JP2013242604A - Execution module optimization device, execution module optimization method and program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To solve the problem of conventional compiler devices, in which optimization of an execution module cannot be performed.SOLUTION: Optimization of an execution module can be performed by an execution module optimization device including: a disassembling unit for disassembling the execution module and obtaining an assembly program; an optimization part detection unit for detecting one or more optimization parts from the assembly program, using optimization part detection information; an altering unit for altering the optimization part using alteration information corresponding to the optimization part detection information used when each optimization part is detected, and obtaining an optimized assembly program; and an execution unit 108 for executing the optimized assembly program.

Description

  The present invention relates to an execution module optimizing device that optimizes an execution module.

  Conventionally, in a compiler apparatus that compiles source code and creates an execution module, a compiler apparatus that optimizes the source code and creates an execution module has been developed. (For example, refer nonpatent literature 1).

“Intel C ++ Compiler 11.1 Professional Edition Details”, [online], Intel Corporation, [searched August 4, 2010], Internet <URL: http://download.intel.com/software/doc/ compilers / reseller_productpage_cpp_compiler_windows.pdf>

  However, the conventional compiler apparatus has a problem that a compiled execution module cannot be optimized.

  The execution module optimizing device according to the first aspect of the present invention includes an execution module storage that can store one or more execution modules, and an optimization that is information for detecting an optimization location that is a location where processing can be optimized. An optimization information storage unit that can store one or more pieces of optimization information having optimization location detection information and change information that is information for changing the optimization location detected using the optimization location detection information; An accepting unit that accepts an instruction having an execution module identifier that identifies an execution module, and an execution module that is identified by the execution module identifier included in the instruction is read from the execution module storage unit, disassembled, and an assembly program corresponding to the execution module From the assembly program using the disassembly unit that obtains the information and one or more optimization location detection information in the optimization information storage unit Optimized change information corresponding to the optimized location detection information used when detecting each optimized location for the optimized location detection unit that detects the above optimized locations and one or more optimized locations A change unit that acquires from the optimization information storage unit, changes one or more optimization locations using the change information, and acquires an optimized assembly program; and an execution unit that executes the optimized assembly program This is an execution module optimization device.

  With such a configuration, the execution module can be optimized.

  In addition, the execution module optimization device according to the second aspect of the invention has the optimization information including the optimization location detection information, the parameter information indicating the parameters to be collected, and the change information. Then, the optimization location detection unit detects the optimization location candidate detection means for detecting one or more optimization location candidates that are loops from the assembly program using the optimization location detection information, and the optimization location candidate detection means detects Information that matches a predetermined pattern is acquired from the one or more candidate optimization locations, and it is determined whether or not the information matches a predetermined condition. Static analysis means for performing static analysis to determine a location as an optimization location candidate, and an assembly for collecting parameters indicated by the parameter information for each of one or more optimization location candidates determined by the static analysis means A program instruction sequence that inserts a program instruction sequence in front of an optimization location candidate and generates a profile program; a dynamic analysis unit that executes the profile program and performs dynamic analysis to acquire one or more parameters; Judgment means for judging whether or not the above parameters meet a predetermined condition, and if one or more parameters meet a predetermined condition, optimize one or more optimization location candidates An execution module optimizing device comprising a determining means for determining a location.

  With such a configuration, the loop portion of the execution module can be optimized.

  The execution module optimizing apparatus according to the third aspect of the present invention is different from the second aspect in that the profile program generating means copies one or more optimization location candidates and copies the one or more optimization locations. An assembly program instruction sequence for collecting parameters indicated by the parameter information immediately before the candidate and jumping immediately before the inserted assembly program instruction sequence immediately before one or more original optimization location candidates Is an execution module optimizing device for inserting

  With such a configuration, the loop portion of the execution module can be optimized.

  The execution module optimizing device according to the fourth aspect of the present invention further includes a profiler storage unit that can store a profiler that executes an assembly program and obtains the load of each instruction, as compared with the second or third aspect of the invention. Then, the optimization location candidate detection means detects one or more loops from the assembly program using the optimization location detection information, executes a profiler for the assembly program, acquires the load of each instruction, An execution module optimizing device that detects a loop having an instruction exceeding a threshold as one or more optimization location candidates.

  With such a configuration, it is possible to optimize the high load portion of the execution module.

  The execution module optimizing device according to the fifth aspect of the present invention is the second to fourth aspects of the invention, wherein the optimization location candidate detection means uses the optimization location detection information to determine one or more from the assembly program. A control flow graph having two or more nodes having each instruction constituting the assembly program as a node is created from the assembly program. From the control flow graph, an arbitrary one node is a starting point. The execution module optimizing device detects one or more optimization location candidates that are dominating loops in which the node immediately before returning to the starting node and the starting node are in a dominating relationship.

  With such a configuration, it is possible to optimize the dominant loop of the execution module.

  In the execution module optimizing device according to the sixth aspect of the present invention, the parameter information is information indicating whether the number of loops and the pointer overlap or not. The analysis unit acquires a loop invariant value and a guidance variable for accessing the array for each of one or more optimization site candidates detected by the optimization site candidate detection unit, and acquires the obtained loop invariant value and the guidance. The variable is used to determine whether or not a memory store for the sequence is performed in the optimization location candidate, and if so, a static analysis is performed to determine the location as the optimization location candidate, and determination means Determines whether the number of loops acquired by the dynamic analysis means is greater than or equal to a threshold and whether the pointers overlap or not is information indicating that the pointers do not overlap, The determining unit determines the optimization location candidate that the determining unit determines that the number of loops is equal to or greater than a threshold value and the information indicating whether or not the pointer overlaps is information indicating that the pointer does not overlap. The optimization part is determined as the optimization part, and the changing unit performs the multi-core parallelization to change the assembly part instruction sequence that can be executed in parallel in a plurality of processors to the optimization part determined by the judging means as the optimization part. Device.

  With such a configuration, execution modules can be optimized by multi-core parallelization.

  In the execution module optimizing device according to the seventh aspect of the invention, the parameter information is information indicating whether the number of loops and the pointer overlap or not. The analysis unit acquires a variable name for each of the one or more optimization location candidates detected by the optimization location candidate detection unit, and uses the acquired variable name to determine the dependency relationship between the variables in the optimization location candidate. If it does not exist, perform static analysis to determine the location as an optimized location candidate, and the determination means has a loop count acquired by the dynamic analysis means that is greater than or equal to a threshold value and a pointer The information indicating whether or not the pointers overlap is information indicating that the pointers do not overlap, and the determining means determines that the determining means is that the number of loops is equal to or greater than a threshold and the pointers overlap. La The optimization part candidate determined that the information indicating whether or not the information is information indicating that the pointer does not overlap is determined as the optimization part, and the changing unit determines the optimization part as the optimization part. This is an execution module optimizing device that performs SIMD conversion to an assembly program instruction that can simultaneously process a plurality of data with a single instruction with respect to an optimization location.

  With such a configuration, the execution module can be optimized by SIMD.

  The execution module optimizing device according to the eighth aspect of the present invention is the first aspect of the invention, wherein the optimization location detection information is a function name for performing 1 processing, and the change information is for performing 1 processing. The name of the function that performs processing at a higher speed than the function indicated by the optimization location detection information, and the changing unit converts one or more optimization locations into the optimization location detection information used when detecting each optimization location. It is an execution module optimizing device that obtains an optimized assembly program by changing to corresponding change information.

  With such a configuration, the execution module can be optimized by API / Library replacement.

  The execution module optimization device according to the ninth aspect of the present invention includes an execution module storage unit that can store one or more execution modules, and information for detecting an optimization location that is a location where processing can be optimized. An optimization information storage unit capable of storing one or more pieces of optimization information having certain optimization location detection information and change information that is information for changing the optimization location detected using the optimization location detection information A reception unit that receives an instruction having an execution module identifier that identifies an execution module, and an execution module identified by the execution module identifier included in the instruction is read from the execution module storage unit, disassembled, and corresponds to the execution module An assembly program is obtained by using the disassembly unit for obtaining an assembly program and one or more pieces of optimization location detection information in the optimization information storage unit. From the optimization location detection unit that detects one or more optimization locations and change information corresponding to the optimization location detection information used for detecting each optimization location for each of the one or more optimization locations A change unit that obtains an optimized assembly program by changing one or more optimization locations using the change information, a storage unit that stores the optimized assembly program, Is an execution module optimizing device.

  With such a configuration, the execution module can be optimized.

  The execution module optimizing device according to the tenth aspect of the present invention further includes a compiling unit that compiles an optimized assembly program and obtains an optimized execution module, as compared with the ninth aspect of the invention. The unit is an execution module optimizing device that also stores optimized execution modules.

  With such a configuration, the execution module can be optimized.

  According to the execution module optimizing apparatus and the like according to the present invention, the execution module can be optimized.

Block diagram of execution module optimizing apparatus 1 according to Embodiment 1 The figure which shows the example of the same control flow graph The figure which shows the example of the information which shows the same control flow graph The figure which shows the example of the information which shows the same control flow graph The figure which shows the example of the information which shows the node which has the same dominance relation The figure which shows the example of the information which shows the node which has the same dominance relation A flowchart showing the overall operation of the execution module optimization apparatus 1 A flowchart for explaining the detection processing of the optimization location The flowchart explaining the detection process of the location which matches the optimization location information Flowchart explaining the detection process of the same optimization location Flowchart for explaining optimization location candidate detection processing using the profiler The flowchart explaining the detection process of the optimization location candidate using the control flow graph Flowchart explaining creation process of the control flow graph Flowchart explaining change processing of the same optimization location The figure which shows the example of the optimization information Diagram showing an example of the assembly program Diagram showing an example of the profile program Diagram showing an example of the multi-core parallelization The figure which shows the example of the same SIMD conversion Diagram showing an example of the replacement Diagram showing an example of the optimized assembly code Block diagram of execution module optimizing device 2 in Embodiment 2 Overview of the computer system Block diagram of the computer system

  Embodiments of an execution module optimizing device according to the present invention will be described below with reference to the drawings. In addition, since the component which attached | subjected the same code | symbol in embodiment performs the same operation | movement, description may be abbreviate | omitted again.

(Embodiment 1)
In the present embodiment, an execution module optimization apparatus 1 that optimizes an execution module will be described.

  FIG. 1 is a block diagram of an execution module optimizing apparatus 1 according to this embodiment.

  The execution module optimization apparatus 1 includes an execution module storage unit 101, an optimization information storage unit 102, a profiler storage unit 103, a reception unit 104, a disassembly unit 105, an optimization location detection unit 106, a change unit 107, and an execution unit 108. Prepare.

  The optimization location detection unit 106 includes optimization location candidate detection means 1061, static analysis means 1062, profile program generation means 1063, dynamic analysis means 1064, determination means 1065, and determination means 1066.

  The execution module storage unit 101 can store one or more execution modules.

  Here, the execution module is a program obtained by compiling a so-called source code, and is called a so-called execution file or execution program. In other words, the execution module is a program that operates on an arbitrary OS (Operation System). The execution module may be a file or a file group indicating the program. The execution module does not include a program written in a so-called script language that operates by an interpreter.

  A program is a concept including a meaning associated with the word “program” in the software field. That is, the program may be a software program or may be written in a programming language.

  The execution module storage unit 101 is preferably a non-volatile recording medium, but can also be realized by a volatile recording medium. The process of storing the execution module in the execution module storage unit 101 is not limited. For example, the execution module may be stored in the execution module storage unit 101 via a recording medium. Further, the execution module transmitted via a communication line or the like may be stored in the execution module storage unit 101. Furthermore, the execution module input via the input device may be stored in the execution module storage unit 101.

  The optimization information storage unit 102 can store optimization information.

  Here, the optimization information is information having at least one piece of optimization location detection information and change information. For example, the optimization location detection information and the change information may or may not correspond one-to-one. The optimization location detection information and the change information may be in any form as long as one of the information is associated with the other information.

  The optimization information may include parameter information. In this case, the parameter information may or may not correspond to the optimization location detection information or the change information on a one-to-one basis.

  The optimized location detection information is information for detecting an optimized location that is a location where processing can be optimized. The optimization location detection information may be, for example, a program code, a program code pattern, information indicating an optimization location detection method, a program, or the like. The program code is, for example, a function name for performing an arbitrary process or an instruction for performing an arbitrary process. The program code pattern is, for example, a pattern indicating a combination of a control syntax and a variable, a so-called regular expression, or the like. Further, the information indicating the method for detecting the optimization location includes, for example, information indicating that a loop is detected, information indicating that a function is detected, and the like. The program is, for example, a program that detects a loop or a program that detects a function.

  The change information is information for changing the optimized location detected using the optimized location detection information. The change information may be, for example, a program code, a program code pattern, information indicating a method for changing an optimization location, a program, or the like. The program code is, for example, a function name for performing an arbitrary process or an instruction for performing an arbitrary process. The program code pattern is, for example, a pattern indicating a combination of a control syntax and a variable, a so-called regular expression, or the like. The information indicating the method of changing the optimization location includes, for example, information indicating that multi-core parallelization is performed on the specified optimization location, information indicating that SIMD processing is performed, and arbitrary information. For example, information indicating replacement. The program is, for example, a program that performs multi-core parallelization, a program that performs SIMD, or a program that replaces arbitrary information with respect to a specified optimization location.

  The parameter information is information indicating parameters to be collected. That is, the parameter information is information indicating that an arbitrary parameter is collected. The parameter information is information used by a profile program generation unit 1063 described later. The parameter information includes, for example, information indicating that the number of loops is collected, information indicating that the number of variables is collected, information indicating whether or not there is a branch, and pointer is over. It may be information indicating that information indicating whether or not to wrap is collected. The parameter information is not limited as long as it is information indicating an arbitrary event or state.

  Optimization is a concept that includes speeding up, memory usage reduction, low power consumption, deletion of redundant program code, and the like, and includes a degree of improvement and is widely understood.

  Further, the optimization information may include detection target information indicating what is to be detected by the optimization location detection information, for example. The detection target information may be, for example, information indicating that the target detected by the optimized location detection information is a loop, information indicating that the target detected by the optimized location detection information is a function, or the like. Further, the detection target information may be included in the optimization information, or when the optimization location detection unit 106 and the optimization location candidate detection unit 1061 described later detect the optimization location and the optimization location candidate. Further, it may be determined and acquired from the optimization location detection information. Further, the detection target information may be acquired by the changing unit 107 described later, judging from the optimization location detection information.

  The optimization information storage unit 102 is preferably a non-volatile recording medium, but can also be realized by a volatile recording medium. Moreover, the process in which optimization information is memorize | stored in the optimization information storage part 102 is not ask | required. For example, the optimization information may be stored in the optimization information storage unit 102 via a recording medium. Further, optimization information transmitted via a communication line or the like may be stored in the optimization information storage unit 102. Further, the optimization information input via the input device may be stored in the optimization information storage unit 102.

  The profiler storage unit 103 can store a profiler.

  Here, the profiler is a program for monitoring processing executed by an operating program and the order in which the processing is executed, and is a program for collecting profile information. The profiler may be a program that executes an assembly program and acquires the load of each instruction, for example. If the OS is Windows (registered trademark), an “event trace provider” (http://www.microsoft.com/whdc/devtools/tools/eventtrace_sample.mspx) is known as the profiler. As the profiler, if the OS is Linux, “OProfile” (http://oprofile.sourceforge.net/) is known. Note that the details of the profiler are well-known techniques, and the description thereof is omitted.

  The load may be, for example, an instruction execution time, a memory usage amount, or any content.

  The profiler storage unit 103 is preferably a nonvolatile recording medium, but can also be realized by a volatile recording medium. The process of storing the profiler in the profiler storage unit 103 is not limited. For example, the profiler may be stored in the profiler storage unit 103 via a recording medium. Further, a profiler transmitted via a communication line or the like may be stored in the profiler storage unit 103. Furthermore, the profiler input via the input device may be stored in the profiler storage unit 103.

  The accepting unit 104 accepts an instruction.

  Here, the instruction is, for example, an instruction to execute an execution module, an instruction to optimize an execution module, or the like. The instruction is an instruction having an execution module identifier. An execution module identifier is information for identifying an execution module.

  For example, if the execution module optimization apparatus 1 is middleware that is always executed when the execution module is executed, the instruction having the execution module identifier is, for example, an icon indicating the execution module. For example, double click or single click. For example, if the execution module optimizing apparatus 1 is an application that runs on the OS, an instruction having an execution module identifier is given by, for example, dragging and dropping an icon indicating the execution module onto the apparatus. is there. That is, the instruction having the execution module identifier may be in any format as long as the instruction has an execution module identifier for identifying the execution module.

  Acceptance refers to accepting information input from an input device such as a keyboard, mouse, or touch panel, receiving information transmitted via a wired or wireless communication line, and reading from a recording medium such as an optical disk, magnetic disk, or semiconductor memory. It is a concept that includes accepting information that has been issued.

  The instruction input means having the execution module identifier in the receiving unit 104 may be anything such as a keyboard, a mouse, or a menu screen. The accepting unit 104 can be realized by a device driver for input means such as a keyboard, control software for a menu screen, or the like.

  The disassembly unit 105 reads out an execution module stored in the execution module storage unit 101, disassembles it, and obtains an assembly program corresponding to the execution module.

  Here, the execution module to be read is usually an execution module identified by the execution module identifier included in the instruction received by the receiving unit 104.

  Also, disassembling is converting an execution module into an assembly program. Since disassembly is a known technique, a detailed description thereof is omitted.

  Further, the disassembly unit 105 may obtain an assembly program corresponding to the execution module and the disassembly of the execution module by using, for example, a disassembler that is a program for performing the disassembly. In this case, the disassembler may or may not be held by the disassembler 105, for example. Otherwise, the disassembler may be stored in any storage area.

  An assembly program is a program written in a so-called assembly language.

  The disassembly unit 105 can be usually realized by an MPU, a memory, or the like. The processing procedure of the disassembly unit 105 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The optimization location detection unit 106 detects one or more optimization locations from the assembly program acquired by the disassembly unit 105. The optimization location detection is performed using one or more optimization location detection information stored in the optimization information storage unit 102. Further, the optimization location is detected by the optimization location candidate detection unit 1061, the static analysis unit 1062, the profile program generation unit 1063, the dynamic analysis unit 1064, the determination unit 1065, and the determination, which will be described later. This may be done by means 1066.

  Here, the optimization location is a location that is changed by the change unit 107 described later using the change information, and is a location in the assembly program acquired by the disassembly unit 105. The information indicating the optimization location may be, for example, a line number indicating the location of the location in the assembly program, a line number indicating the range of the location, or one or more program codes constituting the assembly program. Good. The information indicating the optimization location is not limited as long as it is information that can identify an arbitrary location in the assembly program acquired by the disassembly unit 105.

  Moreover, the information which shows an optimization location may contain the change method information which shows the method for the below-mentioned change part 107 to change the said optimization location, for example. The change method information includes, for example, information indicating that replacement is performed with the change information corresponding to the optimization location detection information used when detecting the optimization location, information indicating that the change is performed using the change information, and the like. There may be. The change method information may be acquired from the change information, for example, or may be acquired from information indicating the detected optimization location.

  Moreover, the detection of an optimization location is acquiring the information which pinpoints the arbitrary locations in the assembly program which the disassembly part 105 acquired using the optimization location detection information.

  The optimization location detection unit 106 can be usually realized by an MPU, a memory, or the like. The processing procedure of the optimized location detection unit 106 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The optimization location candidate detection means 1061 detects one or more optimization location candidates from the assembly program acquired by the disassembly unit 105. The optimization location candidate is detected using one or more pieces of optimization location detection information stored in the optimization information storage unit 102. Further, the optimization location candidate may be detected using a profiler stored in the profiler storage unit 103 or based on a control flow graph created from the assembly program acquired by the disassembly unit 105. Good.

  Here, the optimization location candidate is a location that can be an optimization location, and is a location in the assembly program acquired by the disassembly unit 105. The information indicating the optimization location candidate includes, for example, information indicating a loop, a line number indicating the position of the location in the assembly program, a line number indicating the range of the location, and one or more programs constituting the assembly program It may be a code or the like. The information indicating the optimization location candidate is not limited as long as it is information that can identify an arbitrary location in the assembly program acquired by the disassembly unit 105.

  Moreover, the detection of an optimization location candidate is to acquire information for specifying an arbitrary location in the assembly program acquired by the disassembly unit 105 using the optimization location detection information.

  In addition, the optimization location candidate is detected by, for example, comparing the code of an arbitrary line in the assembly program with the optimization location detection information while shifting the comparison target line by line, and optimizing the matching location. It may be detected as a location.

  The optimization location candidate detection unit 1061 may detect an optimization location candidate by executing a profiler stored in the profiler storage unit 103 for the assembly program acquired by the disassembly unit 105, for example.

  The optimization location candidate detection means 1061 executes, for example, the profiler for the assembly program, acquires the load of each instruction, and uses the optimization location detection information indicating the loop pattern from the assembly program. One or more loops may be detected, and among the detected one or more loops, a loop having an instruction whose load exceeds a threshold may be detected as an optimization location candidate.

  The optimization location candidate detection unit 1061 detects, for example, one or more loops using the optimization location detection information indicating the loop pattern from the assembly program, executes the profiler for the assembly program, The load of each instruction in the detected one or more loops may be acquired, and among the detected one or more loops, a loop having an instruction whose load exceeds a threshold value may be detected as an optimization location candidate.

  In addition, said threshold value may be decided beforehand and may not be so. When it is determined in advance, the threshold value may be stored in an arbitrary storage area. If the threshold is not determined in advance, the threshold value may be, for example, the average of the loads of all the lines in the assembly program, half the maximum load among the loads of all the lines in the assembly program, It may be an average of all instruction loads, half the maximum load among all instruction loads in the assembly program, or the like.

  In addition, the optimization location candidate detection unit 1061 may create a control flow graph from the assembly program acquired by the disassembly unit 105 and detect the optimization location candidate based on the control flow graph.

  Further, the optimization location candidate detection unit 1061 detects, for example, one or more loops using the optimization location detection information indicating the loop pattern from the assembly program, creates a control flow graph from the assembly program, A loop that is a dominant loop is detected from the control flow graph, and a loop that is detected by using the optimized location detection information is compared with a loop that is detected by using the optimized location detection information and a dominant loop that is detected from the control flow graph. Among them, a loop that is a dominant loop may be detected as an optimization location candidate.

  Further, the optimization location candidate detection unit 1061 creates a control flow graph from the assembly program, for example, and detects a loop that is a dominant loop from the control flow graph using optimized location detection information indicating the pattern of the dominant loop. Then, a loop in the assembly program corresponding to the loop in the detected control flow graph may be detected as an optimization location candidate.

  Here, the control flow graph is a control flow graph (Control Flow Graph), which is a graph showing all paths of instructions that may be passed when a program is executed. The control flow graph may have two or more nodes having each instruction constituting the assembly program as a node. The control flow graph is, for example, as shown in FIG.

  The information indicating the control flow graph is, for example, information indicating the relationship between two nodes, that is, the node where the arrow appears in the control flow graph of FIG. Alternatively, the information shown in FIG. 4 may be used. The information indicating the control flow graph is not particularly limited as long as the graph as shown in FIG. 2 can be expressed.

  The dominant loop is a loop starting from any one node, and the node immediately before returning to the starting node and the starting node are in a controlling relationship.

  Further, the dominating relationship means that in any two nodes in the control flow graph, in order to pass one node, the other node must be passed. For example, in the control flow graph as shown in FIG. 2, in order to pass B, you must pass A. Therefore, “B is controlled by A” and “A and B are in a control relationship”. Say. In FIG. 2, in order to pass through E, you must pass through B. Therefore, “E is controlled by B” and “B and E are in a control relationship”. In FIG. 2, it is not always necessary to pass C in order to pass E, and it is not always necessary to pass D in order to pass E. "C and E are not in a dominating relationship", "E is not controlled by D", "D and E are not in a dominating relationship".

  Further, in the control flow graph as shown in FIG. 2, in order to pass through I, it must pass through F, so it is said that “I is controlled by F”, and “F and I are in a control relationship”. Say. In FIG. 2, “F → G → H → I → F” is a loop, and I which is a node immediately before returning to F which is a node which is a starting point of the loop is in a dominating relationship. Such a loop is called a dominant loop. In FIG. 2, since A and E are in a dominating relationship, the loop “A → B → C → E → A” is a dominating loop. Similarly, the loop “A → B → D → E → A” is also a dominant loop.

  Further, the optimization location candidate detection unit 1061 may create a control flow graph, analyze the dominating relationship between the nodes, and accumulate the analysis result in an arbitrary storage area. In this case, the information indicating the analysis result of the dominant relationship is, for example, information indicating the relationship between the ruled node and the ruled node in the control flow graph of FIG. It may be. The information indicating the analysis result of the dominating relationship may be any format as long as the dominating relationship between the nodes is known.

The determination as to whether or not the loop is a dominant loop may be made, for example, by the following procedure.
(1) A loop in the control flow graph is detected. For example, detection is performed in a format such as “A → B → C → A”.
(2) Analyze the dominating relationship between the last two nodes of the loop detected in (1), and obtain the analysis result. For example, the analysis result is acquired in a format such as “dominate: A | dominated: C”.
(3) It is determined whether or not the analysis result obtained in (2) has a dominating relationship.
(4) When it is determined that there is a dominating relationship in (3), it is determined that it is a dominating loop.

  Further, the optimization location candidate detection unit 1061 determines whether to detect the optimization location candidate using the profiler or to detect the optimization location candidate using the control flow graph, and uses the profiler to determine the optimization location candidate. If it is determined that the optimized location candidate is detected using the optimized location detection information and the profiler, and the optimized location candidate is detected using the control flow graph, the optimized location detection information and You may detect an optimization location candidate using a control flow graph. In this case, the determination includes, for example, information indicating whether the optimization location candidate is detected using the profiler or whether the optimization location candidate is detected using the control flow graph. It may be determined using the information. The determination may be, for example, determining whether or not the profiler is stored in the profiler storage unit 103. In addition, the determination is, for example, information for performing multi-core parallelism or information for performing SIMD, in which change information corresponding to each piece of optimization location detection information used when detecting a loop is information. If it is information for performing multi-core parallelization, it is determined that an optimization location candidate is detected using a profiler. If it is information for performing SIMD, optimization is performed using a control flow graph. It may be determined that a location candidate is detected.

  In addition, the detection of the optimization location candidate which the optimization location candidate detection means 1061 performs can be any method as long as the optimization location candidate can be detected.

  The optimization location candidate detection means 1061 can be usually realized by an MPU, a memory, or the like. The processing procedure of the optimized location candidate detection means 1061 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The static analysis unit 1062 performs a static analysis on the optimization location candidate detected by the optimization location candidate detection unit 1061.

Here, the static analysis is a process of determining a location indicated by the optimization location candidate detected by the optimization location candidate detection unit 1061 as an optimization location candidate. The static analysis is, for example, the following processing.
(1) The optimization location candidate detected by the optimization location candidate detection means 1061 is acquired.
(2) Information matching a predetermined pattern is acquired from each optimization location candidate acquired in (1).
(3) It is determined whether the information acquired in (2) matches a predetermined condition.
(4) If it is determined in (3) that they match, the location corresponding to the information acquired in (2) is determined as an optimized location candidate.

  Here, the predetermined pattern is a pattern for acquiring information from the optimization location candidates detected by the optimization location candidate detection means 1061. The predetermined pattern may be, for example, a loop invariant value, an induction variable for accessing an array, or a pattern for acquiring a variable name. That is, in this case, information that matches a predetermined pattern is a loop invariant value, an induction variable for accessing the array, or a variable name.

  The predetermined pattern may be, for example, “mov edx, dword ptr [ebp + # * eax− #]”, “ptr [ebp + ## * eax− #]”, “edp”, or the like. The predetermined pattern is not particularly limited as long as information can be acquired from the optimization location candidates detected by the optimization location candidate detection means 1061. The predetermined pattern may or may not be held by the static analysis unit 1062. Otherwise, the predetermined pattern may be stored in an arbitrary storage area.

  A loop invariant value is a value (variable) that does not change in a loop without depending on the repetition of processing in the loop. For example, in a loop indicated by “for (i = 0; i <10; i ++)”, when program codes “a = 10;” and “b = i * 2;” exist, “a” Since it does not change without depending on repetition of processing in the loop, it is a loop invariant value. Also, “b” is not a loop invariant value because it changes depending on the repetition of processing in the loop.

  The induction variable is a variable indicating repetition of processing in the loop in the loop. For example, in a loop indicated by “for (i = 0; i <10; i ++)”, “i” is a variable indicating the iteration of the loop, and thus is an induction variable.

  The predetermined condition is a condition for determining a location corresponding to information that matches a predetermined pattern as an optimization location. Further, the predetermined condition may be, for example, that information acquired by a predetermined pattern matches an arbitrary condition, and an event that can be determined using the information matches an arbitrary condition. It may be.

  In addition, the predetermined condition may be, for example, that a memory store is performed, or that there is no dependency between variables. For example, when the predetermined condition is that the memory store is not performed, the determination as to whether or not the condition is met is based on the loop invariant value and the induction variable that can be acquired by the predetermined pattern. It may be used to determine whether a memory store is being performed on the array. Also, for example, when the predetermined condition is that there is no dependency relationship between variables, the determination as to whether or not the condition is satisfied is performed using the variable name acquired by the predetermined pattern. It may be determined whether or not there is a dependency relationship between variables.

  That is, the static analysis unit 1062 makes a determination according to a predetermined condition, and determines whether or not the result of the determination matches a predetermined condition.

  The predetermined condition may be, for example, “memory is being stored”, “no dependency between variables exists”, “the pointers do not overlap”, or the like. The predetermined condition is not particularly limited as long as the location corresponding to the information matching the predetermined pattern can be determined as the optimization location. The predetermined condition may or may not be held by the static analysis means 1062. Otherwise, the predetermined condition may be stored in an arbitrary storage area.

  Further, the dependency relationship between variables means that, for example, processing such as assignment of a value to a certain variable depends on processing results of other variables. That is, for example, when there are two variables “A” and “B”, and codes “A = 1” and “B = A + 2” exist, the assignment of a value to “B” is “A”. Depends on the result of assigning the value to. In such a case, it is determined that there is a dependency between variables.

  The static analysis performed by the static analysis means 1062 is not particularly limited as long as the optimization location candidate can be determined.

  The determination of the optimization location candidate is to acquire information for specifying an arbitrary location in the assembly program acquired by the disassembly unit 105 using the optimization location detection information.

  The static analysis means 1062 can usually be realized by an MPU, a memory, or the like. The processing procedure of the static analysis means 1062 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The profile program generation unit 1063 generates a profile program from the assembly program in which the static analysis unit 1062 performs static analysis and the optimization location candidate is determined. Further, the profile program generation unit 1063 may generate a profile program from the assembly program acquired by the disassembly unit 105 based on the optimization location candidates determined by the static analysis unit 1062.

  Here, the profile program is a program for collecting parameters indicated by the parameter information stored in the optimization information storage unit 102.

The profile program is generated by the following process, for example.
(1) The optimization location candidate determined by the static analysis means 1062 is acquired. Here, it is assumed that m optimization location candidates have been acquired.
(2) Set 1 to the counter i.
(3) The i-th optimized location candidate is acquired from the optimized location candidates acquired in (1).
(4) An assembly program instruction sequence for collecting parameters indicated by the parameter information stored in the optimization information storage unit 102 is inserted before the optimization location candidates acquired in (2).
(5) Determine whether i is m. If it is m, the process is terminated; otherwise, the process proceeds to (6).
(6) Increment i by 1 and return to (3).

The profile program may be generated by the following procedure, for example.
(1) The optimization location candidate determined by the static analysis means 1062 is acquired. Here, it is assumed that m optimization location candidates have been acquired.
(2) Set 1 to the counter i.
(3) The i-th optimized location candidate is acquired from the optimized location candidates acquired in (1).
(4) The i-th optimized location candidate obtained in (3) is copied.
(5) An assembly program instruction sequence for collecting parameters indicated by the parameter information stored in the optimization information storage unit 102 is inserted immediately before the optimization location candidate copied in (4).
(6) An assembly program instruction that jumps immediately before the assembly program instruction sequence inserted in (5) is inserted immediately before the optimization location candidate acquired in (3).
(7) It is determined whether i is m. If it is m, the process is terminated; otherwise, the process proceeds to (8).
(8) Increment i by 1 and return to (3).

  Here, acquisition of the optimization location candidate may be, for example, acquisition of information indicating the optimization location candidate, or acquisition of an assembly program indicated by information indicating the optimization location candidate. May be.

  In addition, the term “before the optimization location candidate” only needs to be in front of the optimization location candidate, and may not be immediately before, but usually it is preferable to be immediately before.

  An instruction sequence is two or more instructions, and an instruction is one instruction.

  Further, the assembly program instruction sequence for collecting the parameters indicated by the parameter information stored in the optimization information storage unit 102 may be accumulated in an arbitrary storage area, or held by the profile program generation unit 1063. It may be.

  In addition, the copy of the optimization location candidate is to accumulate program code corresponding to the location in the assembly program specified by the information indicating the optimization location candidate in an arbitrary storage area.

  The profile program generated by the profile program generation unit 1063 may be generated by any method as long as the profile program can be generated.

  The profile program generation unit 1063 can be usually realized by an MPU, a memory, or the like. The processing procedure of the profile program generation means 1063 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The dynamic analysis unit 1064 executes a profile program and performs dynamic analysis to acquire one or more parameters.

  Here, the profile program is usually executed after the assembly program is converted into a machine language. The profile program may be executed after the profile program is converted into the machine language using an assembler that is a program for converting the assembly program into the machine language. In this case, for example, the assembler may or may not be held by the dynamic analysis means 1064. Otherwise, the assembler may be stored in any storage area.

  The dynamic analysis performed by the dynamic analysis unit 1064 is not limited as long as one or more parameters can be acquired.

  The dynamic analysis unit 1064 can be usually realized by an MPU, a memory, or the like. The processing procedure of the dynamic analysis means 1064 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The determination unit 1065 determines whether or not one or more parameters acquired by the dynamic analysis unit 1064 meet a predetermined condition.

  Here, the predetermined condition is a condition for determining an optimization location candidate that has acquired the parameter as an optimization location. The predetermined condition may be, for example, that the number of loops acquired by the dynamic analysis unit 1064 is greater than or equal to a threshold value, and whether or not the pointers acquired by the dynamic analysis unit 1064 overlap. The information to be displayed may be information indicating that there is no overlap.

  That is, for example, when the predetermined condition is “the number of loops acquired by the dynamic analysis unit 1064 is equal to or greater than the threshold”, the determination unit 1065 determines that the number of loops acquired by the dynamic analysis unit 1064 is It is determined whether or not the threshold value is exceeded. For example, when the predetermined condition is “the information indicating whether or not the pointer acquired by the dynamic analysis unit 1064 overlaps is information indicating that there is no overlap”, the determination unit 1065. Determines whether or not the information indicating whether or not the pointers acquired by the dynamic analysis unit 1064 overlap is information indicating that the pointers do not overlap.

In addition, the pointer overlap means that the address pointed to by any two pointers overlaps the address of the data size secured by one pointer, or is included in the address of the data size secured by one pointer It is to be. The determination of whether or not the pointers overlap may be performed, for example, by the following procedure.
(1) The size of the array starting from the first pointer (Ptr_A) is acquired. Let the size acquired at this time be Size_A.
(2) The address of each element of the array starting from Ptr_A is acquired. Assume that the size_A addresses acquired at this time are Addr_A, and the addresses corresponding to the elements of the array are stored in Addr_A [1], Addr_A [2],..., Addr_A [Size_A]. .
(3) The size of the array starting from the second pointer (Ptr_B) is acquired. Let the size acquired at this time be Size_B.
(4) The address of each element of the array starting from Ptr_B is acquired. The size_B addresses obtained at this time are defined as Bddr_B, and the addresses corresponding to the elements of the array are stored in Bddr_B [1], Bddr_B [2],..., Bddr_B [Size_B]. .
(5) Set false to the flag.
(6) Set 1 to the counter i.
(7) Set 1 to the counter j.
(8) It is determined whether or not the condition “Bddr_B [j] is greater than or equal to Addr_A [i] and smaller than Addr_A [i + 1]” is satisfied.
(9) If the condition is met in (8), true is set in the flag. If the condition is not met, j is incremented by 1, and the process returns to (8).
(10) The processes of (8) to (9) are repeated until j matches Size_B.
(11) Increment i by 1 and return to (7).
(12) The processes of (7) to (11) are repeated until i matches Size_A-1.
(13) It is determined whether or not the flag is true. If the flag is true, it is determined that the pointers overlap. If the flag is false, it is determined that the pointers do not overlap.

  The determination of whether or not the pointers overlap may be any method as long as it can be determined whether or not the pointers overlap.

  The determination performed by the determination unit 1065 may be any method as long as it can be determined whether one or more parameters acquired by the dynamic analysis unit 1064 meet a predetermined condition.

  The determination unit 1065 can be usually realized by an MPU, a memory, or the like. The processing procedure of the determination unit 1065 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  When one or more parameters acquired by the dynamic analysis unit 1064 meet a predetermined condition, the determination unit 1066 determines an optimization location candidate that has acquired the parameter as an optimization location. That is, when the determination unit 1065 determines whether or not one or more parameters acquired by the dynamic analysis unit 1064 match a predetermined condition, and determines that they match, the optimization that acquired the parameter Location candidates are determined as optimization locations.

  The determination unit 1066 can be usually realized by an MPU, a memory, or the like. The processing procedure of the determination unit 1066 is usually realized by software, and the software is recorded in a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The change unit 107 normally acquires change information corresponding to the optimization location detection information used when the optimization location detection unit 106 detects the optimization location from the optimization information storage unit 102, and the change information Change the optimization location using to get the optimized program.

  Here, changing the optimization location using the change information means, for example, that the change information is a program pattern, information indicating a program change method, a program for making a change, or the like. The optimization point is changed.

  The change is a concept including replacement and rewriting.

  Further, the change unit 107 may perform multi-core parallelization on the optimized location determined by the determining unit 1066 as the optimized location, and obtain an optimized assembly program.

  Here, multi-core parallelization includes rewriting to a program code that can be executed in parallel in a plurality of processors, or inserting an instruction to be executed in parallel in a plurality of processors. Multi-core parallelization, for example, for an array having 10 elements, loops that process by turning the counter from 1 to 10, and turns the counter from 1 to 5 to process two elements at a time. It may be changed to a loop that can be performed. In addition, multi-core parallelization is, for example, for executing a loop in parallel with a loop that performs processing by rotating a counter from 1 to 10 for an array having 10 elements, immediately before the loop. For example, an instruction may be inserted. In addition, multi-core parallelization, for example, for an array having 10 elements, for a loop that processes by rotating the counter from 1 to 10, an instruction for starting parallel execution immediately before the loop. It may be inserted and an instruction for ending parallel execution is inserted immediately after the loop.

  Further, if the number of program codes divided in multi-core parallelization is referred to as a parallel number, the parallel number may or may not be determined in advance. If determined in advance, the parallel number may be stored in an arbitrary storage area. If not determined in advance, the parallel number may be, for example, the maximum number that does not exceed the number of processors among the number of processors or a divisor of the maximum value of the loop counter. The changing unit 107 may acquire the parallel number when it is not determined in advance.

Multi-core parallelization is, for example, the following processing.
(1) A loop that performs multi-core parallelization is detected.
(2) The maximum value of the loop counter of the loop detected in (1) is acquired.
(3) Obtain the number of processors.
(4) A value obtained by dividing the maximum value of the loop counter acquired in (2) by the number of processors acquired in (3) is acquired. Here, the divided value, that is, the quotient is an integer value, not a decimal value. Further, the remaining value without being divisible, that is, the remainder may be acquired.
(5) The value acquired in (4) is set to the maximum value of the loop counter of the loop detected in (1).
(6) Copy the program code indicating the processing for the array in the loop detected in (1).
(7) The program code copied in (6) is inserted immediately after the copy source program code.
(8) In the process for the array indicated by the program code inserted in (7), the variable for accessing the element of the array is rewritten to “+1”. That is, for example, if “Array [i]”, it is rewritten to “Array [i + 1]”.

  In addition, when a remainder exists in (4) and can be acquired, a program code for performing processing on an array not processed by the remainder may be inserted immediately after the loop.

Further, multi-core parallelization may be the following processing, for example.
(1) A loop that performs multi-core parallelization is detected.
(2) A program code (instruction) for starting parallel execution is inserted immediately before the loop detected in (1).
(3) Immediately after the loop detected in (1), a program code (instruction) for ending parallel execution is inserted.

For example, when the following processing is performed in each unit included in the optimization location detection unit 106, the change unit 107 performs the optimization on the optimization location finally determined by the determination unit 1066. Multi-core parallelization may be performed.
(1) The optimization location candidate detection means 1061 detects an optimization location candidate.
(2) The static analysis means 1062 acquires the loop invariant value and the induction variable from the optimization location candidate detected in (1), and uses these pieces of information to determine the sequence for the optimization location candidate. It is determined whether or not memory storage is being performed, and it is determined that memory storage is being performed.
(3) For the optimization location candidate determined to be performed in (2), the profile program generation unit 1063 obtains information indicating whether the number of loops and the pointer overlap. Insert a profile code and generate a profile program.
(4) The dynamic analysis unit 1064 executes the profile program generated in (3), and obtains information indicating the number of loops and whether or not the pointers overlap.
(5) With respect to the loop count acquired in (4) and information indicating whether or not the pointer overlaps, the determination unit 1065 determines that the loop count is equal to or greater than the threshold and the pointer overlaps. It is determined that the information indicating whether or not to perform is information indicating that the pointers do not overlap.
(6) According to the result determined in (5), the determination unit 1066 determines the optimization location candidate that has acquired information indicating whether the number of loops and the pointer overlap or not as the optimization location.

  The changing unit 107 may obtain an optimized assembly program by applying SIMD to the optimized location determined by the determining unit 1066 as the optimized location.

  Here, SIMD conversion refers to changing one or more assembly program instructions that process one data with one instruction into assembly program instructions that can simultaneously process multiple data with one instruction. . Specifically, for example, replacing four identical instructions that assign values to 4-byte variables with instructions that simultaneously assign values to four 4-byte variables using a 16-byte register, etc. It is.

  In addition, if the number of instructions to be replaced with one instruction in SIMD conversion is called an integrated number, the integrated number may be determined in advance or not. If determined in advance, the integration number may be stored in an arbitrary storage area. If not determined in advance, the integration number may be, for example, the number of processors or a number calculated based on the size of a variable type (number of bits or number of bytes). The changing unit 107 may acquire the number of integrations when the number is not determined in advance.

Further, SIMD conversion is, for example, the following processing.
(1) A loop for applying SIMD is detected.
(2) Obtain the number of bytes that can be handled by the instruction to be replaced by SIMD conversion.
(3) The integration number n is acquired.
(4) A value obtained by dividing the number of bytes acquired in (2) by the integrated number n is acquired.
(5) In the loop detected in (1), program code for performing processing for assigning a value to a variable having the value acquired in (4) as the number of bytes is acquired as a set of n or less.
(6) A program that performs processing of assigning values to n variables when there is no dependency among n variables that are processed by n program codes acquired in (5) Generate code.
(7) Delete the part corresponding to the program code acquired in (5).
(8) The program code generated in (6) is inserted into the location corresponding to the first program code among the locations deleted in (7).

For example, when the following processing is performed in each unit included in the optimization location detection unit 106, the change unit 107 performs the optimization on the optimization location finally determined by the determination unit 1066. SIMD may be applied.
(1) The optimization location candidate detection means 1061 detects an optimization location candidate.
(2) The static analysis unit 1062 obtains a variable name from the optimization location candidate detected in (1), and using this information, there is a dependency between variables in the optimization location candidate. Whether or not it exists.
(3) For the optimization location candidate determined not to exist in (2), the profile program generation unit 1063 obtains information indicating whether the number of loops and the pointer overlap. Insert a profile code and generate a profile program.
(4) The dynamic analysis unit 1064 executes the profile program generated in (3), and obtains information indicating the number of loops and whether or not the pointers overlap.
(5) With respect to the loop count acquired in (4) and information indicating whether or not the pointer overlaps, the determination unit 1065 determines that the loop count is equal to or greater than the threshold and the pointer overlaps. It is determined that the information indicating whether or not to perform is information indicating that the pointers do not overlap.
(6) According to the result determined in (5), the determination unit 1066 determines the optimization location candidate that has acquired information indicating whether the number of loops and the pointer overlap or not as the optimization location.

  Further, the change unit 107 acquires change information corresponding to the optimization location detection information used when the optimization location detection unit 106 detects the optimization location from the optimization information storage unit 102, and the change information The optimized program may be obtained by replacing the optimized part with.

  Here, changing the optimization location with the change information means replacing the optimization location with the change information. That is, for example, when the optimization detection information is an instruction “MOVE” and the change information is an instruction “SMOVE”, “MOVE” is replaced with “SMOVE”.

  Moreover, the change part 107 may change an optimization location according to the optimization location detection information used when detecting an optimization location. For example, when the optimization location detection information is information for detecting a location to be subjected to multi-core parallelization, the changing unit 107 performs the optimization on the optimization location detected by the optimization location detection information. Multi-core parallelization is performed using the change information corresponding to the conversion location detection information. Further, for example, when the optimization location detection information is information for detecting a location to be converted to SIMD, the changing unit 107 applies the optimization location detected by the optimization location detection information to the optimization location. SIMD parallelization is performed using the change information corresponding to the optimized location detection information. For example, when the optimization location detection information is information for detecting a location to be replaced, the changing unit 107 changes the optimization location detected by the optimization location detection information to the optimization location. Replace with the change information corresponding to the detection information.

  Moreover, the change part 107 may change an optimization location according to the change information corresponding to optimization location detection information. For example, when the change information is information indicating a method for performing multi-core parallelization, the change unit 107 performs multi-core parallelization on the detected optimization location using the change information. For example, when the change information is information indicating a method for performing SIMD conversion, the changing unit 107 performs SIMD conversion on the detected optimization location using the change information. For example, when the change information is information for replacement, the change unit 107 replaces the detected optimization location with the change information.

  For example, when the optimization location information includes change method information indicating a method for changing the optimization location, the optimization location may be changed according to the change method information. For example, when the change method information is information indicating that multi-core parallelization is to be performed, the change unit 107 performs multi-core parallelization on the detected optimization location using the change information. Further, for example, when the change method information corresponding to the change information is information indicating that SIMD conversion is to be performed, the change unit 107 performs SIMD conversion on the detected optimization location using the change information. Apply. For example, when the change method information corresponding to the change information is information indicating that replacement is to be performed, the change unit 107 replaces the detected optimization location with the change information.

  The changing unit 107 can be usually realized by an MPU, a memory, or the like. The processing procedure of the changing unit 107 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The execution unit 108 executes the optimized assembly program.

  Here, the assembly program is usually executed after the assembly program is converted into machine language. The assembly program may be executed after the assembly program is converted into the machine language by using an assembler that is a program for converting the assembly program into the machine language. In this case, for example, the execution unit 108 may or may not hold the assembler. Otherwise, the assembler may be stored in any storage area.

  The execution unit 108 can usually be realized by an MPU, a memory, or the like. The processing procedure of the execution unit 108 is usually realized by software, and the software is recorded on a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  Next, the operation of the execution module optimization device 1 will be described.

  FIG. 7 is a flowchart showing the overall operation of the execution module optimizing apparatus 1.

  (Step S701) The receiving unit 104 determines whether an instruction having an execution module identifier has been received. If accepted, the process proceeds to step S702; otherwise, the process returns to step S701.

  (Step S <b> 702) The disassembly unit 105 acquires an execution module stored in the execution module storage unit 101. The execution module acquired here is an execution module identified by an execution module identifier included in the instruction received by the receiving unit 104 in step S701.

  (Step S703) The disassembly unit 105 disassembles the execution module acquired in step S702, and acquires an assembly program.

  (Step S704) The optimization location detection part 106 detects an optimization location from the assembly program which the disassembly part 105 acquired by step S703. Details of this processing will be described with reference to the flowchart of FIG.

  (Step S705) The changing unit 107 changes the optimization location and acquires an optimized assembly program. Details of this processing will be described with reference to the flowchart of FIG.

  (Step S706) The execution unit 108 executes the assembly program acquired by the changing unit 107.

  FIG. 8 is a flowchart showing the optimization location detection processing in step S304 of the flowchart of FIG.

  (Step S801) The optimization location candidate detection means 1061 acquires the assembly program acquired by the disassembly unit 105 in step S703.

  (Step S802) The optimization location candidate detection means 1061 acquires the optimization information stored in the optimization information storage unit 102. Here, it is assumed that m pieces of optimization information have been acquired.

  (Step S803) The optimization location candidate detection means 1061 sets 1 to the counter i.

  (Step S804) The optimization location candidate detection unit 1061 detects a location that matches the i-th optimization location detection information (hereinafter, optimization location detection information [i]) from the assembly program acquired in Step S801. Here, it is assumed that n matching locations have been detected. Details of this processing will be described with reference to the flowchart of FIG.

  (Step S805) The optimization location candidate detection means 1061 determines whether i is m. When it is m, it progresses to step S807, and when that is not right, it progresses to step S806.

  (Step S806) The optimization location candidate detection means 1061 increments the counter i by 1, and returns to step S804.

  (Step S807) The optimization location candidate detection means 1061 sets 1 to the counter i.

  (Step S808) The optimization location candidate detection means 1061 sets 0 to the counter j.

  (Step S809) The optimization location candidate detection means 1061 sets 0 to the counter k.

  (Step S810) The optimization location candidate detection means 1061 judges whether the i-th matching location (henceforth a matching location [i]) is a loop among the matching locations detected by step S804. . When it is a loop, it progresses to step S811, and when that is not right, it progresses to step S813.

  (Step S811) The optimization location candidate detection means 1061 increments j by 1.

  (Step S812) The optimization location candidate detection means 1061 accumulates the matching location [i] in an arbitrary storage area as the j-th loop.

  (Step S813) The optimization location candidate detection means 1061 increments k by 1.

  (Step S814) The optimization location candidate detection unit 1061 accumulates the matching location [i] as an kth optimization location in an arbitrary storage area.

  (Step S815) The optimization location candidate detection means 1061 determines whether i is n. When it is n, it progresses to step S817, and when that is not right, it progresses to step S816.

  (Step S816) The optimization location candidate detection means 1061 increments i by 1, and returns to step S810.

  (Step S817) The optimization location candidate detection means 1061 detects an optimization location from the loop accumulate | stored at step S812, and returns to a high-order process. Details of this processing will be described with reference to the flowchart of FIG.

  FIG. 9 is a flowchart showing a process for detecting a location that matches the optimized location information in step S804 of the flowchart of FIG.

  (Step S901) The optimization location candidate detection means 1061 sets 1 to the counter i.

  (Step S902) The optimization location candidate detection means 1061 determines whether i is larger than the value obtained by subtracting the number of rows m of the optimization location detection information from the number of rows n of the assembly program. If it is larger, the process returns to the upper process, and if not, the process proceeds to step S903.

  (Step S903) The optimization location candidate detection unit 1061 reads the code on the i-th line (hereinafter referred to as code [i]) in the assembly program and the code pattern on the first row in the optimization location detection information (hereinafter referred to as code pattern) [1]) to determine whether or not the code matches the code pattern. If they match, the process proceeds to step S905, and if not, the process proceeds to step S904.

  (Step S904) The optimization location candidate detection means 1061 increments i by 1, and returns to step S902.

  (Step S905) The optimization location candidate detection means 1061 sets i to the start position start of the matching location.

  (Step S906) The optimization location candidate detection means 1061 sets false to the flag flag.

  (Step S907) The optimization location candidate detection means 1061 sets 1 to the counter j.

  (Step S908) The optimization location candidate detection unit 1061 compares the code [start + j-1] with the code pattern [j] and determines whether or not the code matches the code pattern. If they match, the process proceeds to step S909; otherwise, the process proceeds to step S910.

  (Step S909) The optimization location candidate detection means 1061 sets true to flag, and progresses to step S911.

  (Step S910) The optimization location candidate detection means 1061 sets false to flag, and progresses to step S911.

  (Step S911) The optimization location candidate detection means 1061 determines whether j is m. When it is m, it progresses to step S913, and when that is not right, it progresses to step S912.

  (Step S912) The optimization location candidate detection means 1061 increments j by 1, and returns to step S908.

  (Step S913) The optimization location candidate detection means 1061 judges whether flag is true. If true, the process proceeds to step S915, and if not, the process proceeds to step S914.

  (Step S914) The optimization location candidate detection means 1061 sets start + 1 to i, and returns to step S902.

  (Step S915) The optimization location candidate detection means 1061 sets start + m-1 to the end position end of the matching location.

  (Step S916) The optimization location candidate detection means 1061 sets the number of matching locations detected already + 1 to the counter k.

  (Step S917) The optimization location candidate detection means 1061 accumulate | stores start and end in arbitrary storage areas as a kth matching location.

  FIG. 10 is a flowchart showing an optimization location detection process from the loop in step S817 in the flowchart of FIG.

  (Step S1001) The optimization location candidate detection means 1061 determines whether or not to detect an optimization location candidate using a profiler. If so, the process proceeds to step S1002, and if not, the process proceeds to step S1003.

  (Step S1002) The optimization location candidate detection means 1061 detects an optimal location candidate using a profiler. Here, it is assumed that m optimization location candidates have been detected. Details of this processing will be described with reference to the flowchart of FIG.

  (Step S1003) The optimization location candidate detection means 1061 detects an optimization location candidate using a control flow graph. Here, it is assumed that m optimization location candidates have been detected. Details of this processing will be described with reference to the flowchart of FIG.

  (Step S1004) The static analysis means 1062 performs a static analysis on each optimization location candidate detected by the optimization location candidate detection means 1061 in step S1002 or step S1003, and determines an optimization location candidate.

  (Step S1005) The profile program generation means 1063 inserts an assembly program instruction sequence before each optimization location candidate determined in step S1004, and generates a profile program.

  (Step S1006) The dynamic analysis means 1064 executes the profile program generated in step S1005, performs dynamic analysis, and acquires parameters.

  (Step S1007) The judgment means 1065 sets 1 to the counter i.

  (Step S1008) The determination unit 1066 sets the number of optimization points already detected in the counter j.

  (Step S1009) The determination unit 1065 determines whether the parameter acquired in step S1006 matches a predetermined condition, thereby determining the i-th optimized location candidate (hereinafter referred to as the optimized location candidate [ i]) is an optimization location. If the predetermined condition is met and the location is an optimization location, the process proceeds to step S1010. Otherwise, the process proceeds to step S1012.

  (Step S1010) The determination unit 1066 increments j by 1.

  (Step S1011) The determination unit 1066 accumulates the optimization location candidate [i] in an arbitrary storage area as the j-th optimization location.

  (Step S1012) The judgment means 1065 judges whether i is m. If m, the process returns to the upper process, and if not, the process proceeds to step S1013.

  (Step S1013) The judgment means 1065 increments i by 1, and returns to step S1009.

  FIG. 11 is a flowchart showing an optimization location detection process using the profiler in step S1002 of the flowchart of FIG.

  (Step S1101) The optimization location candidate detection means 1061 acquires the profiler stored in the profiler storage unit 103.

  (Step S1102) The optimization location candidate detection means 1061 executes the profiler acquired in Step S1102 on the assembly program acquired in Step S801, and acquires profile information.

  (Step S1103) The optimization location candidate detection means 1061 acquires a threshold for determining whether or not the load in the profile information exceeds the threshold from an arbitrary storage area.

  (Step S1104) The optimization location candidate detection means 1061 sets 1 to the counter i.

  (Step S1105) The optimization location candidate detection means 1061 sets 0 to the counter j.

  (Step S1106) The optimization location candidate detection means 1061 acquires the load corresponding to the i-th loop (hereinafter referred to as loop [i]) acquired in step S812 in the profile information acquired in step S1102. Is greater than the threshold value acquired in step S1103. If so, the process proceeds to step S1107; otherwise, the process proceeds to step S1109.

  (Step S1107) The optimization location candidate detection means 1061 increments j by 1.

  (Step S1108) The optimization location candidate detection means 1061 accumulate | stores loop [i] in arbitrary storage areas as a jth optimization location candidate.

  (Step S1109) The optimization location candidate detection means 1061 determines whether i is the number n of loops acquired in step S812. When it is n, it returns to a high-order process, and when that is not right, it progresses to step S1110.

  (Step S1110) The optimization location candidate detection means 1061 increments i by 1, and returns to step S1105.

  FIG. 12 is a flowchart showing an optimization location detection process using the control flow graph in step S1003 of the flowchart of FIG.

  (Step S1201) The optimization location candidate detection means 1061 creates a control flow graph from the assembly program acquired in step S801. Details of this processing will be described with reference to the flowchart of FIG.

  (Step S1202) The optimization location candidate detection means 1061 analyzes the dominance relationship between each node about the control flow graph created by step S1201, and acquires the analysis result.

  (Step S1203) The optimization location candidate detection means 1061 sets 1 to the counter i.

  (Step S1204) The optimization location candidate detection means 1061 sets 0 to the counter j.

  (Step S1205) The optimization location candidate detection means 1061 determines whether or not the i-th loop acquired in step S812 (hereinafter, loop [i]) is a dominating loop based on the dominating relationship acquired in step S1202. to decide. If so, the process proceeds to step S1206. Otherwise, the process proceeds to step S1208.

  (Step S1206) The optimization location candidate detection means 1061 increments j by 1.

  (Step S1207) The optimization location candidate detection means 1061 accumulate | stores loop [i] in arbitrary storage areas as a jth optimization location candidate.

  (Step S1208) The optimization location candidate detection means 1061 determines whether i is the number n of loops acquired in step S812. When it is n, it returns to a high-order process, and when that is not right, it progresses to step S1209.

  (Step S1209) The optimization location candidate detection means 1061 increments i by 1, and returns to step S1204.

  FIG. 13 is a flowchart showing a control flow graph creation process in step S1201 of the flowchart of FIG.

  (Step S1301) The optimization location candidate detection means 1061 sets 0 to the counter n.

  (Step S1302) The optimization location candidate detection means 1061 sets 1 to the counter i.

  (Step S1303) The optimization location candidate detection means 1061 sets i + 1 to the counter j.

  (Step S1304) The optimization location candidate detection unit 1061 receives the instruction [j] of the assembly program immediately after the instruction in the i-th line code of the assembly program acquired in Step S801 (hereinafter, instruction [i]). Determine if it will be executed. If so, the process proceeds to step S1305; otherwise, the process proceeds to step S1307. This determination is normally made when the instruction [j] is executed immediately after the instruction [i] if the instruction [j] is the next line of the instruction [i]. Further, for example, when the instruction [i] is a branch instruction such as a jump instruction and the instruction [j] is executed immediately after the branch destination by the instruction [i], the instruction [j] It is determined to be executed immediately after [i].

  (Step S1305) The optimization location candidate detection means 1061 increments n by 1.

  (Step S1306) The optimization location candidate detection unit 1061 accumulates the instruction [i] and the instruction [j] in an arbitrary storage area as the n-th node set.

  (Step S1307) The optimization location candidate detection means 1061 determines whether j is the number m of lines of the assembly program acquired in step S801. When it is m, it progresses to step S1309, and when that is not right, it progresses to step S1308.

  (Step S1308) The optimization location candidate detection means 1061 increments j by 1, and returns to step S1304.

  (Step S1309) The optimization location candidate detection means 1061 determines whether i is m-1. If it is m−1, the process returns to the upper process, and if not, the process proceeds to step S1310.

  (Step S1310) The optimization location candidate detection means 1061 increments i by 1, and returns to step S1303.

  FIG. 14 is a flowchart showing the optimization location change processing in step S705 of the flowchart of FIG.

  (Step S1401) The changing unit 107 acquires optimization locations accumulated in an arbitrary storage area.

  (Step S1402) The changing unit 107 sets 1 to the counter i.

  (Step S <b> 1403) The changing unit 107 displays the change information corresponding to the optimized location detection information used when detecting the i-th optimized location (hereinafter referred to as the optimized location [i]) as the optimized information storage unit 102. Get from.

  (Step S1404) The changing unit 107 changes the optimization location [i] based on the change information acquired in step S1403, and returns to the upper process.

  In the flowchart of FIG. 10, the optimization location candidate detection process in steps S1002 and S1003 is performed without performing the determination in step S1001, and the optimization location candidates detected in both the processes are compared, and overlapping optimization is performed. The optimized location candidate may be detected again as an optimized location candidate.

  Next, a specific example of the operation of the execution module optimization device 1 will be described.

  In this specific example, it is assumed that the optimization information storage unit 102 stores the optimization information of FIG.

  In the optimization information of FIG. 15, the first and second optimization location detection information are program patterns. In the optimization information, the third to fifth optimization location detection information is a function.

  In the optimization information of FIG. 15, the first change information is a program pattern for performing multi-core parallelization, and the second change information is a program pattern for performing SIMD. In the optimization information, the third to fifth change information is a function.

  In the optimization location detection information of FIG. 15, “<label>” is a pattern indicating an arbitrary character string, and “<...>” Is a pattern indicating an arbitrary program code. In FIG. 15, “<var>”, “<var1>”, “<var2>”, etc. are patterns indicating an arbitrary character string (variable), and “<count>” is an arbitrary numerical value. It is a pattern to show.

  In the change information in FIG. 15, “$ code” is a variable indicating the program code corresponding to the optimization location detected by the optimization location detection information that is a program pattern, and “$ label” is “< This is a variable indicating a character string that matches the pattern “label>”. In the change information of FIG. 15, “$ var”, “$ var1”, “$ var2”, etc. are characters that match a pattern such as “<var>”, “<var1>”, “<var2>”, etc. This is a pattern indicating a column (variable), and “$ count” is a variable indicating a numerical value matching the pattern “<count>”.

  In addition, the optimization location candidate detection unit 1061 uses a profiler to determine whether a loop that can be subjected to multi-core parallelization among the detected loops can be an optimization location candidate. It shall be detected as an optimization location candidate. In addition, the optimization location candidate detection unit 1061 uses the control flow graph to determine whether a loop that can be subjected to SIMD among the detected loops can be an optimization location candidate. Are detected as optimization location candidates. Further, when the determining unit 1066 determines the optimization location, the condition for determining the parameter in the determining unit 1065 is “the number of loops is 100 or more” and “the pointers do not overlap”. To do.

  First, the user drags and drops an execution module to be optimized onto the execution module optimization apparatus 1 using a mouse. Then, the reception unit 104 receives an instruction to execute an execution module having an execution module identifier.

  Next, the disassembly unit 105 acquires the execution module identified by the execution module identifier included in the instruction received by the reception unit 104 from the execution module storage unit 101. Then, it is assumed that the disassembly unit 105 disassembles the execution module and obtains the assembly program of FIG.

  Next, the optimization location candidate detection means 1061 acquires the optimization information of FIG. 15, and detects a location that matches the optimization location detection information included in the optimization information from the assembly program of FIG. As a result, the 8th to 15th lines are matched with the first optimized part detection information, the 18th line to the 31st line, the 3rd line are matched with the second optimized part detection information. It is assumed that the 39th line is detected as a location that matches the optimized location detection information.

  Next, the optimization location candidate detection means 1061 determines whether or not a location that matches each detected optimization location detection information is a loop, and the 8th to 15th and 18th rows in FIG. The 31st line is judged to be a loop, and the 39th line is judged not to be a loop. As a result, the optimized location detection unit 1061 accumulates the 39th row as an optimized location in an arbitrary storage area. Of the portions determined to be loops, the 8th to 15th lines are the first loop, and the 18th to 31st lines are the second loop.

  Next, the optimization location candidate detection unit 1061 determines from the change information in FIG. 15 and acquires the load of each instruction in the first loop using the profiler for the first loop. And the first loop is detected as an optimization location candidate.

  Next, the optimization location candidate detection unit 1061 determines from the change information in FIG. 15, creates a control flow graph for the second loop, analyzes the dominating relationship between the nodes, and the second loop is the dominating loop. The second loop is detected as an optimization location candidate.

  Next, the static analysis unit 1062 performs a static analysis on each optimization location candidate detected by the optimization location candidate detection unit 1061. For the first loop, the static analysis means 1062 obtains a loop invariant value and an induction variable for accessing the array, determines from these values that memory storage is being performed for the array, One loop is determined as an optimization location candidate. Further, the static analysis unit 1062 acquires a variable name for the second loop, determines that there is no dependency relationship between the variables from these values, and determines the second loop as an optimization location candidate.

  Next, the profile program generation means 1063 inserts an instruction sequence for collecting information indicating whether or not the number of loops and the pointer overlap with each other immediately before the first loop and the second loop in the assembly program of FIG. Assume that the profile program of FIG. 17 is generated. In FIG. 17, lines 8 and 20 are instructions for acquiring the number of loops, and lines 9 and 21 are instructions for acquiring information indicating whether or not the pointers overlap. It is.

  Next, the dynamic analysis unit 1064 executes the profile program of FIG. 17 and acquires “100” as the number of loops and information indicating that the pointers do not overlap for the first loop and the second loop. Here, the parameter acquired for the first loop is the first parameter, and the parameter acquired for the second loop is the second parameter.

  Next, the determination unit 1065 determines whether or not the first parameter and the second parameter match the conditions for optimization, and determines that both the first parameter and the second parameter match the conditions.

  Next, the determination unit 1066 determines the first loop and the second loop, which are the loops that have acquired the first parameter and the second parameter, as optimization locations. Here, the optimization location corresponding to the first loop is the first optimization location, and the optimization location corresponding to the second loop is the second optimization location. In addition, the optimization location that is the location where the above replacement is performed is set as the third optimization location.

Next, the changing unit 107 performs multi-core parallelization on the first optimization location, for example, by the following processing.
(1) The changing unit 107 acquires the program codes from the 8th line to the 15th line in FIG. 16 corresponding to the first optimization location.
(2) The change unit 107 acquires the first change information in FIG.
(3) The change unit 107 replaces “$ code” in the change information acquired in (2) with the program code acquired in (1), and generates a changed program code.
(4) The changing unit 107 replaces the program code from the 8th line to the 15th line in FIG. 16 corresponding to the first optimization location with the program code generated in (3).
It is assumed that the assembly code at the first optimization location has been changed to that shown in FIG. 18 by the above processing.

Further, the changing unit 107 applies SIMD to the second optimization location by the following process, for example.
(1) The change unit 107 acquires program codes from the 18th line to the 31st line in FIG. 16 corresponding to the second optimization location.
(2) The changing unit 107 uses the program code acquired in (1) to match a character string that matches a pattern such as “<label>” or “<var>” in the second optimization location change information in FIG. , Get the variable.
(3) The change unit 107 acquires the second change information in FIG.
(4) The change unit 107 replaces “$ label”, “$ var”, etc. in the change information acquired in (3) with the character string and variable acquired in (2), and changes the program code after the change. Generate.
(5) The changing unit 107 replaces the program code from the 18th line to the 31st line in FIG. 16 corresponding to the second optimization location with the program code generated in (4).
It is assumed that the assembly code at the second optimization location has been changed to that shown in FIG. 19 by the above processing.

  In addition, the changing unit 107 performs replacement for the third optimization location. As a result, it is assumed that the assembly code at the third optimization location is changed to that shown in FIG.

  As a result, it is assumed that the changing unit 107 has acquired the optimized assembly code of FIG.

  Finally, the execution unit 108 executes the assembly program of FIG.

  In the present embodiment, the changing unit 107 has already optimized the execution module identifier included in the instruction received by the receiving unit 104 and the execution module identified by the execution module identifier when the assembly program is optimized. The optimized information indicating that the information has been converted to the optimized assembly program may be associated with each other and stored in an arbitrary storage area. In this case, the accepting unit 104 determines whether or not the execution module identified by the execution module identifier included in the accepted instruction has been optimized based on the optimized information stored in an arbitrary storage area. It may be judged. In this case, if the receiving unit 104 determines that the execution module has been optimized, the execution module optimizing device 1 does not perform the optimization process by the disassembly unit 105 or the like, and the execution unit 108 The optimized assembly program corresponding to the execution module identifier included in the instruction received by the receiving unit 104 may be acquired from an arbitrary storage area, and the assembly program may be executed.

  Further, in the present embodiment, when the changing unit 107 optimizes the assembly program, the execution module identifier included in the instruction received by the receiving unit 104 and the execution module identified by the execution module identifier are already optimal. Assembling the optimized assembly program indicating that it has been optimized and the optimized assembly program, obtaining an execution module, associating the optimized execution module, and storing it in an arbitrary storage area Good. In this case, the accepting unit 104 determines whether or not the execution module identified by the execution module identifier included in the accepted instruction has been optimized based on the optimized information stored in an arbitrary storage area. It may be judged. In this case, if the receiving unit 104 determines that the execution module has been optimized, the execution module optimizing device 1 does not perform the optimization process by the disassembly unit 105 or the like, and the execution unit 108 The optimized execution module corresponding to the execution module identifier included in the instruction received by the reception unit 104 may be acquired from an arbitrary storage area, and the execution module may be executed.

  Further, in the present embodiment, when the changing unit 107 optimizes the assembly program, the execution module identifier included in the instruction received by the receiving unit 104 and the execution module identified by the execution module identifier are already optimal. Optimized information indicating that it has been optimized, accumulated in an arbitrary storage area, assembled an optimized assembly program, obtained an execution module, and received the optimized execution module The execution module may be replaced with an execution module identified by an execution module identifier included in the instruction received by the unit 104. In this case, the accepting unit 104 determines whether or not the execution module identified by the execution module identifier included in the accepted instruction has been optimized based on the optimized information stored in an arbitrary storage area. It may be judged. In this case, if the receiving unit 104 determines that the execution module has been optimized, the execution module optimizing device 1 does not perform the optimization process by the disassembly unit 105 or the like, and the execution unit 108 The execution module identified by the execution module identifier included in the instruction received by the reception unit 104 may be executed.

  As described above, according to the execution module optimizing apparatus 1 according to the present embodiment, it is possible to acquire assembly code from an execution module and optimize the assembly code.

(Embodiment 2)
In the present embodiment, an execution module optimizing device 2 that optimizes an execution module will be described. Unlike the execution module optimization apparatus 1 of the first embodiment, the execution module optimization apparatus 2 in the present embodiment does not execute the optimized assembly program. The execution module optimizing device 2 only optimizes an assembly program or an execution module.

  FIG. 22 is a block diagram of the execution module optimizing apparatus 2 in the present embodiment.

  The execution module optimization apparatus 2 includes an execution module storage unit 201, an optimization information storage unit 202, a reception unit 203, a disassembly unit 204, an optimization location detection unit 205, a change unit 206, a compilation unit 207, and an accumulation unit 208. .

  Note that the execution module storage unit 201, the optimization information storage unit 202, the reception unit 203, the disassembly unit 204, the optimization location detection unit 205, and the change unit 206 are the same as those in the first embodiment, so the description thereof will be given. Omitted.

  The compiling unit 207 compiles the optimized assembly program acquired by the changing unit 206 and acquires the optimized execution module.

  Here, compiling means converting an assembly program into an execution module. Since the compilation is a known technique, a detailed description is omitted.

  Further, the compiling unit 207 may compile an optimized assembly program and obtain an optimized execution module corresponding to the optimized assembly program, for example, using a compiler that is a compiling program. Good. In this case, for example, the compiler may or may not be held by the compiling unit 207. Otherwise, the compiler may be stored in any storage area.

  The compiling unit 207 can usually be realized by an MPU, a memory, or the like. The processing procedure of the compiling unit 207 is usually realized by software, and the software is recorded in a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  The accumulation unit 208 accumulates the optimized assembly program acquired by the changing unit 206 in an arbitrary storage area. Further, the storage unit 208 may store the optimized execution module acquired by the compiling unit 207 in an arbitrary storage area.

  Further, the storage unit 208 may store the optimized execution module by overwriting the execution module before the optimization.

  The storage unit 208 can usually be realized by an MPU, a memory, or the like. The processing procedure of the storage unit 208 is usually realized by software, and the software is recorded in a recording medium such as a ROM. However, it may be realized by hardware (dedicated circuit).

  A specific example of the optimization process of the execution module optimization device 2 and the optimization process of the execution module optimization device 2 is the same as that of the first embodiment, and thus the description thereof is omitted.

  As described above, according to the execution module optimizing apparatus 2 according to the present embodiment, it is possible to acquire assembly code from an execution module and optimize the assembly code. Further, according to the execution module optimizing apparatus 2 according to the present embodiment, the execution module can be optimized.

  In the present embodiment, the changing unit 206 has already optimized the execution module identifier included in the instruction received by the receiving unit 203 and the execution module identified by the execution module identifier when the assembly program is optimized. The optimized information indicating that the information has been converted to the optimized assembly program may be associated with each other and stored in an arbitrary storage area. In this case, the accepting unit 203 determines whether or not the execution module identified by the execution module identifier included in the accepted instruction has been optimized based on the optimized information accumulated in an arbitrary storage area. It may be judged. In this case, when the receiving unit 203 determines that the execution module has been optimized, the execution module optimizing device 2 does not perform the optimization process by the disassembly unit 204 or the like.

  Further, in each of the above embodiments, it goes without saying that two or more communication units existing in one apparatus may be physically realized by one medium.

  The execution module optimizing device in each of the above embodiments may be, for example, a stand-alone device or a server device in a server / client system. In the latter case, for example, the execution module may be optimized as an ASP (Application Service Provider).

  In each of the above embodiments, each process or each function may be realized by centralized processing by a single device or a single system, or distributed by a plurality of devices or a plurality of systems. It may be realized by being processed.

  In each of the above embodiments, each component may be configured by dedicated hardware, or a component that can be realized by software may be realized by executing a program. For example, each component can be realized by a program execution unit such as a CPU reading and executing a software program recorded on a recording medium such as a hard disk or a semiconductor memory.

  The software that implements the execution module optimizing device in each of the above embodiments is the following program. In other words, this program stores, on the recording medium, one or more execution modules, optimization location detection information that is information for detecting an optimization location that is a location where processing can be optimized, and the optimization location detection. One or more pieces of optimization information having change information which is information for changing the optimization location detected using the information is stored, and an instruction having an execution module identifier for identifying the execution module A receiving unit for receiving the program, an execution module identified by the execution module identifier included in the instruction, reading from the recording medium, disassembling, and obtaining an assembly program corresponding to the execution module, and storing in the recording medium One or more optimization locations are detected from the assembly program using one or more optimization location detection information. To obtain from the recording medium change information corresponding to the optimized location detection information used when detecting each optimized location for the optimized location detection unit and the one or more optimized locations, It is a program for functioning as an execution unit for executing the optimized assembly program, and a change unit for changing the one or more optimization locations using change information and obtaining an optimized assembly program.

  The program may be executed by being downloaded from a server or the like, or a program recorded on a predetermined recording medium (for example, an optical disk such as a CD-ROM, a magnetic disk, a semiconductor memory, or the like) is read out. May be executed. Further, this program may be used as a program constituting a program product.

  Moreover, the computer which performs the said program may be single, and plural may be sufficient as it. That is, centralized processing may be performed, or distributed processing may be performed.

  FIG. 23 is an overview of the computer system 9 that executes the above-described program to realize the execution module optimizing device and the like of the above-described embodiment. The above-described embodiments can be realized by computer hardware and a computer program executed thereon.

  In FIG. 23, the computer system 9 includes a computer 901 including a CD-ROM (Compact Disk Read Only Memory) drive 9011, an FD (Flexible Disk) drive 9012, a keyboard 902, a mouse 903, and a monitor 904.

  FIG. 24 is a block diagram of the computer system 9. 24, in addition to the CD-ROM drive 9011 and the FD drive 9012, a computer 901 includes a CPU (Central Processing Unit) 9013 and a ROM (Read-Only Memory) 9014 for storing a program such as a boot-up program. A random access memory (RAM) 9015 connected to the CPU 9013 for temporarily storing application program instructions and providing a temporary storage space; and a hard disk 9016 for storing application programs, system programs, and data , A CD-ROM drive 9011, an FD drive 9012, a CPU 9013, and the like. Although not shown here, the computer 901 may further include a network card that provides connection to a LAN.

  A program that causes the computer system 9 to execute functions such as the execution module optimization device of the above-described embodiment is stored in the CD-ROM 9101 or FD 9102, inserted into the CD-ROM drive 9011 or FD drive 9012, and It may be transferred to the hard disk 9016. Alternatively, the program may be transmitted to the computer 901 via a network (not shown) and stored in the hard disk 9016. The program is loaded into the RAM 9015 when executed. The program may be loaded directly from the CD-ROM 9101, the FD 9102, or the network.

  The program does not necessarily include an operating system (OS), a third-party program, or the like that causes the computer 901 to execute functions such as the execution module optimization device of the above-described embodiment. The program only needs to include an instruction portion that calls an appropriate function (module) in a controlled manner and obtains a desired result. How the computer system 9 operates is well known and will not be described in detail.

  The present invention is not limited to the above-described embodiments, and various modifications are possible, and it goes without saying that these are also included in the scope of the present invention.

  As described above, the execution module optimization apparatus according to the present invention can optimize an execution module without changing the source code. It is useful as an execution module optimizing device.

1, 2 Execution module optimization device 101, 201 Execution module storage unit 102, 202 Optimization information storage unit 103 Profiler storage unit 104, 203 Reception unit 105, 204 Disassembly unit 106, 205 Optimization location detection unit 107, 206 Change Unit 108 Execution unit 207 Compiling unit 208 Storage unit 1061 Optimization location candidate detection unit 1062 Static analysis unit 1063 Profile program generation unit 1064 Dynamic analysis unit 1065 Judgment unit 1066 Determination unit

Claims (14)

An execution module storage that can store one or more execution modules;
Optimization location detection information that is information for detecting an optimization location that is a location where processing can be optimized, and information for changing the optimization location detected using the optimization location detection information An optimization information storage unit capable of storing one or more pieces of optimization information having certain change information;
A receiving unit that receives an instruction having an execution module identifier for identifying an execution module;
A disassembly unit that reads an execution module identified by the execution module identifier included in the instruction from the execution module storage unit, disassembles the assembly module corresponding to the execution module, and
An optimization location detection unit that detects one or more optimization locations from the assembly program using one or more optimization location detection information of the optimization information storage unit;
For each of the one or more optimization locations, change information corresponding to the optimization location detection information used when detecting each optimization location is acquired from the optimization information storage unit, and the change information is used. Changing the one or more optimization locations and obtaining an optimized assembly program;
An execution module optimization apparatus comprising: an execution unit that executes the optimized assembly program. The optimization information is
The optimization location detection information, parameter information indicating parameters to be collected, and the change information,
The optimization location detection unit
Using the optimized location detection information, optimized location candidate detection means for detecting one or more optimized location candidates that are loops from the assembly program;
Information that matches a predetermined pattern is acquired from one or more optimization location candidates detected by the optimization location candidate detection means, and it is determined whether or not the information matches a predetermined condition. A static analysis means for performing a static analysis to determine a location corresponding to the information as an optimization location candidate when they match,
For each one or more optimization location candidates determined by the static analysis means, an assembly program instruction sequence for collecting parameters indicated by the parameter information is inserted before the optimization location candidates, and a profile program is Profile program generating means for generating;
Dynamic analysis means for executing dynamic analysis for executing the profile program and acquiring one or more parameters;
Determination means for determining whether the one or more parameters meet a predetermined condition;
The execution module optimizing device according to claim 1, further comprising: a determining unit that determines each of the one or more optimization location candidates as an optimization location when the one or more parameters meet a predetermined condition. The profile program generation means includes
Each of the one or more optimization location candidates is copied, an assembly program instruction sequence for collecting parameters indicated by parameter information is inserted immediately before each of the one or more optimization location candidates copied, and the one or more elements 3. The execution module optimizing apparatus according to claim 2, wherein an assembly program instruction that jumps immediately before the inserted assembly program instruction sequence is inserted immediately before each optimization location candidate. A profiler storage unit that can store a profiler that executes the assembly program and obtains the load of each instruction;
The optimized location candidate detection means includes:
Using the optimization location detection information, one or more loops are detected from the assembly program, the profiler is executed on the assembly program, the load of each instruction is acquired, and an instruction whose load exceeds a threshold is detected. 4. The execution module optimizing device according to claim 2, wherein a loop having the same is detected as one or more optimization location candidates. The optimized location candidate detection means includes:
Using the optimization location detection information, one or more loops are detected from the assembly program, and a control flow graph having two or more nodes having each instruction constituting the assembly program as a node is created from the assembly program. Then, from the control flow graph, it is a loop starting from any one node, and is a controlling loop in which the node just before returning to the starting node and the starting node are in a controlling relationship. The execution module optimization apparatus according to claim 2, wherein one or more optimization location candidates are detected. The parameter information is information indicating the number of loops and whether or not the pointers overlap.
The static analysis means includes
For each of one or more optimization location candidates detected by the optimization location candidate detection means, a loop invariant value and an induction variable for accessing the array are obtained, and the obtained loop invariance value and the induction variable are obtained. Using, determining whether or not a memory store for the sequence is performed in the optimization location candidate, and if so, perform a static analysis to determine the location as an optimization location candidate,
The determination means includes
Determining whether the number of loops acquired by the dynamic analysis means is equal to or greater than a threshold and whether or not the pointers overlap is information indicating that the pointers do not overlap;
The determining means includes
The optimization part candidate determined that the information indicating whether the number of loops is equal to or greater than the threshold and whether the pointers overlap is information indicating that the pointers do not overlap is determined as the optimization part. Decided on
The changing unit is
6. The execution module optimization according to claim 2, wherein the optimization part determined as the optimization part by the determining means is subjected to multi-core parallelization for changing to an assembly program instruction sequence that can be executed in parallel by a plurality of processors. apparatus. The parameter information is information indicating the number of loops and whether or not the pointers overlap.
The static analysis means includes
A variable name is acquired for each of one or more optimization location candidates detected by the optimization location candidate detection means, and there is a dependency between variables in the optimization location candidate using the acquired variable name. If it does not exist, perform static analysis to determine the location as an optimization location candidate,
The determination means includes
Determining whether the number of loops acquired by the dynamic analysis means is equal to or greater than a threshold and whether or not the pointers overlap is information indicating that the pointers do not overlap;
The determining means includes
The optimization part candidate determined that the information indicating whether the number of loops is equal to or greater than the threshold and whether the pointers overlap is information indicating that the pointers do not overlap is determined as the optimization part. Decided on
The changing unit is
6. The method according to claim 2, wherein the optimization part determined as the optimization part by the determination unit is subjected to SIMD conversion to an assembly program instruction that can simultaneously process a plurality of data with one instruction. Execution module optimization device. The optimized location detection information is
Is the name of the function that performs the processing of 1.
The change information is
A function name that performs the process 1 and performs the process at a higher speed than the function indicated by the optimization location detection information,
The changing unit is
2. The execution according to claim 1, wherein the one or more optimized locations are changed to change information corresponding to the optimized location detection information used when detecting each optimized location, and an optimized assembly program is acquired. Module optimization device. An execution module storage that can store one or more execution modules;
Optimization location detection information that is information for detecting an optimization location that is a location where processing can be optimized, and information for changing the optimization location detected using the optimization location detection information An optimization information storage unit capable of storing one or more pieces of optimization information having certain change information;
A receiving unit that receives an instruction having an execution module identifier for identifying an execution module;
A disassembly unit that reads an execution module identified by the execution module identifier included in the instruction from the execution module storage unit, disassembles the assembly module corresponding to the execution module, and
An optimization location detection unit that detects one or more optimization locations from the assembly program using one or more optimization location detection information of the optimization information storage unit;
For each of the one or more optimization locations, change information corresponding to the optimization location detection information used when detecting each optimization location is acquired from the optimization information storage unit, and the change information is used. Changing the one or more optimization locations and obtaining an optimized assembly program;
An execution module optimizing device comprising: an accumulator that accumulates the optimized assembly program. A compiling unit that compiles the optimized assembly program and obtains an optimized execution module;
The storage unit
The execution module optimization apparatus according to claim 9, wherein the optimized execution module is also accumulated. On the recording medium,
One or more execution modules;
Optimization location detection information that is information for detecting an optimization location that is a location where processing can be optimized, and information for changing the optimization location detected using the optimization location detection information One or more pieces of optimization information having certain change information are stored,
An execution module optimization method performed using a reception unit, a disassembly unit, an optimization location detection unit, a change unit, and an execution unit,
A receiving step in which the receiving unit receives an instruction having an execution module identifier for identifying an execution module;
A disassembly step in which the disassembly unit reads an execution module identified by an execution module identifier included in the instruction from a recording medium, disassembles, and obtains an assembly program corresponding to the execution module;
An optimization location detection step in which the optimization location detection unit detects one or more optimization locations from the assembly program using one or more optimization location detection information stored in a recording medium;
The change unit acquires, from the recording medium, change information corresponding to the optimization location detection information used when detecting each optimization location for each of the one or more optimization locations, and A changing step of using the one or more optimization locations to change and obtaining an optimized assembly program;
An execution module optimization method, wherein the execution unit includes an execution step of executing the optimized assembly program. On the recording medium,
One or more execution modules;
Optimization location detection information that is information for detecting an optimization location that is a location where processing can be optimized, and information for changing the optimization location detected using the optimization location detection information One or more pieces of optimization information having certain change information are stored,
Computer
A receiving unit that receives an instruction having an execution module identifier for identifying an execution module;
A disassembly unit that reads an execution module identified by the execution module identifier included in the instruction from a recording medium, disassembles the assembly module, and obtains an assembly program corresponding to the execution module;
An optimization location detection unit that detects one or more optimization locations from the assembly program using one or more optimization location detection information stored in a recording medium;
For each of the one or more optimized locations, change information corresponding to the optimized location detection information used when detecting each optimized location is acquired from a recording medium, and the change information is used to obtain the one or more optimized locations. A change unit that changes the optimization part of, and obtains an optimized assembly program;
A program for functioning as an execution unit for executing the optimized assembly program. On the recording medium,
One or more execution modules;
Optimization location detection information that is information for detecting an optimization location that is a location where processing can be optimized, and information for changing the optimization location detected using the optimization location detection information One or more pieces of optimization information having certain change information are stored,
An execution module optimization method performed using a reception unit, a disassembly unit, an optimization location detection unit, a change unit, and a storage unit,
A receiving step in which the receiving unit receives an instruction having an execution module identifier for identifying an execution module;
A disassembly step in which the disassembly unit reads an execution module identified by an execution module identifier included in the instruction from a recording medium, disassembles, and obtains an assembly program corresponding to the execution module;
An optimization location detection step in which the optimization location detection unit detects one or more optimization locations from the assembly program using one or more optimization location detection information stored in a recording medium;
The change unit acquires, from the recording medium, change information corresponding to the optimization location detection information used when detecting each optimization location for each of the one or more optimization locations, and A changing step of using the one or more optimization locations to change and obtaining an optimized assembly program;
An execution module optimizing method, wherein the accumulating unit comprises an accumulating step for accumulating the optimized assembly program. On the recording medium,
One or more execution modules;
Optimization location detection information that is information for detecting an optimization location that is a location where processing can be optimized, and information for changing the optimization location detected using the optimization location detection information One or more pieces of optimization information having certain change information are stored,
Computer
A receiving unit that receives an instruction having an execution module identifier for identifying an execution module;
A disassembly unit that reads an execution module identified by the execution module identifier included in the instruction from a recording medium, disassembles the assembly module, and obtains an assembly program corresponding to the execution module;
An optimization location detection unit that detects one or more optimization locations from the assembly program using one or more optimization location detection information stored in a recording medium;
For each of the one or more optimized locations, change information corresponding to the optimized location detection information used when detecting each optimized location is acquired from a recording medium, and the change information is used to obtain the one or more optimized locations. A change unit that changes the optimization part of, and obtains an optimized assembly program;
A program for functioning as a storage unit for storing the optimized assembly program.

Images