JP3206535B2 - Measured value display method of performance measurement tool - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a measurement value display method of a performance measurement tool, and more particularly, to a performance measurement tool for efficiently performing performance analysis and tuning at a source code level in a translation and execution environment where optimization is involved. It relates to a measurement value display method.

[0002]

2. Description of the Related Art Generally, a performance measurement tool has a function of measuring a performance value such as an execution time of a specific structure in a program and displaying the result in a table or a graph format.
The specific structure to be measured is often a routine, a loop, a basic block, or the like in consideration of the usefulness of the measured performance value and the efficiency of the measurement.

An example of a performance measurement tool for measuring execution time is disclosed in Japanese Patent Laid-Open No. Hei 5-20135. According to the invention of this publication, at the time of compilation, a function is decomposed into basic blocks that do not include branches, an execution count addition code is embedded in each basic block, and at execution, the execution count for each basic block is measured from the accumulated value and calculated in advance. The CPU time per one time of each basic block is multiplied by the measured number of executions, and the execution time for each basic block is calculated and displayed.

In addition to this, a time measurement code is inserted before and after a loop or a routine, a performance measurement tool for directly measuring the execution time of the structure, an interrupt is generated at regular intervals during execution, and There is generally known a performance measurement tool for estimating the performance of each routine from the distribution state of the addresses being executed.

[0005] All of the conventional performance measurement tools described above measure the performance of a structure in the measured object code.
This is a method of displaying the data as it is in correspondence with the structure in the object code.

[0006]

However, according to the above-mentioned conventional display method for displaying the measurement result as it is for the measured structure in the object code, the display of the measurement result is difficult for a user with a recent compiler that performs optimization. There is a problem that it is difficult to understand.

That is, since recent compilers perform complicated program transformation, the structure in the source code understood by the user and the structure in the object code actually measured and displayed are one-to-one. Not often. Therefore, in order for the user to know the performance of each part of his or her source code from the displayed numerical values, it is necessary to perform conversion based on the optimization history and the like, which is troublesome and requires advanced knowledge on optimization. Required.

In view of the above problems, an object of the present invention is to reduce a measurement result performed on a structure in object code to a structure in source code regardless of the degree of optimization performed by a compiler. To display to the user.

[0009]

According to a first aspect of the present invention, there is provided a performance measuring tool for displaying a measured value, wherein a source code is translated into an object code, a measuring code is embedded, and an execution time of a structure in the object code is measured. In the measured value display method, an instruction CPU time table in which CPU time required to execute each instruction of hardware for executing an object code is described in advance, and each instruction in the object code created when the object code is generated is included in the source code. A code correspondence table describing which part of the structure has the origin, a measured value distribution table describing a ratio of allocating the measured execution time of the structure in the object code to the corresponding structure in the source code, and the code The structure in the object code as the unit of measurement based on the correspondence table and the instruction CPU time table CPU time for each instruction that is present within
And when the structure of the source code is the origin of those instruction have multiple divided Ri equally oscillation in their respective structures, by taking the ratio of the integrated values integrated in each corresponding structure in the source Sukodo A measurement value distribution calculating means for calculating a distribution ratio and describing the measurement value distribution table in the measurement value distribution table; and executing the execution time of the structure in the object code measured by executing the object code in the source code based on the measurement value distribution table. Execution time display means for distributing and displaying a plurality of structures.

According to a second aspect of the present invention, there is provided a method for displaying a measured value of a performance measurement tool, comprising the steps of: translating a source code and generating an object code in which a measurement code is embedded. Comprises the instruction CPU time table and the measured value distribution calculating means, outputs the measured value distribution table and the object code in which the measurement code is embedded, and the performance display processing unit inputs the measured value distribution table when executing the measurement. The display is performed by the execution time display means.

According to a third aspect of the present invention, there is provided a method for displaying a measured value of a performance measuring tool according to the second aspect of the present invention, wherein the compiler unit includes a plurality of instruction CPU time tables corresponding to different execution computers. And selecting one of them to output the measured value distribution table.

According to a fourth aspect of the present invention, there is provided a method for displaying a measured value of a performance measuring tool, wherein the compiler section translates a source code and generates an object code in which a measuring code is embedded. Comprises the measured value distribution calculating means, reads the instruction CPU time table from an external storage device, and outputs an object code in which the measured value distribution table and the measurement code are embedded. It is characterized in that a value distribution table is inputted and displayed by the execution time display means.

According to a fifth aspect of the present invention, there is provided a method for displaying a measured value of a performance measurement tool, comprising the steps of: translating a source code and generating an object code in which a measurement code is embedded. Outputs the code correspondence table and the object code in which the measurement code is embedded, and the performance display processing unit includes the instruction CPU time table and the measurement value distribution calculating means, and performs the measurement by using the code correspondence table when executing the measurement. A value distribution table is created and displayed by the execution time display means.

[0014]

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a block diagram showing the configuration of the first embodiment of the present invention. In the first embodiment of the present invention, as shown in FIG. 1, a compiler 1 that inputs a source code 5 and outputs a measured value distribution table 3 and an object code 6 into which the measured code is inserted, And outputs the performance data 7 measured in the structural unit in the object code, and the measurement in which the performance data 7 is input and converted into the structural unit in the source code with reference to the measured value distribution table 3. The performance display tool 2 includes an execution time display unit 21 for displaying data on the display 9 and outputting the data as the performance display list 8.

The compiler 1 includes a syntax analyzer 11 for analyzing the syntax of the source code 5 and an object for generating an object code 6 by performing optimization based on the processing result of the syntax analyzer 11 and creating a code correspondence table 13. A code generation unit 12, a measurement code embedding unit 15 for inserting a measurement code into a generated object code, and an instruction CPU time table 14 in which CPU time required for executing each instruction of the execution computer 4 is registered in advance. From the code correspondence table 13 and the instruction CPU time table 14
Measurement value calculation means 16 for calculating the ratio of allocating the measurement data of the structure in the object code measured in the above to the corresponding structure in the source code and outputting the measurement value distribution table 3
And a control unit 17 for controlling these.

Hereinafter, the operation of the first embodiment configured as described above will be described in detail. The source code 5 input to the compiler 1 is sent to the object code generation unit 12 together with the analysis result by the syntax analysis unit 11. The object code generation unit 12 performs optimization processing based on the analysis result, and calls the measurement code embedding unit 15 when the object code is generated.

The measurement code embedding means 15 recognizes a structure serving as a measurement unit such as a loop existing in the generated object code, and embeds a measurement code for performing timekeeping processing before and after the recognized structure. Thereby, the object code 6 in which the measurement code is inserted is generated,
It is possible to measure the execution time of the structure in the object code at the time of execution.

Next, the object code generation unit 12
The code correspondence table 13 is generated based on the analysis information used for generating the object code. Code Correspondence Table 13
Means that each instruction included in the structure in the object code is
This is a correspondence table that describes which part in the source code has the origin, and describes the relationship between the structure in the source code and the structure in the object code that no longer corresponds one-to-one due to optimization. It was done. When the creation of the code correspondence table 13 is completed, the object code generation unit 12 calls the measured value distribution calculation unit 16.

The measurement value distribution calculation means 16 acquires the CPU time of each instruction existing in the structure in the object code as one measurement unit from the instruction CPU time table 14,
With reference to the code correspondence table 13, the multiplication is added to the structure in the source code which is the display unit from which the instruction originates. On this occasion,
If one instruction originates from a plurality of instructions in the source code, the instructions are equally divided into a plurality of structures and integrated. When all instructions have been integrated, the ratio of the execution time of the measured structure in the object code to the corresponding multiple structures in the source code, by taking the ratio of the integrated value between the structures in the source code Is calculated, and the distribution ratio is recorded in the measured value distribution table 3 and output.

FIG. 2 is a flowchart showing the operation of the measured value distribution calculating means 16. When activated, the structure in which the measurement code is embedded is detected from the object code (step S1), and the processing is started (step S1).
2). If there is an unprocessed statement in the structure (step S
3), the statement is read out, and the CPU time corresponding to the instruction is obtained from the instruction CPU time table 14 (step S4). It is determined whether there is more than one (step S5). As a result, if there is a plurality, the CPU time obtained in step S4 is equally distributed to a plurality of corresponding structures in the source code and added (step S6), and if not, the CPU time is added to the corresponding structure as it is (step S6). S7). The processes in steps S3 to S7 are repeated until there is no sentence in the structure. When a measurement code indicating the end of the structure to be measured is detected and no sentence is found, the process proceeds to step S8 to calculate the integrated value between the structures in the source code. The distribution ratio is calculated by calculating the ratio, described in the measured value distribution table 3, and the process returns to step S1. By repeating the above operation until there is no unprocessed structure, the measurement value distribution table 3 is created for all the measurement target structures in the object code.

Using the object code 6 created by the compiler 1 and the measured value distribution table 3 as described above, the execution computer 4 measures the structure of the object code as a unit of measurement and outputs it as performance data 7. The execution time value is stored in the execution time display unit 2 of the measurement display tool 2.
By means of 1, it is reduced to the corresponding structure in the source code and displayed. As a result, the performance value (execution time) for each structure such as loops and routines in the source code 5 is displayed on the display 9 in a format that is easy for the user to understand and has high utility value, and is output as the performance display list 8.

Next, the operation of the above-described first embodiment will be described with a specific example. Here, Compiler 1 is For
This is a tran language compiler that inputs the source code 5 shown in FIG.
It is assumed that the object code 6 shown in FIG. Note that the instruction CPU time table 14 is given as shown in FIG. 3 corresponding to the execution computer 4.

With respect to the input of the source code 5 shown in FIG. 4A, the compiler 1 optimizes the fusion of the loop 1 starting from the third line and the loop 2 starting from the sixth line. I do. As a result, the object code 6 output by the object code generation unit 12 is
(B). The fourth to tenth lines in FIG. 4B form one loop (loop P). This loop P is generated as a result of the fusion of loop 1 and loop 2 in the source code. It is a thing. Since the measurement unit is a loop, f_loopbegin () and f_loopbegin () are routines for performing measurement by the measurement code embedding unit 15.
_Loopopen () call statement is loop P
Are inserted immediately before and after.

The source code 5 shown in FIG.
In addition to generating the object code 6 in (b), the object code generation unit 12 generates a code correspondence table 13 as shown in FIG. The instruction of each line of the object code 6 indicates which line of the source code 5 corresponds to the instruction.

The measurement value distribution calculation means 16 sequentially processes the statements on each line in the loop P of FIG. 4B. FIG.
The sentence on the fourth line in FIG.
Since it corresponds to both the third and sixth lines in (a), it originates in both loop 1 and loop 2. 4 in FIG.
The CPU time in the row is the instruction CPU time table 14 shown in FIG.
Since 5 で あ 2, 5 ÷ 2 = 2.5 is added to both the loop 1 and the loop 2. The next sentence in the fifth and sixth lines in FIG. 4B is a sentence generated from the fourth line in FIG. FIG.
Since the total CPU time corresponding to the fifth and sixth lines in (b) is 10 + 20 = 30 from the instruction CPU time table 14 in FIG. 3, 30 is added for loop 1. Similarly, FIG.
The sentences on the seventh and eighth lines in (b) are shown in FIG.
This is a sentence generated from the seventh line of (a), and both are loop 2
You can see that it has origin. The sum of the CPU time in the seventh and eighth lines in FIG.
Since 20 = 60, 60 is added to loop 2. The sentence on the ninth and tenth lines in the following FIG. 4B has the origin in both the loop 1 and the loop 2 as in the fourth line, so that 5 ÷ 2 =
2.5 is added to both loop 1 and loop 2. As described above, a total of 35 is added to the loop 1 and 65 is added to the loop 2. Therefore, the ratio at which the execution time of the loop P in FIG. 4B is distributed to the loop 1 and the loop 2 is 35: 65 = 0.35: 0.65, and the measured value distribution calculating means 16 calculates this ratio. It is recorded in the measured value distribution table 3 as shown in FIG.

Next, assuming that the execution time of the loop P is 10 seconds as a result of executing the object code 6 by the execution computer 4, the execution time display means 21
Seconds are allocated to Loop 1 and Loop 2 according to the distribution ratio described in the measured value distribution table 3, and the execution time of Loop 1 is set to 3.
Output to the performance display list 8 or display on the display 9 assuming that the execution time of the loop 2 is 6.5 seconds for 5 seconds.

In the description of the first embodiment, it is assumed that the instruction CPU time table 14 corresponding to the hardware of the execution computer 4 for performing the measurement is provided in the compiler 1. However, since the instruction CPU time table needs to be changed correspondingly for different execution computers,
By reading from the external storage device the one corresponding to the execution computer used for measurement, it is possible to support a plurality of execution computers and the adaptability is improved. A similar effect can be obtained even if a plurality of instruction CPU time tables corresponding to different execution computers are provided in the compiler so that any one can be selectively used.

FIG. 5 is a block diagram showing the configuration of the second embodiment of the present invention. In the second embodiment of the present invention shown in FIG. 5, a compiler 1a that inputs a source code 5 and outputs an object code 6 into which a measurement code is inserted and a code correspondence table 13a, and executes an object code 6 to execute an object structure The execution computer 4 that outputs the performance data 7 measured in units, the performance data 7 is input, and the measured value distribution table 23 is created from the code correspondence table 13a and the object code 6 and converted into the structural unit of the source code. A performance display tool 2a for displaying measured data on a display 9 and outputting the data as a performance display list 8.

The difference from the first embodiment shown in FIG. 1 is that the compiler 1a comprises a syntax analysis unit 11, an object code generation unit 12a, a measurement code embedding unit 15, and a control unit 17a. Outputting the object code 6 and the code correspondence table 13a without including the means,
When the execution computer 4 executes the object code 6, the measured value distribution calculating means 22 of the performance display tool 2a creates the measured value distribution table 23 using the instruction CPU time table 24, and displays the execution time using this. This means that the means 21 performs display by reducing to the structure in the source code. Therefore, there is an advantage that the object code 6 and the code correspondence table 13a created at the time of compiling can be used as they are even if the execution computer 4 is different.

[0031]

As described above in detail, according to the measured value display method of the performance measuring tool of the present invention, even when the compiler deforms a complicated program, the measured execution of the structure in the object code is performed. Since the time is distributed and displayed in the structure in the source code understood by the user, there is an effect that the measurement result is easy to understand and use for the user.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a first exemplary embodiment of the present invention.

FIG. 2 is a flowchart showing a processing procedure of a measured value distribution calculating means of FIG. 1;

FIG. 3 is an explanatory diagram showing a data configuration example of an instruction CPU time table of FIG. 1;

FIG. 4 is an explanatory diagram showing a specific example of the operation of the first embodiment of FIG. 1;

FIG. 5 is a block diagram showing a configuration of a second exemplary embodiment of the present invention.

[Explanation of symbols]

 1, 1a Compiler 2, 2a Performance display tool 3, 23 Measurement value distribution table 4 Execution computer 5 Source code 6 Object code 7 Performance data 8 Performance display list 9 Display 11 Syntax analyzer 12, 12a Object code generator 13, 13a Code correspondence table 14, 24 Instruction CPU time table 15 Measurement code embedding means 16, 22 Measured value distribution calculation means 17, 17a Control unit 21 Execution time display means

──────────────────────────────────────────────────続 き Continued on the front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G06F 11/28-11/36 G06F 9/06-9/45

Claims (5)

(57) [Claims] In a measurement value display method of a performance measurement tool for translating a source code into an object code, embedding a measurement code, and measuring an execution time of a structure in the object code, each instruction of hardware for executing the object code is executed. An instruction CPU time table in which the CPU time required for execution is described in advance, and a code correspondence table that is created when the object code is generated and describes where in the source code each instruction in the object code originates are measured. Distribution table which describes a ratio of allocating the execution time of the structure in the object code to the corresponding structure in the source code, the code correspondence table and the instruction C
The CPU time for each instruction that is present in the structure in the object code to be measured unit based on PU timetable, equally on each of their structure when the structure of the source code is the origin of those instruction there are multiple and Ri divided vibration measurements, and corresponding measurements allocation calculation means for describing the allocation ratio calculated the measured value allocation table by taking the ratio of the integrated and the integrated value in each structure in the source Sukodo, by the execution of the object code Execution time display means for distributing and displaying the execution time of the structure in the object code obtained based on the measured value distribution table to a plurality of corresponding structures in the source code. Measurement value display method. 2. A compiler section for translating a source code and generating an object code in which a measurement code is embedded includes the instruction CPU time table and a measurement value distribution calculating means, wherein the measurement value distribution table and the measurement code are embedded. 2. A method for displaying a measured value of a performance measurement tool according to claim 1, wherein an object code is output, and a performance display processing unit inputs the measured value distribution table and displays the measured value distribution table when the measurement is executed. . 3. The method according to claim 2, wherein said compiler unit comprises a plurality of instruction CPU time tables corresponding to different execution computers, and selects one of them to output said measurement value distribution table. Measurement value display method of the performance measurement tool. 4. A compiler section for translating a source code and generating an object code in which a measurement code is embedded includes the measurement value distribution calculation means, reads the instruction CPU time table from an external storage device, and reads the measurement value distribution table. 2. The performance measurement device according to claim 1, wherein an object code in which the measurement code is embedded is output, and a performance display processing unit inputs the measurement value distribution table and displays the measurement value distribution table when the measurement is executed. Tool measurement value display method. 5. A compiler for translating a source code and generating an object code in which a measurement code is embedded outputs the code correspondence table and the object code in which the measurement code is embedded, and a performance display processing unit outputs the instruction CP.
2. The apparatus according to claim 1, further comprising a U time table and a measurement value distribution calculating means, wherein the measurement value distribution table is created using the code correspondence table at the time of measurement execution, and is displayed by the execution time display means. Measurement value display method of the performance measurement tool.