JP5042487B2 - Computer operating method, program, computer - Google Patents

Description

  The present invention relates to an interpreter that interprets and executes a program written in a source program or an intermediate language, and in particular, a method for interpreting and executing a source program language or an intermediate language into a sequentially optimized object program language, and The present invention relates to a technique for improving the execution speed of an interpreter by an optimization method using a method of executing a program by using a direct execution mechanism provided on the hardware side of a computer.

Hereinafter, an intermediate language that can be interpreted by an interpreter positioned between the source program language and the machine language is referred to as “IR” (Internal Representative).
When executing a computer program comprising a source program language (high-level language) and an IR sequence, the source program language and IR are sequentially translated into an optimized object program language (machine language, etc.) and executed. The system is called “JIT” (Just In Time Compiler).
On the other hand, a hardware mechanism built in a computer that directly executes a source program language or IR program is referred to as “HAE” (Hardware Accelerate Execution).

  Conventionally, when a programmer using an interpreter creates a computer program, an interpreter language (more specifically, a source program language or an IR positioned between the language and the machine language) is arranged. Write down as The interpreter sequentially interprets and executes the interpreter language (sequentially interprets and executes the machine language). In other words, the interpreter system is originally configured to execute the source program language itself or the IR one step before the language is converted into a machine language while sequentially interpreting it.

  However, comparing the interpreter system with a compiler system that compiles (translates) and executes an ordinary compiler language, the interpreter system has many advantages specific to the interpreter system, such as being able to execute in real time. However, it has many disadvantages. The greatest of these disadvantages is the slow execution speed as the fate of the sequential interpretation (interpretation) method.

  Therefore, in order to increase the execution speed while leaving the advantage of the interpreter system that can be executed in real time, the above-mentioned JIT method and the above-described interpreter method incorporating the HAE ( The following two methods have been developed, and to date, both methods have been developed individually.

  However, each of these methods has advantages and disadvantages, and has not been a satisfactory system. The general advantages and disadvantages of the JIT method and the HAE method will be summarized below.

  First, as a merit of the above-mentioned JIT method, it can be mentioned that it has a very great effect on a loop executed frequently and a processing step in which repeated processing exists.

  However, as a disadvantage of the JIT method, there is a processing step in which an interpreter (corresponding to compiler translation) is converted into an object program that can be executed by a computer such as a microcomputer at the time of execution. Necessary. For this reason, in particular, immediately after the execution program is started, there are many uninterpreted execution steps, so the execution speed becomes slow.

  Further, in addition to the memory for storing the source program, an extra memory for storing the converted object program (generally an execution program in a machine language format) is required. In general, the size of the object program increases as the means (hereinafter referred to as “optimization means”) prepared by the interpreter for extracting high-speed performance becomes more sophisticated. The more program parts that are targeted, the larger the tendency.

  Furthermore, even if the minimum optimization means is implemented, if the optimization is performed more effectively to improve the speed performance, the target of the optimization means in the program written in the interpreter language An unevenly distributed part (that is, a part where a series of hierarchical processing procedures targeted by the optimization means are concentrated and many times are repeatedly executed, (Referred to as “Hotspot”), and for the discovery and determination of this Hotspot, a certain amount of processing time is required when executing the program.

  On the other hand, as a merit of the above-mentioned HAE method, it is known that generally less power is required than the JIT method. In addition, since initial investment (that is, execution of interpretation processing for realizing the JIT system) is unnecessary, it can be expected that an average high execution speed can be obtained immediately after the start of execution of the program. Further, it can be executed directly only by a source program or IR, and therefore, an object program as in the case of the JIT method is not required, and therefore an amount of memory for storing the object program is also unnecessary.

  However, as a demerit of the HAE method, an additional mechanism is required for computer hardware such as a microcomputer, and the equipment cost is increased accordingly. Therefore, normally, not all of the source program language and IR are directly executed, but only a part of them is still executed directly.

  The peak performance when the source program and IR are viewed in micro units (more specifically, when viewed in instruction units) can exceed the JIT method, but the HAE method can also be optimized globally. Unlike the method, the sequence of program steps cannot be targeted for optimization, so the JIT method is better especially for high-load program parts that affect the overall execution speed performance of the program. Yes. Further, if software processing is entered midway, it is necessary to stop execution by the hardware mechanism and return to execution by software (ie, interpreter program). Examples of such software processing include garbage collection (GC), interrupt processing, and exception processing.

  Regarding a conventional interpreter, Patent Document 1 discloses a method and apparatus that allows a user to effectively acquire debug information while minimizing a decrease in speed performance and an increase in memory consumption accompanying generation of debug information. Is disclosed.

JP 2004-54807 A

  In the conventional interpreter described in the background art, for example, in the case of an interpreter based on the JIT method (generally equipped with an optimization process), if an attempt is made to convert the entire source program or IR into an object program, Improve this point and apply optimization methods only to the above-mentioned Hotspot, as realistic memory space and interpreting processing capability are required and the performance effect will not increase as expected. This time, there was a problem that it took a lot of processing time to find this Hotspot.

  Anyway, as long as the hotspot mentioned above can be found, the interpreter can exert its maximum effect even if it is the minimum interpretation target. However, in order to find this Hotspot, the interpreter program must first statically find the program structure that is likely to have the effect of the JIT method, and then look at the result of moving the program to some extent and dynamically I only have the means to confirm it. In other words, if it can be statically confirmed that a certain processing procedure having a source program structure that is likely to have the effect of the JIT method is a processing procedure that is frequently called, this processing procedure is temporarily optimized. If the frequency of operation of this processing procedure is verified at the time of execution, this processing procedure can be considered as a hot candidate for Hotspot to which optimization means should be applied. If there is, it can be confirmed. For example, the number of loops in a loop structure that is layered in layers is only known dynamically, and as a result of running a program to some extent, the execution of this loop structure can be performed a sufficiently large number of times. May be repeated.

  The Hotspot candidates to which the optimization means found in this way should be applied are converted into objects by applying the optimization means of the interpreter, and the object is executed from the next execution. Realizes faster execution speed. However, to execute an object program with improved speed performance by applying optimization means, of course, it is determined whether a series of processing procedures is a Hotspot candidate to which optimization means should be applied. This requires another processing time, that is, it is necessary to carry out processing such as extra evaluation and judgment unrelated to the original program operation. This is also a problem of the conventional interpreter.

  Further, as is clear from the above, with respect to processing procedures (for example, processing procedures with a non-complex hierarchical structure) that have not been determined as Hotspot candidates to which the optimization means are applied based on the above determination, there is actually some degree. Even if it has an execution frequency, the optimization means is not applied for efficient use of the memory space, and such a part is executed as an object program having no performance. become.

  On the other hand, the HAE method cannot be generally described because the technology is related to hardware design, but generally has the disadvantages described above. In other words, in addition to the extra hardware cost, the speed performance also has a complicated hierarchical structure, and the processing procedure has a large number of iterations (the number of loops). The speed performance of the JIT interpreter provided with the optimization means cannot be exceeded. Furthermore, if an interrupt or the like that requires processing on the software side occurs during task execution, it is naturally impossible to continue execution with the hardware mechanism alone, so continued execution of the task by the hardware side Must be stopped and the control flow left to the software side.

  Note that in order to execute the process of collecting the memory area occupied by unused object programs and data (that is, garbage collection), all threads must be stopped. It is no exception whether the task is executed on the software side or the hardware side. However, when executing on the hardware side, since there is a HAE execution start / stop instruction on the hardware side, the processing cost for stopping is higher than when executing on the software side. Resulting in. For this reason, when executing a processing procedure that tends to generate garbage collection at a high frequency during execution, execution on the software side (that is, execution by a general interpreter having no JIT method or optimization means) It may be better to leave it up.

  When an external interrupt occurs, it may be necessary to take control flow directed to the task and transfer the control flow to the interrupt processing routine.

  In addition, in the HAE method, an arbitrary task can generally be operated at a high speed. However, if a certain task is a task having a processing procedure of a nature that frequently generates an interrupt, As described above, there are cases where it is better to leave the entire task or execution of the processing procedure to the software side. In the case where a program is executed by a conventional HAE interpreter, the judgment around this is not fed back well.

  The present invention has been made in view of the above-described conventional problems, and by interpreting information necessary for optimization statically and dynamically, an interpreter (JIT system) provided with optimization means, Provide an interpreter that can improve the execution speed by incorporating only the advantages of the hardware direct execution method (HAE method) and taking into account the characteristic processing of individual applications (interrupt processing procedures, etc.) The purpose is to do.

  Another object of the present invention is to collect information necessary for the optimization statically and dynamically, so that an interpreter provided with optimization means (JIT method), a hardware direct execution method (HAE method), and It is an object of the present invention to provide an interpreter execution speed improvement technique capable of improving the execution speed by taking into account only the advantages of the above and taking into consideration the characteristic processing (interrupt processing procedures, etc.) of individual applications.

  The present invention has been made in view of the above-described problems of the conventional interpreter. A program written in a high-level language or an intermediate language is partially fetched sequentially, and the fetched program portion is arranged in a machine language sequence. An interpreter that includes a sequential execution unit that is executed while being converted, and that is executed by a computer to which a hardware mechanism capable of directly executing a part of the language is added. Optimization means capable of converting a program portion into a sequence of machine languages optimized in terms of execution speed, and the sequential execution means or the optimization means via the optimization means as an execution system of the program portion Execution system selection means for selecting one of sequential execution means or the hardware mechanism; and The Mareta program portion, there is provided an interpreter, characterized in that it and a means for executing the selected executing by said execution system selection means.

  Further, the interpreter includes an application exclusion determination unit that determines whether or not the partially fetched program part includes more than a predetermined number of instructions that cannot be executed by the hardware mechanism; and the application exclusion determination Means for causing the program part to be executed by the sequential execution means when it is determined that the program part includes more than a predetermined number of instructions that cannot be executed by the hardware mechanism; and Second execution for causing the program part to be executed by the hardware mechanism when the application exclusion determining unit determines that the program part does not include more than a predetermined number of instructions that cannot be executed by the hardware mechanism. And means for said partially captured program part after said execution The optimization determining means for determining whether or not the optimization means has achieved an effect that exceeds a predetermined standard, and the optimization determination means provides an effect that exceeds the predetermined standard by the optimization means. The program part determined to have been executed is provided with third execution means for executing the program part by the sequential execution means via the optimization means at the next and subsequent executions.

  Further, the interpreter verifies whether or not the program part generates more executions of instructions that cannot be executed by hardware when the number of executions of the program part reaches a predetermined number. If the program part causes the execution of an instruction that cannot be executed by hardware more than a predetermined number of times by the application exclusion verification means and the application exclusion verification means, And fourth execution means for causing the program part to be executed by the sequential execution means.

  Further, the interpreter executes the processing by the system excluding the application exclusion determination unit, the optimization determination unit, and the application exclusion verification unit from the system when the execution exceeds the predetermined number of times. .

  In addition, the program portion whose effectiveness of the optimization is verified by the optimization means is a portion in which the number of internal loops is a predetermined number or more, or a portion including a nested structure exceeding a predetermined depth. It is characterized by. Here, the nested structure refers to a program structure including conditional branching (if statement), a certain number of repetitions (for statement), conditional repetition (while statement), and the like.

  In the interpreter, the optimization determination unit corrects the estimation of the optimization effect by the optimization unit according to the degree of power saving desired by the user. For example, when the degree of power saving is greater than a predetermined level, the optimization effect by the optimization means is estimated to be less than actual.

  Further, the present invention provides a computer having a hardware mechanism capable of directly executing a part of a program written in a high-level language or an intermediate language. An operation method of an interpreter comprising sequential execution means for execution while converting a machine language sequence, wherein the fetched program portion is optimized to convert a machine language sequence optimized in terms of execution speed And an execution system selection step of selecting one of the sequential execution step, the sequential execution step via the optimization step, or the hardware mechanism as an execution system of the program part And the captured program part is selected by the execution system selection step. There is provided a method of operating an interpreter, characterized in that it has a an execution step of executing the system.

  Further, in the operation method of the interpreter, it is determined whether the partially fetched program part includes more than a predetermined number of instructions that cannot be executed by the hardware mechanism, and the program part includes If it is determined that the number of instructions that cannot be executed by the hardware mechanism is greater than a predetermined number, the sequential execution means executes the program part, and the program part has a predetermined number of instructions that cannot be executed by the hardware mechanism. If it is determined that the program part is not included, the program part is executed by the hardware mechanism, and after the execution, the optimization is performed in the optimization step for the partly taken-in program part. To determine whether an effect exceeding For the program part determined to have achieved the effect exceeding the predetermined standard by the optimization in the optimization step, the program part is executed by the sequential execution means through the optimization step at the next and subsequent executions. Features.

  Further, in the operating method of the interpreter, when the number of executions of the program part reaches a predetermined number, the program part generates more than a predetermined number of instructions that cannot be executed by hardware. And when the program part causes the execution of instructions that cannot be executed by hardware more than a predetermined number of times, the program part is executed by the sequential execution means during execution exceeding the predetermined number of times. It is made to perform.

  In the operation method of the interpreter, when the execution exceeds the predetermined number of times, the processing by the application exclusion determination step, the optimization determination step, and the application exclusion verification step is omitted.

  Further, in the interpreter operating method, in the optimization step, the program portion whose effectiveness of the optimization is verified has a portion in which the number of internal loops is a predetermined number or more, or a nested structure exceeding a predetermined depth. It is the part which contains. Here, the nested structure refers to a program structure including conditional branching (if statement), a certain number of repetitions (for statement), conditional repetition (while statement), and the like.

  In the operation method of the interpreter, the optimization determination step refers to a power saving level set in advance by a user, and estimates an optimization effect by the optimization unit according to the power saving level. It is characterized by correcting.

  As described above, according to the interpreter of the present invention, a program is fetched in units of methods, and a method that includes many instructions that cannot be processed by HAE, while applying the HAE method to the fetched method. Excludes the application of the HAE method, and for the method in which Hotspot is found by the first execution, the JIT method having the optimization means is applied from the next execution, so that an interpreter with a high execution speed is provided. be able to.

  For methods that include many instructions that cannot be processed by HAE, the method that has been found to have caused the execution of instructions that cannot be processed by hardware alone after executing a sufficient number of executions, while excluding the application of the HAE method. Since the application of the HAE method is excluded, it is possible to provide an interpreter that can suppress the occurrence of a phenomenon that the execution speed decreases even when the HAE method is applied.

  Also, after execution of a sufficient number of executions, statistical processing program steps for optimization means, etc. are removed, so that an overhead is eliminated, an execution speed is high, and an interpreter with a small amount of required memory is provided. Can do.

  Further, when the degree of power saving desired by the user is large, the application of the JIT method is excluded, so that it is possible to provide an interpreter capable of saving power consumption in a mobile device or the like.

  The best mode for carrying out the interpreter of the present invention will be described below in detail with reference to the accompanying drawings. 1 to 7 are diagrams illustrating embodiments of the present invention. In these drawings, the same reference numerals denote the same components, and the basic configuration and operation are the same. To do.

  First, the principle of improving execution performance by the interpreter of the present invention will be described below.

  As described above, an object of the present invention is to compensate for the weak points of each of the JIT interpreter and the HAE interpreter and increase the execution speed in consideration of the tendency of individual applications. By a macro analysis of the program structure, there is a series of processing procedures (that is, Hotspot) in which the application of the JIT method is determined to be effective in a certain processing procedure (hereinafter referred to as “method”) having a group. As for the method, it can be assumed that the execution speed can be improved as compared with the HAE method which can only improve the processing speed in units of instructions. Therefore, in the interpreter according to the present invention, the JIT method is applied to a series of processing procedures in the method. The highest priority is given to determining whether a hotspot is possible. Hereinafter, such a determination is referred to as “Hotspot determination”.

  In the interpreter of the present invention, for a method to be interpreted, a series of processing procedures excluded from the application of the JIT method or a series of processing procedures in the middle of determining whether the JIT method can be applied, All are treated as execution targets in the HAE method. However, there are many cases where IR instructions that can be executed by the HAE method are limited, and among the processing procedures that include many IR instructions that cannot be executed, there are some instructions that should be executed while interpreting with an interpreter. Hereinafter, such an exception is referred to as “HAE exclusion determination by IR factor”.

  Furthermore, in the interpreter of the present invention, at the time of execution, processing procedures in which garbage collection, exception processing, interrupt processing, etc. frequently occur are statistically collected, and those that should be executed while being interpreted by the interpreter are selected. Of course, since such garbage collection processing occurs irregularly, it is possible that there is no law to find the processing procedure to be selected, and interrupt processing occurs in a specific processing procedure. It may be a structure to do. However, in the interpreter of the present invention, it is assumed that the processing procedure to be selected is found by using a statistical method. Hereinafter, this is referred to as “HAE exclusion determination by an interrupt factor”.

  In addition to these determinations, one other parameter needs to be considered. That is how much power the user who executes the program requires. As described above, the HAE method generally requires less power consumption than the JIT method. This is an important judgment material (performance index) when the computer on which the interpreter is executed is a mobile device. How much importance is attached to this judgment material, it is necessary to leave the room for adjustment to the user, but the interpreter of the present invention allows the user to input the degree of power saving desired by the user as a parameter at the time of execution. The level value indicating the degree is used as a factor coefficient for shifting in a direction of relaxing or strictening the Hotspot judgment in the JIT method, for example, with a numerical value of 1 to 10. Hereinafter, the power saving intensity that becomes such a factor coefficient is referred to as a “power saving coefficient”.

  In the interpreter of the present invention, these “Hotspot judgment”, “HAE exclusion judgment by IR factor”, “HAE exclusion judgment by interrupt factor”, and the like, and further using the above “power saving coefficient” are executed. A feedback system for each processing procedure included in the target program is formed. As execution is repeated, the determination by such statistical methods increases in accuracy, and after a sufficient number of executions, it is possible to select the optimum execution form of the user program. Thus, this program is executed in a highly optimized state. Hereinafter, such a state is referred to as an “optimized execution state”. When the “optimized execution state” is reached, the processing system for performing the above-mentioned four determinations for the program and the processing system for feeding back the result are not necessary. Can be executed as a slim execution system with a small number of steps. Also, the execution speed of the program can be further increased by shifting to an execution system that eliminates unnecessary judgment processing.

  Hereinafter, a system for setting the interpreter of this embodiment on a circuit board for a microcomputer and confirming the operation will be described. FIG. 1 is a configuration diagram showing the overall configuration of a system for setting an interpreter according to an embodiment of the present invention on a circuit board for a microcomputer and confirming the operation.

  In the figure, an interpreter of this embodiment (hereinafter simply referred to as “interpreter”) is set on a circuit board for a microcomputer, and a system for confirming the operation is executed by the circuit board 100 on which the interpreter is set and the interpreter. HAE built-in microcomputer 110, a mouse 120 for determining a coordinate position to be input on a display 140, which will be described later, and a keyboard for inputting commands and data to an interpreter and an application program (hereinafter abbreviated as "application"). 130, a display 140 that displays information output by the interpreter or application, a host PC (personal computer) 150 that starts the interpreter and application, and a file system that stores the interpreter and application. Includes a file system built compact flash (memory) 160 that includes a memory (RAM) 170 serving as an interpreter and an application execution region. On the circuit board 100 on which the microcomputer 110 incorporating the HAE is mounted, there is mounted at least a compact flash 160 with a built-in file system, its diskette (not shown), and a memory 170 composed of RAM. There must be.

  Hereinafter, the function of the system for setting the interpreter of this embodiment to the circuit board for microcomputers and confirming the operation will be described. The platform on which the interpreter and application (such as IR) operate is the circuit board 100 on which the microcomputer 110 incorporating the above-described HAE is mounted, and this board is a device that is incorporated into various devices (such as mobile devices).

  The interpreter and the application executed by the interpreter are stored in a file system (hereinafter simply abbreviated as “file system”) stored in the compact flash 160 with a built-in file system, and are started and executed by the host PC 150. Be started. Of course, after being incorporated into the above-described device, the device is started from this device instead of the host PC 150. More specifically, the execution is started by a hardware switch provided in the device or by power-on. The interpreter that has started execution is first loaded from the file system into the memory 170, and after preparing itself for execution, the execution target application is also loaded from the file system 160 into the memory 170.

  Some applications have an input / output function, which is executed by receiving text input with the keyboard 130 and coordinate information with the mouse 120, for example. When there is a graphic output, it is drawn on the display 140 from the circuit board 100 via a graphic system such as a VGA (Video Graphic Array). All of these operations are performed by the application using functions on the interpreter side.

  FIG. 2 shows an example of an area map of a memory system including software necessary for the operation of the interpreter. However, the HAE 210 is not a software but a hardware mechanism (the HAE built-in microcomputer 110 incorporates the above-described HAE 210, and as soon as there is a HAE start instruction, an array of IR instructions is sent to the HAE 210 so that it can be executed at high speed. Waiting).

  The software stored in the memory system shown in FIG. 2 includes the real-time OS 220 that handles the operating system, the file system, and the graphics in the order that the software interface is closer to the hardware side, in order from the lower level (closest to the hardware side). And a network middleware group 221, an interpreter 230 including a JIT 231, an interpreter library 240 composed of an array of IRs, and an application program 250 written in IR. A program that cannot be executed by an interpreter for performance reasons must be realized in a compiler language as in the prior art, and is arranged as a normal native program 260 in parallel with the interpreter and the program. The real-time OS 220 starts up with the start of the system. This OS also includes a middleware group 221. The middleware group 221 can be used simultaneously with the OS.

  Next, an interpreter 230 incorporating a JIT 231 with an interpreter library 240 is loaded onto the memory 170, and the application 250 is loaded as described above. Only in this state, execution can be started from the entrance of the application 250. Since the interpreter language is usually described in a series of processing procedures (hereinafter referred to as “methods”), the interpreter starts executing in units of methods.

  In the following description, an execution method is selected. Basically, one of the following three execution methods is selected for each method (execution system selection means). The three execution methods are an execution method using a normal interpreter that does not have an optimization means (sequential execution means) and an execution method using a JIT interpreter that has optimization means (optimization means + sequentially). Execution means) and HAE (hardware mechanism).

  FIG. 3 is a flowchart showing a flow of processing for selecting an execution method of the interpreter according to the present embodiment. Here, a conceptual flowchart showing the selection method of the three methods is shown. In the selection method of the three methods shown in FIG. 3, based on the execution by the HAE, the HAE application exclusion is gradually determined while collecting information at the time of execution. That is, the execution method proceeds from an undifferentiated state to an optimized state along the structure of the application.

First, it is determined whether or not there is a Hotspot that is effective when a JIT interpreter is applied (step S300).
As described above, since this Hotspot is a program part in which processing is unevenly distributed when the entire program is looked at, if a certain method is called intensively, this method is a promising candidate for Hotspot. Alternatively, if the number of calls is not so high, but the number of internal loops is large, or the nesting is deep and the execution time is long, such a part is also a strong candidate for Hotspot. In any case, if it is predicted that if a certain method becomes faster, the whole program will be faster, this method can be regarded as a Hotspot, and the execution by the JIT interpreter is This is most effective when a method having such a Hotspot is applied.

  In order to determine whether or not a certain program part is such a Hotspot, data as a result of the structural analysis related to the above-described program part is required. Therefore, in the Hotspot determination step (step S300) Must refer to the analysis / feedback data 320 at the time of execution, which is a collection of data obtained by performing such a structural analysis at the time of the past execution. Another factor data to be referred to is a power saving coefficient 310. In general, when a JIT interpreter is executed, the number of execution instructions increases, so that execution by HAE does not consume power. The degree of power saving desired by the user is specified from the outside (for example, the keyboard 130 (FIG. 1)) as a parameter. In the Hotspot determination step (step S300), this parameter is referred to convert to a corresponding coefficient, and the determination as to whether or not it is the aforementioned Hotspot is adjusted using this coefficient. More specifically, when the user strongly desires power saving (that is, when the coefficient is large), the judgment that it is a Hotspot is made strict, and only the selected method is subject to JIT execution (step S340). And so on. The above judgments are collected in step S330.

Next, HAE is excluded due to an interrupt factor (step S350).
A method in which an interrupt factor frequently occurs must be returned to software execution while it is being executed in HAE, and switching is expensive. If it is predicted that external interrupts will frequently occur during execution of a certain method, it is more likely that this method can maintain high-speed performance if it is excluded from the target to be executed by HAE. Such a determination is also made with reference to the analysis / feedback data 320 at the time of execution. The method determined to be excluded is sent to the interpreter execution (step S400). In addition, in the HAE exclusion determination based on the IR factor (step S380), it is verified whether the method should be excluded from the execution target by the HAE based on the IR factor. In step S390, the verification result is determined, and If the IR instruction that can be executed by the HAE is included only in a predetermined ratio or less, the method is sent to the interpreter execution (step S400). The above-mentioned ratio including IR instructions that can be executed by the HAE can be obtained by referring to the static analysis data 370 of the IR, which is a summary of the results of static analysis of the IR. There is no need. The method remaining in this way is executed by HAE (step S410).

  FIG. 4 is a flowchart showing an optimum execution form by the interpreter according to this embodiment. In the figure, the execution program of a certain application is completely optimized by the static analysis as described above and the feedback of the analysis result at the time of execution. Shows the execution flow by the interpreter when is determined for each method.

  First, immediately after the process for starting the execution of the method (step S500), the table 520 showing the execution method is referred to, and the execution method of the method that is written in this table and is about to start executing is acquired ( Step S510). Next, in step S530, as an execution method related to the method, if “J” is written in the table, for example, in step S540, the method is executed by the JIT method (including a repetitive part). The process ends in S570. However, when executing in the JIT system at this timing, since the execution by the JIT system has already been repeated many times, the object program that has been interpreted and output by the JIT system interpreter is already in the memory. Just run the program.

  If, for example, “H” is written in the table as an execution method related to the method in step S530, the method is executed by HAE (including a repetitive part) in step S550, and the process is performed in step S570. Exit. However, there is a case where it is necessary to switch to execution on the software side during execution by this HAE. In such a case, the process proceeds to step S560 where execution is performed by a normal interpreter (including repetitive parts) in the middle. Of course there is a case to do. When the execution on the software side is completed after this transition, the control flow can be returned to the hardware side again, and the execution (including the repetitive part) can be performed by the HAE.

  In step S530, if “I” is written in the table as an execution method related to the method, execution is performed by a normal interpreter (including repetitive parts) in step S560, and the process ends in step S570. To do. However, in this case as well, the object program that has been interpreted and output by the interpreter can be used after the execution by the normal interpreter has already been repeated. Reaching step S570 described above is not limited to the case where the method is completed, but also includes the case where another method is called on the way. When this program is called and executed, the return method in step S570 depends on the processing system.

  FIG. 5 is an explanatory view showing a list as an example of information fed back for dynamic optimization processing by the interpreter according to the embodiment of the present invention. In the figure, information fed back in units of methods is shown. The information items displayed in a list in the figure are “method name” 600, “execution method” 610, “execution count” 620, “Hotspot” 630, “HAE non-executable IR count” 640, “total IR count” 650, “GC (garbage collection) occurrence count” 660, “interrupt occurrence count” 670, and “exception occurrence count” 680. A method of using the information shown in FIG. 5 will be described later (see FIG. 7).

  FIG. 6 is an explanatory diagram showing, as an example, a power saving coefficient that is one parameter for optimization processing by the interpreter according to the embodiment of the present invention. In the figure, a “power saving coefficient execution parameter” 700 for selecting whether or not to execute power saving is shown, and a “power saving coefficient” 710 for describing the power saving coefficient indicated by numerical values 1 to 10 is shown. . A method of using the information shown in FIG. 6 will be described later (see FIG. 7).

  FIG. 7 is a flowchart showing a processing procedure when the interpreter according to the embodiment of the present invention executes the dynamic optimization process and selects the optimal execution form. The processing flow shown in the figure shows an execution form before the interpreter according to the embodiment of the present invention reaches the optimized interpreter shown in FIG. 4, and here, dynamic optimization processing is performed. In addition, it is assumed that the execution form is executed while selecting the optimum execution form. In the figure, a table 820 shows a specific example as an example of the feedback information shown in FIG.

  First, immediately after the process for starting the execution of the method (step S800), an entry corresponding to the method in the table 820 is searched, and if there is no entry, an entry corresponding to the method is newly established (step S810). Next, since the method is called, the number of executions of this method is counted up (ie, +1), and the corresponding field (“execution number”) in the table 820 is updated (step S830). Next, with reference to the “execution method” field of the table 820, whether or not the execution method is already determined is verified. If the execution method is already determined, step S890 is executed. If the method has not been determined yet, the process proceeds to step S880 (step S840).

  In step 890, the execution method of the method is determined by referring to the three execution methods described in the “Execution method” field of the table 820, as in the case of the optimal execution mode shown in the flowchart of FIG. Depending on the execution method, one of step S900, step S910, and step S920 is executed, and then the process proceeds to step S930.

  In step S850, since the execution method has not yet been determined, the “total number of IRs” and the “number of non-HAE executable IRs” in the table 820 are referred to. If these settings have not yet been made, New setting is made, and the process proceeds to step S860.

  In step S860, it is determined whether or not the percentage of the number of IRs that cannot perform HAR exceeds a predetermined threshold (application exclusion determination unit). If the number of IRs exceeds the threshold, In step S870, after confirming that “the execution method is an interpreter” (that is, “I” is described in the “execution method” field of the table 820), the process proceeds to step S890 (first execution means). If the IR number does not exceed the threshold value, the process proceeds to step S880.

  In step 880, after confirming that “the execution method is HAE” (that is, “H” is described in the “execution method” field of the table 820), the process proceeds to step S890 (second execution means).

  The flow after step S890 is almost the same as in the optimal execution mode shown in the flowchart of FIG. 4, except for the processing step (step S930) for determining Hotspot and the HAE execution by exception. An adjustment process (step S940) is added.

  In step S930, the process flow up to immediately before is verified to determine whether or not a Hotspot exists (optimization determination means). Since the specific processing content for determining the presence of this Hotspot is a well-known technique, it is not described here, but the definition of Hotspot is as described above. If the above-mentioned Hotspot can be found for this method, at this step, “ON” is entered in the “Hotspot” field of the table 820 and the execution method is determined to be a JIT interpreter (ie, In this case, “J” is overwritten in the “execution method” field of the table 820 (third execution means).

  Next, in step S940, it is determined whether or not an interrupt caused by the above-described interrupt factor (GC (garbage collection), interrupt, exception processing) has occurred frequently in this method. It can be judged only when the number of times reaches a sufficient number. The number of occurrences of interrupts due to such interrupt factors is counted at the time of execution up to the present, and is recorded as a numerical value in the corresponding field of the table 820 for each interrupt factor. If the sufficient number of executions is indicated and the number of occurrences of the interrupt due to the interrupt factor exceeds a predetermined threshold, it is determined that execution by the interpreter is advantageous, and the “execution method” field of the table 820 "I" is overwritten to indicate that the execution method is an interpreter (fourth execution means). As shown in FIG. 7, an interrupt routine (960, 970, 980) is separately provided for counting the number of interrupt factor occurrences. The number of times of the corresponding field is counted by looking at what method is being executed. Is up.

  In addition, after executing for a sufficiently long time in the state of execution in the execution mode shown in FIG. 7, the execution method of the table 820 reaches a state where it can be regarded as statistically determined. At this time, the processes of steps S830 to S880 and the processes of steps S930 and S940 and steps S960 to S980 can be removed.

  Although this algorithm depends on the processing system, the execution method of the table 820 is statistically determined when the execution method field on the table 820 of all methods is not rewritten after being executed for a sufficiently long time. Since it can be determined that it has been confirmed, it is easy to add such processing.

  The interpreter of the present invention has been described above with specific embodiments, but the present invention is not limited to these. A person skilled in the art can make various changes and improvements to the configurations and functions of the invention according to the above-described embodiments or other embodiments without departing from the gist of the present invention.

  The present invention is a system for immediately executing a computer program written in a programming language between a computer and a user (programmer), without partial translation, and partially interpreting the program immediately. It can be applied to an interpreter having optimization means.

It is a block diagram which shows the whole structure of the system which sets the interpreter which concerns on embodiment of this invention to the circuit board for microcomputers, and confirms operation | movement. An example of an area map of a memory system including software necessary for the operation of the interpreter is shown. It is a flowchart which shows the flow of the process which selects the execution system of the interpreter which concerns on this Embodiment. It is a flowchart which shows the optimal execution form by the interpreter which concerns on this Embodiment. It is explanatory drawing which shows the list as an example of the information which the interpreter which concerns on embodiment of this invention feeds back for dynamic optimization processing. It is explanatory drawing which shows the power saving coefficient made into 1 parameter for the optimization process by the interpreter which concerns on embodiment of this invention as an example. It is a flowchart which shows the process sequence in the case of performing, while selecting the optimal execution form while the interpreter which concerns on embodiment of this invention performs a dynamic optimization process.

Explanation of symbols

110 HAE built-in microcomputer 120 mouse 130 keyboard 140 display 150 host PC (personal computer)
160 Compact Flash with built-in file system (memory)
170 Memory (RAM)
220 Real-time OS (Operating System)
221 Middleware 230 Interpreter (JIT method)
240 Library for interpreter 250 Application program (using intermediate language IR)
600 “Method Name” (Hereinafter, brackets indicate item names)
610 “Execution Method”
620 “Number of executions”
630 “Hotspot” (optimization target)
640 “Number of non-HAE executable IRs”
650 “Total number of IRs”
660 “Number of GC (garbage collection) occurrences”
670 “Number of interrupt occurrences”
680 “Number of exception occurrences”
700 “Power-saving coefficient runtime parameters”
710 “Power Saving Factor”

Claims (7)

Method of operating a computer comprising a hardware mechanism capable of directly executing a part of instructions of a program written in a high-level language or an intermediate language, and a storage device storing a table describing the execution method for each program part Because
An application exclusion determination step for determining whether or not the program portion includes more than a predetermined number of instructions that cannot be executed by the hardware mechanism;
An optimization determination step for determining whether a JIT method can be applied to the program portion;
An execution method determination step comprising:
In the application exclusion determination step, when it is determined that the program part includes more than a predetermined number of instructions that cannot be executed by the hardware mechanism, the program part is sequentially executed while converting the machine language. Set the execution method of the program part of the table to sequentially execute the program part,
In the application exclusion determination step, when it is determined that the program part does not include more than a predetermined number of instructions that cannot be executed by the hardware mechanism, a part of the program part that can be executed by the hardware mechanism Is executed by the hardware mechanism and sets the execution method of the program part of the table so that the part that cannot be executed by the hardware mechanism is sequentially executed while converting the sequence of machine language. A portion that can be executed by the hardware mechanism is executed by the hardware mechanism, and a portion that cannot be executed by the hardware mechanism is sequentially executed while converting a sequence of machine language;
If it is determined in the optimization determination step that the JIT method can be applied to the program part, the program part is optimized in terms of execution speed regardless of the determination result in the application exclusion determination step. An execution method determination step, wherein an execution method of the program part of the table is set so as to be converted into an optimized object program and executed, and the program part is converted into the optimized object program and executed. ,
An execution method determination step for determining whether or not the execution method of each program part in the table is unchanged as a result of repeatedly executing the program part by the execution method determination step;
Have
In the execution method determination step, when it is determined that the execution method of each program portion in the table has not changed, the program is executed according to the execution method stored in the table. . In the execution method determination step, it is determined whether or not the program part causes execution of an instruction that cannot be executed by hardware more than the predetermined number of times when the execution number of the program part reaches a predetermined number of times. And further comprising an exemption verification step to verify,
In the execution method determination step,
In the application exclusion verification step, when it is determined that the program part generates more than a predetermined number of instructions that cannot be executed by hardware, the program part is converted into a sequence of machine languages. The execution method of the program portion of the table is set so as to be executed sequentially, and the program portion is executed sequentially while converting the sequence of machine language at the time of execution exceeding the predetermined number of times. 2. The operation method according to 1.   In the optimization determination step, it is determined that the JIT method can be applied to a program part in which the number of internal loops is a predetermined number or more, or a program part including a nested structure exceeding a predetermined depth. The operation method according to claim 1, wherein the operation method is characterized.   The optimization determination step refers to a value representing a degree of power saving preset by a user, and performs the determination using the predetermined number or the predetermined depth according to the value. 3. The operation method according to 3. The program for making the said computer perform the operation | movement method of any one of Claim 1 to 4. The computer stored in said storage device according to claim 5, wherein the program. 5. The computer that operates according to the operation method according to claim 1.

Images