JP2003316611A - Program performance analysis method, execution device therefor and processing program therefor - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a technique capable of quickly and accurately recognizing the tuning point of a program and supporting the development of the program of high performance at a low cost. <P>SOLUTION: A program performance analysis method for gathering the performance information of respective functions constituting the program and analyzing the performance comprises: a step of executing the program and outputting function start information or function end information indicating the start or end of the function when the function constituting the program is started or ended; a step of outputting address information corresponding to the calling origin function of the function during execution in the case that the function start information or the function end information is outputted; and a step of storing the performance information of the function during the execution in a storage place indicated by the outputted address information. <P>COPYRIGHT: (C)2004,JPO

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a performance analysis device for analyzing the performance of a program, and particularly a technique effectively applied to a performance analysis device for collecting the performance information of each function constituting the program and performing the performance analysis. It is about.

[0002]

2. Description of the Related Art As a conventional technique related to program performance analysis, a technique described in Japanese Patent Laid-Open No. 6-259245 is known. This conventional technique is to divide a program into a group of instructions called a block and collect the operation information of each block when the program is executed to calculate and display the performance information of each function after the execution of the program.

[0003]

In the above-mentioned prior art,
In order to analyze the performance information for each function entity in the object, for example, in a program in which both functions A and B call function C, the execution time of function C depends on whether it is called from function A or from function B. It is calculated as a sum, and in which case the processing load is large cannot be determined, and in order to make such a determination, it is necessary to modify the source program.

An object of the present invention is to solve the above problems, to obtain a faster and more accurate understanding of the tuning points of a program,
It is to provide technology capable of supporting high-performance and low-cost program development.

[0005]

According to the present invention, in a performance analysis device that collects performance information of each function constituting a program and analyzes the performance, the performance information of the function is collected for each calling function. is there.

In the performance analysis apparatus of the present invention, when a source program is first compiled to generate a load module which is an executable program, function start information or function end indicating the start or end of the function that constitutes the program. An instruction to output information is embedded.

Next, the program of the execution format generated as described above is executed, and when the function constituting the program is started or ended, the function start information or the function end information is output by the instruction. Then, the address information indicating the storage location corresponding to the combination of the function being executed and the calling function that called the function is output, and the performance information of the function being executed to the storage location indicated by the output address. To store.

As described above, according to the present invention, the performance information of a function is stored in a different storage location for each caller function. Therefore, when a function is executed, which caller function is called, the execution time and the instruction are executed. It is possible to determine whether the number is large, and it is possible to develop a high-performance and low-cost program by grasping and improving the execution part when called from the calling function as a tuning point. Becomes

As described above, according to the performance analysis apparatus of the present invention, the performance information of the function is collected for each calling function, so that the tuning point of the program can be grasped more quickly and accurately, and the high performance and low cost program can be obtained. It is possible to support development.

[0010]

BEST MODE FOR CARRYING OUT THE INVENTION A performance analysis apparatus of one embodiment for collecting performance information of each function constituting a program and analyzing the performance will be described below.

FIG. 1 is a diagram showing a schematic configuration of the performance analysis apparatus of this embodiment. As shown in FIG. 1, the performance analysis device 102 of the present exemplary embodiment includes a load module execution unit 103, a table management unit 104, a performance information collection unit 106, and a performance information management unit 107.

The load module execution unit 103 executes the load module 101, which is an executable program, and when a function forming the program is started or ended, function start information indicating the start or end of the function. Alternatively, it is a processing unit that outputs the function end information to the table management unit 104.

The table management unit 104 stores the function start information or the function end information in the load module execution unit 103.
Output from the performance information management unit 107, as the address information corresponding to the caller function of the function being executed, a different address in the performance information table for each combination of the function being executed and the caller function is output to the performance information management unit 107. It is a processing unit.

The performance information collection unit 106 is a processing unit that collects performance information of a function being executed and outputs it to the performance information management unit 107. The performance information management unit 107 is a processing unit that stores the performance information output from the performance information collection unit 106 in the performance information table at the storage location indicated by the address information output from the table management unit 104.

The performance analysis device 102 includes a load module execution unit 103, a table management unit 104, and a performance information collection unit 1.
The program for functioning as 06 and the performance information management unit 107 is recorded on a recording medium such as a CD-ROM and stored in a magnetic disk or the like, and then loaded into a memory and executed. The recording medium for recording the program may be a recording medium other than the CD-ROM. The program may be installed in the information processing apparatus from the recording medium and used, or the recording medium may be accessed through the network to use the program.

In FIG. 1, 101 is a load module, 102 is a performance analysis device, 103 is a load module execution unit, 104 is a table management unit, 105 is an address conversion table, 106 is a performance information collection unit, and 107 is a performance information management unit. , 108 is a performance information table group, 109 is a function management table group, 110 is a performance information display device, and 111 is an output result.

Here, the load module is an executable object program, and when executing each function in the object program, function start information indicating that execution of the corresponding function has started, or execution of the corresponding function Indicates that the function end information that means that has ended is output. Here, the function start information must be information that can identify each function. Examples include the name of each function and the address on the virtual memory of the object program corresponding to each function (hereinafter, function address).

In the present embodiment, the mechanism for outputting the function start information or the function end information is incorporated when the load module is generated from the source program.
Here, the source program is C, C ++, FORTR.
A program written in a high-level language such as AN.

As shown in FIG. 1, the performance analysis device 102 is
Load module execution unit 103, table management unit 10
4, address conversion table 105, performance information collection unit 10
6, performance information management unit 107, performance information table group 10
8. A function management table group 109 is provided.

As shown in FIG. 1, the load module execution unit 1
03 executes the load module 101, outputs function start information or function end information during execution, and the table management unit 104 inputs function start information or function end information output during execution of the load module 101. , The information including the performance information table address in the function management table of the function is output.

FIG. 6 is a diagram showing an example of the function management table of this embodiment. Upon inputting the function start information or the function end information, the table management unit 104 refers to the function management table group 109 and refers to the function management table example 601 entry constituent element, which is the caller function name 602 and the caller function management table. Either the pair of the address 603 and the performance information table address 604 or only the performance information table address 604 is output.

The address conversion table 105 stores a correspondence relationship between the function start information and the function management table output immediately after the execution of each function.
The contents of 05 are the table management unit 1 when the function start information for the function called during the execution of the load module 101 is not registered in the address conversion table 105.
It is updated at any time in the processing of 04.

FIG. 5 is a diagram showing an example of the address conversion table 105 of this embodiment. As shown in FIG. 5, the address conversion table example 501 is a set of entries having the function name 502, the function address 503, and the function management table address 504 as constituent elements. Here, the function address 50
3 represents an address on the virtual memory of the object program corresponding to the function name 502, and the function management table address 504 represents an address of the function management table corresponding to the corresponding function.

As shown in FIG. 1, the performance information collecting unit 106
The performance information collected by the execution of the load module 101 is output. Here, the performance information refers to general numerical values related to performance such as the time required to execute the program, the total number of executed instructions, and the number of floating-point arithmetic operations.

Further, the performance information management unit 107 outputs the performance information table address output from the table management unit 104,
Also, the performance information output from the performance information collection unit 106 is input, and the performance information table of the input performance information table address is updated.

The performance information display device 110 receives the performance information table group 108 and the function management table group 109 as input, calculates the performance information of each function for each performance information table in the performance information table group 108, and outputs the output result 111. Is displayed. The above is the description of the processing in FIG.

FIG. 2 is a diagram showing an outline of a process of generating the load module 101 from the source program of this embodiment. As shown in FIG. 2, the load module generation device 20
2 includes a compiler 203 and a linker 206, and the compiler 203 internally has a function start / end information output instruction embedding unit 204. The compiler 203 and the linker 206 may be installed in the performance analysis device 102 so that the performance analysis device 102 also functions as the load module generation device 202.

As shown in FIG. 2, the compiler 203 receives the source program 201 as an input, and the intermediate object 2
05 is output. The intermediate object 205 represents a group of object programs, and each is not necessarily executable.

Function start / end information output instruction embedding unit 20
4 plays a role of embedding an instruction to output function start information or function end information in the process of the compiler 203 generating the intermediate object 205. The linker 206 receives the intermediate object 205 as an input and loads the load module 10.
1 is output. The above is the description of the processing in FIG.

FIG. 4 shows the source program 20 of this embodiment.
It is a figure showing an example of one format of No. A series of flows of the program performance analysis system will be described below based on a source program example 401 in the FORTRAN language, which is a format example of the source program 201 shown in FIG.

The source program example 401 of FIG. 4 represents an example composed of one main program and three subroutine programs. The subroutine X and the subroutine Y both call the subroutine Z.

When the source program example 401 is input to the load module generator 202, the function start / end information output instruction embedding unit 204 causes the subroutine X,
The function start information output instruction is embedded in the portion corresponding to the beginning of the Y and Z execution statements, and the function end information output instruction is embedded in the portion corresponding to the end of the execution statement.
It is output as 1. Here, the function start information is a function address.

When the load module 101 is input to the performance analysis device 102 and executed by the load module execution unit 103, the performance information of each function is collected by the performance information collection unit 106. The storage destination of the performance information of each function is determined by the table management unit 104 and the performance information management unit 107.

The flow of processing of the table management unit 104 for determining the storage location of the performance information of each function will be described with reference to FIG.

FIG. 3 shows the table management unit 104 of this embodiment.
5 is a flowchart showing the processing procedure of step S1. When the processing of the table management unit 104 is started, as shown in FIG. 3, the table management unit 104 checks in step 301 whether the input information sent from the load module execution unit 103 is function start information or function end information. If the input information is the function start information, the process proceeds to step 302.

In step 302, the function address which is the input function start information is converted into the address conversion table 10.
5 is checked, and if the function address is not registered, the process proceeds to step 303.

In step 303, after allocating the area of the function management table of the function, the function name and function address of the function and the address value of the allocated function management table are newly registered in the address conversion table 105.

As a result of checking in step 302, if the function address which is the input function start information is already registered in the address conversion table 105, step 30
Go to 4.

In step 304, the function management table address value corresponding to the function address of the input function start information is acquired from the function management table address 504 of the address conversion table 105.

In step 305, the obtained function management table address value is PUSHed into the stack area S prepared in advance by the table management unit 104. Then, in step 307, the function management table of the corresponding function management table address is set as the function management table of the function currently being executed.

On the other hand, in step 301, if the input information to the table management unit 104 is not the function start information,
That is, when the input information is the function end information, the process proceeds to step 306, POPs the value of the stack area S, and then acquires the latest value (function management table address) from the stack area S. Sets the function management table of the function management table address as the function management table of the function currently being executed.

After setting the function management table, in step 308, it is checked whether or not there are two or more values in the stack area S. If there are two or more values in the stack area S, the process proceeds to step 309. , Get the value one older than the latest.

In step 310, it is checked whether or not the acquired function management table address value exists as a calling function management table address on the current function management table. If it exists, step 312
Then, the procedure proceeds to and outputs the caller function name, caller function management table address and performance information table address of the entry.

As a result of checking in step 310, if the acquired function management table address value does not exist as the calling function management table address on the current function management table, the process proceeds to step 311.

In step 311, different areas of the performance information table are allocated for each combination of the function being executed corresponding to the current function management table and the calling function corresponding to the acquired function management table address value. Then, a new entry is added to the current function management table, and the name of the caller function, the acquired function management table address value, and the address of the assigned performance information table are called for that entry. Register to the original function name, caller function management table address and performance information table address. Then, in step 312, the registered caller function name, caller function management table address, and performance information table address are output.

On the other hand, as a result of checking in step 308, if two or more values do not exist in the stack area S, O
The process proceeds to step 313 assuming that it is the MAIN process called from S.

In step 313, the performance information table address registered on the current function management table is read. If the performance information table address is not registered in the current function management table, the area of the performance information table for MAIN called from the OS is allocated, and then a new entry is added to the current function management table. After registering the assigned performance information table address in the performance information table address of the entry, the performance information table address is used. Then, in step 312, the performance information table address is output.

5 and 6 respectively show an example of the address conversion table 105 and an example of the function management table of the function Z accumulated by the processing of the table management unit 104 described above. Since the source program example 401 is composed of four functions, the number of entries in the address conversion table 105 is four. Since there are only two caller functions, function X and Y, of function Z, the number of entries in the function management table is also two.

When the performance information table address is output from the table management unit 104, the performance information management unit 107 switches the performance information tables, and the performance information collection unit 10
A process of storing the performance information output from No. 6 in the performance information table after switching is performed. In addition, performance information tabulation processing is also performed at the time of switching.

For example, after the function X is called from MAIN, the function Z is called from the function X, and the table management unit 1
When “ZX” of the performance information table address 604 is output from 04, the performance information of the function X having the call source of MAIN is aggregated, and the time required to execute the function X, the total number of executed instructions, and the number of floating point arithmetic operations Etc. are calculated and stored in the performance information table corresponding to the combination of MAIN and function X, and then the output performance information table address value “Z
The performance information table is switched to the area of “X”, and the performance information of the function Z that calls the function X is stored in the performance information table of “ZX”.

FIG. 7 is a diagram showing an example of the output result of the performance information for each caller function of this embodiment. The performance information display device 110 totalizes the performance information stored in each performance information table as described above based on the information in the function management table, and outputs the performance information for each calling function in FIG. 7 as an output result example. It is displayed as 701. In FIG. 7, the information of the function management table example 601 of the function Z is reflected, and the performance information of the function Z is displayed separately for the function X and the function Y that are the calling functions of the function Z.

As described above, according to the performance analysis apparatus of this embodiment, since the performance information of the function is collected for each calling function, the tuning point of the program can be grasped more quickly and accurately, and high performance and low cost can be obtained. It is possible to support the program development of.

[0053]

According to the present invention, since the performance information of a function is collected for each calling function, it is possible to support a quick and accurate grasp of the tuning point of the program and the development of a high-performance and low-cost program. Is.

[Brief description of drawings]

FIG. 1 is a diagram showing a schematic configuration of a performance analysis apparatus of the present embodiment.

FIG. 2 is a diagram showing an outline of a process of generating a load module 101 from a source program of this embodiment.

FIG. 3 is a flowchart showing a processing procedure of a table management unit 104 of this embodiment.

FIG. 4 is a diagram showing an example of a format of a source program 201 of this embodiment.

FIG. 5 is a diagram showing an example of an address conversion table 105 of this embodiment.

FIG. 6 is a diagram showing an example of a function management table of the present embodiment.

FIG. 7 is a diagram showing an example of an output result of performance information for each caller function of the present embodiment.

[Explanation of symbols]

101 ... Load module, 102 ... Performance analysis device, 1
05 ... Address conversion table, 108 ... Performance information table group, 109 ... Function management table group, 110 ... Performance information display device, 111 ... Output result, 103 ... Load module execution unit, 104 ... Table management unit, 106 ... Performance information collection Section, 107 ... Performance information management section, 201 ... Source program, 202 ... Load module generation device, 203 ...
Compiler, 204 ... Function start / end information output instruction embedding section, 205 ... Intermediate object, 206 ... Linker, 4
01 ... Source program example, 501 ... Address conversion table example, 502 ... Function name, 503 ... Function address, 50
4 ... Function management table address, 601 ... Function management table example, 602 ... Caller function name, 603 ... Caller function management table address, 604 ... Performance information table address, 701 ... Output result example.

   ─────────────────────────────────────────────────── ─── Continued front page    (72) Inventor Shinichi Tanaka             5030 Totsuka Town, Totsuka Ward, Yokohama City, Kanagawa Prefecture             Ceremony company Hitachi Ltd. software division F-term (reference) 5B042 HH20 MC24 MC33

Claims (5)

[Claims] 1. A program performance analysis method for collecting performance information of each function constituting a program and analyzing the performance thereof, when a program is executed and a function constituting the program is started or ended. When outputting the function start information or the function end information indicating the start or end of the function, and when the function start information or the function end information is output, the address information corresponding to the calling function of the function being executed is displayed. A program performance analysis method comprising: a step of outputting and a step of storing performance information of a function being executed in a storage location indicated by the output address information. 2. An address of a performance information table that differs for each combination of the function being executed and its calling function is output as the address information, and the performance information storage destination is switched to a different performance information table for each calling function. The program performance analysis method according to claim 1, wherein the performance information of each function is generated for each correspondence with the calling function. 3. The stored performance information is displayed for each calling function.
The program performance analysis method described in any one of. 4. A performance analyzer that collects performance information of each function that constitutes a program and analyzes the performance thereof when the program is executed and the function that constitutes the program is started or ended. A load module execution unit that outputs function start information or function end information indicating the start or end of a function, and an address corresponding to the caller function of the function being executed when the function start information or function end information is output. A performance analysis apparatus comprising: a table management unit that outputs information, and a performance information management unit that stores performance information of a function being executed in a storage location indicated by the output address information. 5. A program for causing a computer to function as a performance analysis device that collects performance information of each function that constitutes a program and analyzes the performance of the program, starts the function that constitutes the program, or executes the program. A load module execution unit that outputs function start information or function end information that indicates the start or end of the function when the end is performed, and a function that is being executed when the function start information or function end information is output. Of the table function that outputs the address information corresponding to the calling function of the computer and the performance information management unit that stores the performance information of the function being executed in the storage location indicated by the output address information. Characteristic program.