence of sentences from to the lower layer to the sentence to be analyzed.

Here, the line 1508 shows the “DO statement” of the “DO loop”, which is not the analysis target but only in the higher hierarchy of the analysis target sentence, and therefore the target type as the argument of the tag is It becomes "NO". Further, the line 1509 is the analysis target, and the target type that is the argument of the tag is “YES”.

Further, lines 1511 to 1514 show ending sentences, and these are also columns of sentences.

The above tag "ST" (lines 1504 to 150)
6), tag "SK" (lines 1507-1510), tag "E
The sentences indicated by "N" (lines 1511-1514) are
In column 1406 of FIG. 14, "start", "skip", and "e"
It corresponds to the sentence (1403) indicated by "nd".

Next, the operation of each processing unit in the analysis processing unit 100 of FIG. 1 which inputs the analysis target file 182 will be described.

The syntax analysis unit 110 uses the input program 18 displayed on the input program display screen 1401 of FIG.
1 is input, an intermediate word 171 is generated and output. Here, the intermediate language 171 is a tree-like representation of the input program, but since the content is almost the same as the input program 181, the input program displayed on the input program display screen 1401 will be referred to as an intermediate language in the following description. 171.

The hierarchical control flow graph generator 120
The intermediate word 171 is input to create the hierarchical control flow graph 172 illustrated in FIG.

FIG. 16 is an explanatory diagram showing a structural example of the hierarchical control flow graph in FIG.

Hierarchical control flow graph 1 shown in FIG.
Reference numeral 72 is created by regarding the input program displayed on the input program display screen 1401 of FIG. 14 as the intermediate language 171 and inputting this intermediate language.

The nodes of the hierarchical control flow graph 172 include an ingress node (indicated as "S" in FIG. 16), an exit node (at "E"), a loop node (at "LOOP"), and a basic block node (at "S"). BB ”), branch node (the same“ B
R ”), a confluence node (same as“ JOIN ”), and a procedure call node. Further, each node is connected by an edge according to the control flow.

As described above, in the hierarchical control flow graph, the loop and the procedure are one hierarchy. Each hierarchy is a directed graph with no back edges, starting at the ingress node and ending at the egress node. Here, the back edge is an edge that faces the entrance node. 1610, 1620, and 1630 in the figure are each a hierarchy.

In FIG. 16, a loop node 1601
Is connected to the hierarchy 1610 corresponding to the loop body. Furthermore, the loop node 161 in the hierarchy 1610
Layers 1620 and 1630 corresponding to the respective loop bodies are connected to 4, 1616.

In the following, the hierarchical control flow graph is abbreviated as "LCF graph", and one hierarchy thereof is referred to as "part L
CF graph ". Further, the loop node 1601 is called an upper node of an arbitrary node included in the partial LCF graph, and conversely, an arbitrary node included in the partial LCF graph is called a lower node of the loop node 1601.

The analysis target input unit 130 inputs the analysis target file 182 and writes the analysis target in the hierarchical control flow graph 172, and the analysis target list 173 representing the calling procedure sequence from the analysis target loop to the analysis target reference point.
Create and output. c

FIG. 2 is a flow chart showing an example of the processing operation of the analysis target input section in FIG.

In the analysis target input section 130 shown in FIG.
First, in step 201, the analysis target file 182 shown in FIG. 15 is input, and then in step 202, in the analysis target file 182 of FIG. 15, the tags “ST”, “SK”, and “EN” are used. Point to the indicated reference point or sentence with "st", "sk", "en". Then, in step 203, an analysis target list 173 having the contents shown in FIG. 17 is created.

FIG. 17 is an explanatory diagram showing a configuration example of the analysis target list in FIG.

The analysis target list 173 in this figure is obtained as a result of processing by the analysis target input unit 130 in FIG. These correspond to "st", "sk", and "en" in order from the top.
Here, each field of each table constituting the list will be described with reference to FIG.

FIG. 13 is an explanatory diagram showing a table configuration example of the analysis target list in FIG.

FIG. 13 shows the contents of each field of the table of the analysis target list 173 having the configuration shown in FIG. 17, and 1301 is a pointer to the next table, 13
02 represents a pointer to the immediately preceding table.

Reference numeral 1303 represents the statement number of the statement represented by the table, and 1304 represents the node type of the node corresponding to the statement. There are four node types shown in 1310. And 1305 is a pointer to the corresponding node of the LCF graph.

In FIG. 17, the start tables 1701, 1703, 1706 of each list correspond to the analysis target loop. Therefore, the sentence number is "10" and the node type is "LOOP". For the tables after the start table of “st”, “sk”, and “en”, the statement numbers of each table in the list correspond to the statement numbers of the corresponding tags in FIG.

Next, referring to FIG. 2, the analysis target input unit 13
In step 204, 0 sets a pointer from the analysis target list to the node of the LCF graph lcf. In the start tables 1701, 1703, and 1706 of each list corresponding to the analysis target loop in FIG. 17, it corresponds to the node 1601 of the LCF graph.

In the tables after the start table of "st", "sk", and "en", the pointer to the LCF graph is
1612, 1614 and 1622 in FIG. 17, respectively.
Corresponding to 1616 and 1632.

Then, in FIG. 2, the analysis target input unit 1
30 flags the nodes of the LCF graph lcf in step 205. These flags are 16 in FIG.
"St", "sk", "en" beside 12, 1622, 1632.
Is shown.

The innermost branch analysis unit 140 in FIG. 1 inputs such a hierarchical control flow graph 172, branches to a sentence including the analysis target reference point, and branches from the sentence including the analysis target reference point. The nodes other than the innermost loop entrance and the edges corresponding to the branch are detected, and a flag is set so that the control flow does not pass through these edges. The processing operation of the innermost branch analysis unit 140 will be described with reference to FIG.

FIG. 3 is a flow chart showing an example of the processing operation of the innermost branch analysis unit in FIG.

In this example, the innermost branch analysis unit 140 of FIG.
6 shows an example of processing operation (algorithm) based on the hierarchical control flow graph shown in FIG.
Since “sk” is designated in the hierarchical control flow graph shown in FIG. 16, 01 becomes “Y”, and the processing proceeds to the next step 302 and subsequent steps.

In this step 302, the node 1622 including “sk” is set to “node”, and in step 303, the LCF graph 1620 including “node” 1622 is lc.
Let f. Then, since the “node” 1622 controls the exit 1623 of the lcf 1620 in the lcf (LCF graph) 1620, it becomes “Y” in step 304.

Lcf (LCF graph) 1620 is "st"
Since neither "en" is included, "N" is obtained in step 305, and lcf (LCF graph) is obtained in step 308.
The node 1614, which is the upper node of 1620, is "node".
And

At step 309, since lcf (LCF graph) 1620 does not correspond to the loop to be analyzed,
It becomes “N”, the process returns to step 303, and node1
A partial LCF graph 1610 including 614 is designated as lcf.

In the next step 304, node16
14 is the exit node 16 of the lcf 1610 within the lcf 1610
Since it does not control 17, it becomes “N”, and step 310
Then, the process proceeds to step 311.

At step 310, lc is output from node 1614.
Following the entrance direction of f1610, the node 1614 finds the first node that does not dominate, and sets the non-passage flag at the edge between that node and the node immediately after it. In the example shown in FIG. 16, the nodes and edges obtained by the processing in step 310 are 1613 and 1640.

In step 311, node1614
From the exit of lcf1610 to node1614
Finds the first node that does not dominate and flags the edge between that node and the node immediately preceding it as a non-passing flag.
In the example shown in FIG. 16, the nodes and edges obtained by the processing in step 311 are 1615 and 1650. As described above, the non-passage flag is set on the two edges 1640 and 1650.

In step 305, lcf (LC
F graph) contains "st" or "en",
Proceed to the process of step 306, and execute "sk" (or "e
n ”) is in the exit direction from“ st ”(or“ sk ”), and if there is not, in step 307, lc
It is determined whether or not there is a loop between f (LCF graph) and the loop to be analyzed.

In step 306, "Y"("sk" or "en" is in the exit direction from "st" or "sk"), and in step 307, "N" (lcf
If there is no loop between the analysis target loop and the analysis target loop) or if “Y” (lcf is a partial LCF flag corresponding to the analysis target loop) in step 309, the process proceeds to step 312 and the “condition There is no path that satisfies the above condition ", and the processing of this tool ends.

The hierarchical control flow graph 172 thus created is included in the inclusion relation analysis unit 150 in FIG.
Is input to and the processing shown in FIG. 4 is performed.

The inclusion relation analysis unit 150 in FIG. 1 comprises an inclusion flag setting unit 151 and a passage flag setting unit 152. The inclusion flag setting unit 151 inputs the hierarchical control flow graph 172 in FIGS. 1 and 16, and sets a flag indicating whether or not each node includes the analysis target reference point. The algorithm of the inclusion flag setting unit 151 is shown in FIG.
It will be explained based on.

FIG. 4 is a flow chart showing an example of the processing operation of the inclusion flag setting section in FIG.

In this example, the maximum range for obtaining the program slice is the loop to be analyzed, and in step 400, L
The entry node 1611 of the partial LCF graph corresponding to the loop body of the analysis target loop is taken as the CF start node. Hereinafter, the process A (steps 401 to 411) is called for the node 1611.

In the process A, the node 1611 becomes “Y” in step 403, and the process proceeds to step 408.
The process A is sequentially called for all the nodes in the current layer (layer 1610 in FIG. 16).

In this process, the next node 1612 contains "st", so it becomes "Y" in step 404, and in the process of step 409, the "st" flag of the upper node 1601 is turned "ON".

The next branch node 1613 is step 401.
Nothing corresponds to any of ~ to 404, and no processing is performed.
The next loop node 1614 becomes “Y” in step 401, and the process A is called for the loop entry node 1621.

For this ingress node 1621, the process proceeds from step 403 to step 408, and the hierarchy 162
The process A is called for all the nodes in 0. First, node 1
Since 622 includes “sk”, the process proceeds from step 404 to step 409, and “s” is assigned to its upper node 1614.
Set the "k" flag.

The next egress node 1623 is step 401.
To 404, the process is not performed, and the process for the loop node 1614 in step 406 is completed. In the next step 411, the “sk” flag of the upper node 1601 is set to the current node 1614. It is set to ON by the "sk" flag of.

A junction node 1615 next to the node 1614
Does not correspond to any of steps 401 to 404 and is not processed.

The next loop node 1616 is processed in the same manner as the loop node 1614, and the "en" flag is set in this loop node 1616 and the upper loop node 1601.

[St], [s] set by the above processing
The “k” and “en” flags correspond to the node 160 in FIG.
1, 1614 and 1616 are displayed with underlines beside them.

With respect to such an inclusion flag setting unit 151, the passage flag setting unit 152 determines, among the analysis target reference points,
Flag a node that does not include the analysis target reference point that is specified not to pass. Hereinafter, the passage flag setting unit 152
The processing operation of will be described with reference to FIG.

FIG. 5 is a flow chart showing an example of the processing operation of the passage flag setting section in FIG.

In FIG. 5, the passage flag setting unit 15 of FIG.
The algorithm of No. 2 is described below. The passage flag setting unit 152 first sets the LCF in step 500.
The entry node (node 1611 in FIG. 16) of the partial LCF graph corresponding to the loop body of the loop to be analyzed is taken as the start node of, and the process B is applied to this node 1611.

In the process B, the node 1611 is "N" in step 501 and "Y" in step 505.
Then, the process B is called for all the nodes in the current hierarchy 1610 (step 506).

Nodes 1612 and 16 in the current hierarchy 1610
Both 13 are "N" in both steps 501 and 505 and are not processed. The next node 1614 becomes "Y" in step 501. Then, since the "sk" flag is set in the node 1614 by the processing of the inclusion flag setting unit 151 described above, step 502
Becomes “N” in the lower hierarchy, and the entry node 162 of the lower hierarchy
The process B is called for 1 (step 504).

This node 1621 becomes "N" in step 501 and "Y" in step 505,
The process B is called for all the nodes in the current hierarchy 1620. However, the nodes 1622 and 1623 in the current hierarchy 1620 are both “N” and not processed in any of steps 501 and 505.

The processing of the layer 1620 is completed, and the upper layer 1
Returning to the processing of 610, first, the node 1615 becomes “N” in both steps 501 and 505 and is not processed, but the next node 1616 is a loop node, becomes “Y” in step 501, and in step 502, The “sk” flag is “OFF” and becomes “Y”, and in step 503, the passage flag is set in the node 1616.

After that, in step 504, the process B is called for the entry node 1631 of the lower layer, but the process for the lower layer is not performed as in the case of the node 1614.

The pass flag set by the above processing is indicated by "pass" with an underline beside the node 1616 in FIG.

In this way, the inclusion flag setting unit 151
The hierarchical control flow graph with the flag set in the passage flag setting unit 152 is input to the control flow analysis unit 160 in FIG.

The control flow analysis unit 160 inputs the hierarchical control flow graph 172 and the analysis target list 173, analyzes the control flow connecting the analysis target reference points designated by the user, and analyzes the control flow. 174 is output again, and the control flow list 174 is input again to analyze the statement or program slice in the program on the control flow, and the analysis result is the statement information file 18
Output to 3. Such a control flow analysis unit 160 will be described with reference to FIGS.

FIG. 6 is a flow chart showing a first processing operation example of the control flow analysis unit in FIG.
3 is a flowchart showing a second processing operation example of the control flow analysis unit in FIG. 1.

6 and 7 explain the algorithm of the control flow analysis unit 160 in FIG. 1, and here, only the processing related to the hierarchical control flow graph 172 shown in FIG. 16 will be described. .

First, the control flow analysis unit 160 determines in step 600 of FIG. 6 that the node including “st” (see FIG. 16).
7 is called for the node 1612) in FIG. Regarding this node 1612, it becomes “Y” in step 700 of FIG. 7, and in step 721, the process C is called to the next node 1613 shown in FIG.

The node 1613 is a branch node, which becomes “Y” in step 705, and the process proceeds to step 716. In the processing of step 716, the non-passage flag of the edge 1640 is “ON”, the edge toward the merging node 1615 is selected, and the merging node 1
The process C is called for 615.

Regarding this merge node 1615, it becomes "Y" at step 706, and at step 717, the sentence number in the program displayed on the input program display screen 1401 of FIG. 14 which is the last sentence in the node 1613 immediately before the merge. A table having the sentence number of the sentence “11” as the sentence number and the node type “JOIN” is created as the first node of the control flow list (174), and the process C is called to the next node 1616. .

Since this node 1616 is a loop node and becomes “Y” in step 704 and does not include “sk”, it becomes “N” in step 712 and “en”.
Therefore, the analysis target list (173) from the current loop node 1616 to “en” is added to the control flow list (174) (step 714). FIG. 18 shows the control flow list obtained as a result of this.

FIG. 18 is an explanatory diagram showing a configuration example of the control flow list in FIG.

In FIG. 18, 1801 is a table created by the processing by the control flow analysis unit 160 in step 717 in FIG. 7, and 1802 and 1803 are tables created by the processing in step 714 in FIG. . These two tables 1802, 1803
Is the same as the tables 1707 and 1708 in FIG.

Then, the control flow list (174) thus obtained is subjected to slice analysis in step 715a. This slice analysis will be described with reference to FIG.

FIG. 8 is a flowchart showing a detailed example of slice analysis in FIG.

FIG. 8 shows an algorithm of slice analysis. In step 801, the tables 1801 to 1801 of the control flow list 174 shown in FIG. 18 are shown.
The program slice is analyzed according to the control flow represented by 803.

The details of this program slice analysis are described in "Swift Explorer: An Interactive and Interprocedural Parallelizer" by P.W.Liao, Amar Daiwan, Robert Peabosch Jr., Anwer Guroum, and Monica Eslam. 37 to 48, 1999) (Shih-Wei L
iao, Amer Diwan, Robert P. Bosch Jr., Anwar Ghulou
m, Monica S. Lam, SUIF Explorer: An Interactive an
d Interprocedural Parallelizer, PPoPP '99, pp. 37-
48, 1999) and will not be described here.

The program slice is shown in FIG. 1 and FIG.
The analysis is performed using the reference point designated as the analysis target file 182 in FIG. Analysis target file 1 shown in FIG.
In FIG. 14, in FIG.
Is analyzed for reference point 1405 at.

At the next step 802, the serial number
num is incremented by 1 from 0 to “1”. And
In step 803, num is used as a serial number, and the statement number of the statement on the program slice is output to the statement information file 183 of FIG. 1 according to the output file format.
The sentence information file 183 will be described with reference to FIGS. 11, 12 and 19.

FIG. 11 is an explanatory view showing an example of the tag structure of the sentence information file in FIG. 1, FIG. 12 is an explanatory view showing an example of the tag format in the sentence information file of FIG. 11, and FIG. 19 is shown in FIG. It is explanatory drawing which shows the structural example of the sentence information file output by slice analysis.

As shown in FIG. 11, in this example, the sentence information file 183 has a total of four tags. Each tag begins with the symbol "#" and ends with a line break. Tags used in pairs are shown by connecting them with a curve on the left side in the figure. For example, the tag 1101 and the tag 1103 are used as a pair.

One of the tags used in pairs begins with the symbol "#" and the other begins with the symbol "# /". Although not shown in FIG. 11, there are tags with arguments. Tags that start with the symbol "# /" have no arguments.

In FIG. 11, 1102 is a set of tags, which are defined by 1104-1109.
1102 is self in its definition 1107
Is included. This represents that another procedure is called from one procedure.

In FIG. 12, reference numeral 1201 indicates the tag format for each tag. Here, "#LIN
"E" and "#CALL" are sorted in line number order within one "#PROC". Immediately after "#CALL", the procedure information "#PROC" of the called procedure follows. In addition, the "CALL" statement is not included in the sequence of statements indicated by "#LINE".

The sentence information file 183 shown in FIG. 19 is output by the control flow analysis unit 160 in FIG. 1 by the slice analysis processing in steps 715a and 718 in FIG.

In FIG. 19, among lines 1901-1906, line 1903 represents one sentence consisting of sentence number “11”.
Line 1904 represents a group of four consecutive sentences from sentence numbers “15” to “18”.

The sentence information display unit 192 in FIG. 1 inputs the sentence information file 183 and the input program 181, and as shown in FIG. 20, the sentence information obtained by the analysis processing unit 100 is input into the input program 181. Display on top.

FIG. 20 is an explanatory view showing an example of a display result by the sentence information display section in FIG.

In the example of the display screen 2000 shown in FIG. 20, the sentences output to the sentence information file 192 with the contents shown in FIG. 19 are displayed in a shaded manner. As a result, there is only one control flow from the sentence with the sentence number “11” that is the loop entry to the minute number “17” without passing through the sentence with the sentence number “13”. The program slices can be easily read by being shaded.

In the example of FIG. 20, all the sentences are displayed at once, but since the sentences are arranged in the sentence information file 183 in the order of execution, they may be displayed in that order or in the reverse order.

As described above, with respect to the array "x",
One iteration of the loop shows the control flow from the loop entry to the use point of the array "x" without passing through the definition point of the array "x". Since this control flow is executed when the value of "i" is neither "1" nor "5", it is executed when the value of "m" is "2" or more in the sentence number "10". Therefore, at this time, it can be seen that privatization cannot be applied to the variable “x”.

As described above with reference to FIGS. 1 to 20, in the program analysis technique of this example, a computer for inputting analysis target designating information designating an analysis target in a program and outputting analysis information corresponding to the analysis target designating information. In the system, the identification information for the analysis target loop, the definition point and the use point of a certain variable is input, and one of the loop entrance of the analysis target loop, the variable definition point, and the use point
One of loop exit, variable definition point, and usage point 1
A control flow that has reached a certain point and does not cross the loop of the analysis target is detected and extracted, and the statement information on this control flow is output to a file or the statement information is highlighted on the source program.

Alternatively, from the loop entrance of the loop to be analyzed, one of the variable definition point and the use point is not reached but the other is reached, or one of the variable definition point and the use point is not passed through the other. The control flow that reaches the loop exit of the analysis target loop is detected and extracted, and the statement information on this control flow is output to a file or the statement information is highlighted on the source program.

The program slice along the detected / extracted control flow for the array subscript or the scalar variable at the definition point or use point of the variable is used as the sentence information. Alternatively, the statement slice is the intersection of the program slice for the array subscript or scalar variable at the definition point or use point of the variable and the detected / extracted control flow, and such statement information is displayed in the program execution order or the reverse order. .

As a result, the user can easily verify the first condition of privatization, "the definition point of the target variable is executed", by looking at the few sentences in the program. You can reduce the time to know the need for privatization.

In the program, the second condition of privatization, "when the use point of the target variable is executed, the definition point of the variable is executed before the execution within the same loop iteration", is less in the program. Since it can be easily verified by looking at the sentence, it is possible to reduce the time for the user to know that the target variable can be privatized.

Further, since the first condition for privatization or the second condition for privatization can be verified by displaying the statements in the program execution order or the reverse order thereof, the user can easily check a program having a complicated control flow. In time, you can be sure to carry out an investigation.

The present invention is not limited to the examples described with reference to FIGS. 1 to 20, and various modifications can be made without departing from the spirit of the invention. For example, the present invention can be used for purposes other than allowing a user to verify the privatization possibility of a variable described in this embodiment.

For example, it is known that a loop cannot be parallelized if there is a loop-carrying flow dependency between two reference points for a certain variable. By displaying the control flow that may cause this data dependence to the user, the user can verify the loop-carrying flow dependence that could not be definitely determined by the static analysis by the compiler.

For this reason, the control flow from one definition point for a variable to the loop exit is used for the two reference points that the compiler considers to be possibly data-dependent by using the statement or program slice above it. To display. In addition, the control flow from the loop entry to the point of use for that variable is also displayed. By displaying these two one by one or simultaneously using two screens, the user can verify the existence of data dependence, that is, verify the loop parallelization.

Further, in this example, the recording medium is a CD-R.
Although it is described that OM is used, FD (Flexible D
isk) or the like may be used as the recording medium. Also,
Regarding the installation of the program, the program may be downloaded and installed via the network via the communication device.

[0139]

According to the present invention, for a given loop, a loop-independent control flow from a loop entry point to a variable definition point or a variable entry point from a loop entry point is not passed. Since the loop-independent control flow to the point of use of the variable is displayed in the program execution order or the reverse order using the statement above it or the program slice along the control flow, conventionally, the target variable is When the user investigates whether or not it can be privatized, the system says, "When the usage point of the target variable is executed, the definition point of that variable is executed before the execution within the same loop iteration." Since the information is not displayed, the user has to check it by himself, which can solve the problem that the user's load is heavy. That is, according to the present invention,
The user does not need to check it by himself, and it is possible to reduce the time required for the user to make a privatization determination of the target variable and to reduce the load on the user.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration example of a program analysis system according to the present invention.

FIG. 2 is a flowchart showing an example of processing operation of an analysis target input unit in FIG.

FIG. 3 is a flowchart showing an example of processing operation of an innermost branch analysis unit in FIG.

FIG. 4 is a flowchart showing an example of processing operation of an inclusion flag setting unit in FIG.

5 is a flowchart showing an example of processing operation of a passage flag setting unit in FIG.

FIG. 6 is a flowchart showing a first processing operation example of a control flow analysis unit in FIG.

FIG. 7 is a flowchart showing a second processing operation example of the control flow analysis unit in FIG.

8 is a flowchart showing a detailed example of slice analysis in FIG.

FIG. 9 is an explanatory diagram showing a tag structure example of an analysis target file.

10 is an explanatory diagram showing an example of a tag format of a file to be analyzed in FIG.

11 is an explanatory diagram showing an example of a tag structure of the sentence information file in FIG.

12 is an explanatory diagram showing an example of a tag format in the sentence information file of FIG.

13 is an explanatory diagram showing a table configuration example of an analysis target list in FIG. 1. FIG.

14 is an explanatory diagram illustrating an example of a designation operation of a program to be analyzed by the program analysis system in FIG.

FIG. 15 is an explanatory diagram showing an example of contents of an analysis target file in FIG.

16 is an explanatory diagram showing a structural example of a hierarchical control flow graph in FIG.

17 is an explanatory diagram showing a configuration example of an analysis target list in FIG.

18 is an explanatory diagram showing a configuration example of a control flow list in FIG.

19 is an explanatory diagram showing a configuration example of a sentence information file output by the slice analysis in FIG.

FIG. 20 is a diagram of the program analysis system in FIG.
It is explanatory drawing which shows the example of an analysis result corresponding to the program of FIG.

[Explanation of symbols]

100: analysis processing unit, 110: syntax analysis, 120: hierarchical control flow graph creation unit, 130: analysis target input unit,
140: innermost branch analysis unit, 150: inclusion relation analysis unit, 1
51: inclusion flag setting unit, 152: passage flag setting unit,
160: control flow analysis unit, 171: intermediate language, 172:
Hierarchical control flow graph, 173: analysis target list, 1
74: control flow list, 181: input program, 1
82: analysis target file, 183: sentence information file, 1
91: analysis target designation unit, 192: sentence information display unit, 901
~ 914: tag, 1001: tag format, 100
2: analysis method type, 1003: sentence type, 1004: target type, 1101 to 1109: tag, 1201: tag format, 1301, 1302: pointer, 130
3: sentence number, 1304: node type, 1305: pointer, 1310: node type, 1401: input program display screen, 1402: analysis target line number display unit, 140
3: Analysis target line display unit, 1404: Sentence number display unit, 14
05: analysis target reference point, 1406: item column, 1501
1516: line, 1601, 1611 to 1617, 162
1-1623, 1631-1633: Loop node, 1
610, 1620, 1630: Hierarchy, 1701, 170
3, 1706: start table, 1702, 1704, 1
705, 1707.1708, 1801 to 1803: table, 1901 to 1906: row, 2000: display screen.

Claims (7)

[Claims] 1. A program analysis method by a computer system for reading a program stored in a storage device and analyzing the program and outputting an analysis result, comprising: a loop entrance on a program designated as an analysis target; The definition point and the usage point of the variable defined in the loop are stored in a storage device, and from either one of the stored entry point and the definition point and the usage point of the variable, the exit point of the loop and the Among the control flows that reach either one of the definition point and the usage point, a control flow that does not cross the repetition of the loop is extracted, and sentence information on the extracted control flow is output. analysis method. 2. A program analysis method by a computer system for reading out a program stored in a storage device and analyzing the program and outputting an analysis result, comprising: a loop entrance on a program designated as an analysis target; A control flow in which the definition point and the use point of the variable defined in the loop are stored in a storage device, and the other is reached from the stored entry point of the loop without passing through either the definition point or the use point of the variable, And, the control flow reaching the exit of the loop from one of the definition point and the use point of the variable without passing through the other is extracted, and the sentence information on the extracted control flow is output. Program analysis method. 3. The program analysis method according to claim 1 or 2, wherein the sentence information is extracted from an array subscript or a scalar variable at a definition point or a use point of the variable. A program analysis method characterized by outputting a program slice according to the control flow described above. 4. The program analysis method according to claim 1 or 2, wherein the sentence information includes a program slice for an array subscript or a scalar variable at a definition point or a use point of the variable. A program analysis method characterized by outputting the intersection with the extracted control flow. 5. The program analysis method according to claim 1, wherein the sentence information is displayed in a program execution order or a reverse order. 6. The computer according to any one of claims 1 to 5.
A program for executing each step in the program analysis method according to any one of 1. 7. A computer system for reading a program stored in a storage device to analyze the program and output an analysis result, and a syntax analysis means for generating an intermediate word from a program to be analyzed and outputting the intermediate word to the storage device. A hierarchical control flow graph creating means for reading the intermediate word from the storage device and generating a hierarchical control flow graph based on the intermediate word and outputting the hierarchical control flow graph to the storage device; and an analysis target loop and analysis by the user on the analysis target program. An analysis target designating unit that receives the designation of the target reference point, generates an analysis target file corresponding to the received designated content, and outputs the file to the storage device, and reads the analysis target file and the hierarchical control flow graph from the storage device. , Write the analysis target indicated by the analysis target file in the hierarchical control flow graph, and Analysis target input means for generating an analysis target list representing a calling procedure sequence from the phantom loop to the analysis target reference point and outputting it to the storage device; and reading the hierarchical control flow graph from the storage device, and including the analysis target reference point. A node that branches to a sentence and a node that branches from a sentence that includes a reference point to be analyzed detects an edge other than the innermost loop entrance and an edge that corresponds to the branch, and prevents the control flow from passing through the edge. The innermost branch analysis means for setting a flag of and outputting to the storage device, and reading out the hierarchical control flow graph from the storage device, analyzes whether or not each node includes the analysis target reference point, and analyzes the analysis target reference point. Of which
An inclusion relation analysis unit that sets a second flag to a node that does not include an analysis target reference point that is designated not to pass by the first flag and outputs the node to the storage device; the analysis target list from the storage device; , Reading the hierarchical control flow graph with the second flag set, analyzing the control flow connecting the user-specified reference points for analysis, outputting the control flow as a control flow list to the storage device, and Control flow analysis means for inputting a list, analyzing statements or program slices in a program on the control flow, and outputting the analysis result to a storage device as a statement information file, and the statement information file and the analysis from the storage device Read and input the target program, and on the screen of the display device, display the sentence information in the sentence information file Program analysis system; and a statement information display means for highlighting on the ram.