JP3580394B2 - Program conversion device and program conversion method - Google Patents

Description

【００３６】
この条件に沿って、図１１（ａ）に示すソースプログラムの最適化因子Ｘを計算すると、図１１（ｂ）に示すように、最適化因子Ｘの計算結果は−３となる。従って、この場合は、最適化Ｂ→最適化Ａの順で処理を行う。そうすると、図８に示すように、効率よく最適化が行われる。
一方、図１２（ａ）に示すソースプログラムの最適化因子を計算すると、図１２（ｂ）に示すように、最適化因子Ｘの計算結果は０となり、最適化Ａ→最適化Ｂの順番で処理を行なう。この場合、図９に示すように最適化されるので、効率よく最適化が行われる。
【００３７】
このように、本実施の形態では、最適化方法毎に処理対象の命令の数を最適化因子として集計し、その結果に基づいて最適化方法及びその実行順を自動的に選択して最適化を行なうため、最適化の効率がよく、プログラムサイズの縮小、処理速度のより一層の向上や、メモリ使用量のより一層の減少が達成される。
なお、上述の実施の形態においては、最適化方法として二重定義の削除による最適化及び基本ブロックをまたがって移動する命令の削除による最適化の場合について説明したが、本発明が適用可能な最適化方法はこれに限定されるものではなく、その他の最適化方法についても適用できることは勿論である。この場合、最適化方法毎にその処理対象となる命令の数を集計して、その値を最適化因子とする。
【００３８】
【発明の効果】
以上説明したように、本発明に係るプログラム変換装置によれば、最適化選択部において各種最適化方法毎に処理対象となる命令の数を最適化因子として集計し、適用すべき最適化方法及びその実行順を選択して、選択された最適化方法及び実行順で最適化処理部において最適化処理を実行するので、効率が良い最適化を行なうことができて、プログラムサイズの減少、処理速度の向上及びメモリ使用量の減少などのソースプログラムに応じた最適化がなされる。
【００３９】
また、本発明に係るプログラム変換方法によれば、ソースプログラムから各最適化方法毎に処理対象となる命令の数を最適化因子として集計し、その最適化因子の値に応じて最適化方法及びその実行順を決定するので、常にソースプログラムに応じた効率よい最適化がなされる。これにより、プログラムサイズの減少、処理速度の向上及びメモリ使用量の減少等の効果が得られる。
【図面の簡単な説明】
【図１】本発明の実施の形態に係るプログラム変換装置を示す模式図である。
【図２】最適化方法及びその実行順の選択処理の一例を示す模式図である。
【図３】最適化方法及びその実行順の選択処理の他の例を示す模式図である。
【図４】図２に示す選択処理例のより具体的な処理を示す模式図である。
【図５】二重定義の削除による最適化を示す図である。
【図６】基本ブロックをまたがって移動する命令の削除による最適化を示す図である。
【図７】最適化Ａ→最適化Ｂの順で最適化を行ったときの最適化の例を示す図である。
【図８】最適化Ｂ→最適化Ａの順で最適化を行ったときの最適化の例を示す図である。
【図９】最適化Ａ→最適化Ｂの順で最適化を行ったときの最適化の他の例を示す図である。
【図１０】最適化Ｂ→最適化Ａの順で最適化を行ったときの最適化の他の例を示す図である。
【図１１】最適化因子の集計の例を示す図である。
【図１２】最適化因子の集計の他の例を示す図である。
【図１３】従来のプログラム変換装置を示す模式図である。
【符号の説明】
１１ 入力部
１２ 字句解析部
１３ 構文解析部
１４ 情報解析部
１４ａ データの依存関係の解析部
１４ｂ 最適化因子の情報収集部
１５ 最適化選択部
１６ 最適化処理部
１７ 出力部[0001]
TECHNICAL FIELD OF THE INVENTION
The present invention relates to a program conversion device (compiler) and a program conversion method for converting a source program (a source program) into an object program (a target program), and in particular, to speed up the finally obtained object program or to reduce the amount of memory used. The present invention relates to a program conversion apparatus and a program conversion method for performing optimization such as reduction of the program.
[0002]
[Prior art]
In recent years, microcomputers having various architectures have been developed, and these program conversion devices are required to reduce the amount of memory used, improve the execution speed, and the like. For this reason, the program conversion apparatus not only converts a source program into an object program, but also reduces redundant instructions (including functions and arithmetic expressions; the same applies hereinafter) or replaces them with instructions that use less memory. So-called optimization for performing such processing may be performed.
[0003]
FIG. 13 is a schematic diagram showing a conventional program conversion device. The lexical analysis unit 12 analyzes the token of the source program input to the input unit 11 of the program conversion device. That is, the lexical analysis unit 12 reads the source program, decomposes the lexical data into basic lexical characters such as meaningful names, reserved words, constants, and delimiters, and converts them into symbols so that the subsequent processing can be executed efficiently. Expand to a table etc. In addition, the lexical analysis unit 12 performs a process of skipping a comment line in a source program, a process of separating a start line and a continuation line, and the like. Further, in this process, macro expansion processing and the like are also performed.
[0004]
The syntax analyzer 13 performs syntax analysis of the program. That is, the syntax analysis unit 13 checks the hierarchical structure of the statements constituting the source program, and determines whether the statement is grammatically correct.
The information analysis unit 14 includes a data dependency analysis unit 14a, and analyzes the data dependency used in the program based on the syntax information and the symbol table obtained by the syntax analysis. Perform For example, when the declaration of a variable is recognized, processing such as registering information such as the name, type, and initial value of the variable in a symbol table is performed. When an executable statement is recognized, the type of executable statement is checked, and various variables, functions, procedures, and the like used in the statement are checked. As a result, data dependencies are entered in various tables such as a symbol table or stored internally in an internal form (internal form) for the input source program.
[0005]
After analyzing the information in this way, the optimization processing unit 16 performs an optimization process. Generally, the optimization processing unit 16 executes a plurality of types of optimization methods in a predetermined order.
The output unit 17 generates and outputs a code expressed in an assembly language, an absolute address machine language, a relocatable machine language, or another programming language from the optimized internal format. .
[0006]
Examples of the optimization method include the following.
(1) When an operation result is obtained in advance such as addition, subtraction, multiplication, and division of optimization constants by folding, the operation expression is replaced with the operation result.
(2) Optimization by Deletion of Common Expression An operation expression that appears repeatedly but has the same operation result is called a common expression. The common expression appearing after the second time is replaced with the result of the first common expression.
[0007]
(3) Optimization of Loop If there is an operation expression whose operation result does not change during the loop, the operation expression is moved from within the loop to immediately before the loop.
(4) Optimization by register allocation Among variables stored in a place where the access time is long, such as an external memory, for a frequently accessed variable, the variable is accessed like an internal register of a microcomputer. By placing it in a place where the time is short, data transfer time and calculation time are reduced.
[0008]
(5) The efficiency is improved by changing the operation order within a range that does not change the optimization operation result by changing the operation order of the expression. For example, although the same operation result is obtained with ab + ac and a (b + c), the former expression adds the operation result of (a × b) and the operation result of (a × c), and the number of operations is three. Become. In the latter equation, the result of calculating (b + c) is multiplied by a, and the number of calculations is two. In this case, although the calculation results are the same, the latter reduces the number of calculations and provides an efficient object program.
[0009]
(6) Optimization by deleting double definition If one variable is defined twice and its constant is not used between the first definition and the next definition, the variable is defined first. Delete the expression you are using.
(7) Optimization by deleting an instruction that moves across basic blocks A basic block is a set of statements constituting a source program, and has no branch between its beginning and end and is continuous. A unit. When the same instruction is transmitted from a plurality of basic blocks to a certain basic block, the instruction is limited to the destination basic block, and the instruction of the source basic block is deleted.
[0010]
(8) Replacement of Instruction If the same processing result is obtained but an instruction with a high operation speed exists, the instruction is replaced with the high-speed instruction.
There are various other optimization methods.
[0011]
[Problems to be solved by the invention]
In a conventional program conversion device, a plurality of optimization methods are executed in a previously determined statistically determined order in which the efficiency becomes highest. However, when the optimization method and the execution order are constant as described above, an optimal object can be generated in a specific source program, but optimization may not be sufficient in another source program.
[0012]
In the optimization depending on the architecture, by changing the optimization method and the execution order depending on the input source program, the number of patterns to be optimized increases, and there is a possibility that a good object can be generated.
An object of the present invention is to provide a program conversion apparatus and a program conversion method that can automatically select an optimization method and an execution order according to a source program and always perform efficient optimization.
[0013]
[Means for Solving the Problems]
The above-described problem is solved by selecting an information analysis unit that aggregates the number of instructions to be processed for each optimization method as an optimization factor from a source program, and selects an optimization method and an execution order based on the optimization factor. The problem is solved by a program conversion device comprising: an optimization selection unit that performs optimization and an optimization processing unit that executes an optimization method selected in the execution order selected by the optimization selection unit.
[0014]
In addition, the above-mentioned problems include a step of totalizing the number of instructions to be processed for each optimization process as an optimization factor from a source program, and selecting an optimization method and an execution order according to the optimization factor. And a step of executing the selected optimization method in the selected execution order.
In the present invention, a processing block described below may refer to a basic block or may refer to an aggregate of a plurality of basic blocks that execute a series of processes. Is the same.
[0015]
In the present invention, the number of instructions to be subjected to optimization processing is totaled for each optimization method, and this is acquired as an optimization factor for each optimization method. For example, in order to collect optimization factors for optimization by register allocation, among variables stored in a place having a long access time such as an external memory, a variable that is frequently accessed (a variable whose access frequency is equal to or more than a predetermined number) is used. Variable)). The total value is an optimization factor for optimization by register allocation.
[0016]
In addition, in order to total the optimization factors for the optimization by deleting the double definition, the number of instructions having a double definition is totaled. The total value is an optimization factor for the optimization by deleting the double definition.
Furthermore, in order to total the optimization factors related to the optimization by deleting the instructions that move across the basic blocks, the number of the same instructions that cross the basic blocks is totaled. The totalized value becomes an optimization factor due to deletion of an instruction that moves across basic blocks.
[0017]
In this way, by counting the number of instructions to be processed for various optimization methods as optimization factors, it is possible to understand the tendency of the source program and determine which optimization method is most effective. it can. Basically, the larger the optimization factor, the greater the effect of the optimization. Therefore, for example, optimization is performed efficiently by giving priority to an optimization method having a large optimization factor.
[0018]
The optimization factors may be totaled for the entire source program, an optimization method and an execution order may be selected based on the totalization result, and the selected optimization method may be executed for the entire source program. However, even more efficient optimization is possible by dividing the source program into multiple processing blocks, totalizing the optimization for each processing block, selecting the optimization method and its execution order, and executing the optimization. It is.
[0019]
BEST MODE FOR CARRYING OUT THE INVENTION
Hereinafter, embodiments of the present invention will be described with reference to the accompanying drawings.
FIG. 1 is a schematic diagram showing a program conversion device according to an embodiment of the present invention. The lexical analysis of the source program input to the input unit 11 of the program conversion device is performed by the lexical analysis unit 12 as in the related art. Further, the syntax analysis unit 13 analyzes the syntax of the program after the lexical analysis to check the hierarchical structure of the sentences constituting the program and to perform a grammatical check.
[0020]
The information analysis unit 14 includes a data dependence analysis unit 14a and an optimization factor information collection unit 14b. The data dependency analysis unit 14a analyzes the data dependency based on the syntax information and the symbol table obtained by the syntax analysis in the same manner as in the related art, and compares the data dependency with various data such as a symbol table. And store it in the form of the internal format of the input source program. Based on the analysis result of the data dependency, it is possible to extract an instruction that does not affect other parts by the optimization.
[0021]
The optimization factor information collecting unit 14b totalizes the number of instructions to be processed for each optimization method as an optimization factor using the analysis result of the data dependency.
The optimization selecting unit 15 selects an optimization method to be applied and an execution order thereof based on the values of the optimization factors totalized by the optimization factor information collecting unit 14b. Then, the optimization processing unit 16 executes the optimization methods selected by the optimization selection unit 15 in the selected execution order.
[0022]
The output unit 17 generates and outputs a code expressed in an assembly language, a machine language at an absolute address, a relocatable machine language, or another programming language from the optimized internal format.
The selection of the optimization method is performed as follows. For example, as shown in FIG. 2, five types of optimization patterns 1 to 5 in which the optimization method and its execution order are patterned are prepared. Then, after inputting the source program and totalizing the optimization factors (collecting information), one of the optimization patterns 1 to 5 is selected based on the values of the optimization factors, and the optimization processing unit is selected. In step 15, the optimization method defined in the optimization pattern is executed in a predetermined execution order, and the result is output.
[0023]
The execution of the optimization processing may be performed as follows. That is, FIG. 3 shows that one of the optimization methods 1 to 5 is appropriately selected and called based on the result of the information collection, the method is executed, and then any one of the optimization methods 1 to 5 is further executed. One is selected and called, and optimization is performed a plurality of times.
FIG. 4 illustrates an example of an optimization selection process based on the optimization process example of FIG. In this example, the optimization factors are x and y, and the values of these optimization factors x and y are shown on the rectangular coordinates, and point P indicates the state of the optimization factor of this program. . In this example, since the point P is in the first quadrant (x> 0, y> 0), this program performs optimization processing in the optimization method and processing order determined by pattern 1. When the point P is located in the second quadrant (x <0, y> 0) or the third quadrant (x <0, y <0), the optimization is performed in the optimization method and processing order determined in the pattern 2. The processing is performed, and when it is located in the fourth quadrant (x> 0, y <0), the optimization processing is performed according to the optimization method and processing order determined by pattern 3.
[0024]
The optimization factor information collecting unit 14b sets the optimization factor of a certain optimization method as a positive value and sets the optimization factor of another optimization method as a negative value, and increases or decreases from an initial value determined for each architecture. The optimization method and the execution order may be selected based on the result of the aggregation.
In addition, since these procedures may be performed for each processing unit of the program, the optimization method and the execution order can be changed for each processing block even in one source program. This allows for more efficient optimization.
[0025]
Hereinafter, the present invention will be described more specifically.
FIG. 5 is a diagram showing optimization by deleting a double definition (hereinafter, referred to as optimization A). As shown in FIG. 5A, in the source program, the variable a is double-defined as “a = b” and “a = c” in the basic block. It is assumed that the variable “a” is not used after “a = b” is defined and before “a = c” is defined. In this case, as shown in FIG. 5B, the number of processes is reduced by deleting the previous instruction “a = b”.
[0026]
FIG. 6 is a diagram illustrating optimization by deleting an instruction that moves across basic blocks (hereinafter referred to as optimization B). As shown in FIG. 6A, when there is a common instruction (“a = b”) in the two basic blocks BK1 and BK2 whose control is transferred to the basic block BK3, as shown in FIG. Then, the number of processes is reduced by moving the common instruction to the block BK3.
[0027]
7 and 8 show examples of processing results when the execution order of the optimization method is changed, and show how the optimization is performed according to the execution order of the optimization A and the optimization B and the result is different. Is shown.
FIG. 7A shows that among the basic blocks BK1, BK2, and BK3 that constitute the source program, the block BK1 and BK2 have the instruction “a = b”, and the block BK1 and BK2 have the instruction “a” assigned to the block BK3 to which the processing shifts. = C "is a program. It is assumed that optimization B is executed after execution of optimization A for this program. In this case, even if the optimization A is executed first, the program is not changed, as shown in FIG. 7B, since the same block is not defined twice. Thereafter, when the optimization B is executed, the common instruction “a = b” in the blocks BK1 and BK2 moves to the block BK3 as shown in FIG. 7C. When the optimization is performed in the order of optimization A → optimization B in this manner, the block BK3 after the optimization processing has a double definition for the variable a, and it cannot be said that the optimization is sufficient. .
[0028]
On the other hand, assume that optimization A is executed after optimization B is executed for the same source program. Then, as shown in FIG. 8A, by executing the optimization B, the instruction “a = b” common to the blocks BK1 and BK2 moves to the block BK3. Next, when the optimization A is executed, the earlier instruction of the instructions “a = b” and “a = c” which are double-defined in the block BK3 is deleted, and as shown in FIG. In the block BK3, the instruction regarding the variable a is only "a = c".
[0029]
That is, the source program of FIG. 7A can be optimized efficiently by performing optimization in the order of optimization B → optimization A.
However, if the optimization is performed in this order, the optimization may not be sufficient. 9 and 10 show examples of processing results when the execution order of the optimization method is changed for another source program. FIG. 9A shows a source program composed of basic blocks BK1, BK2 and BK3. In the block BK1, an instruction "a = d" is followed by an instruction "a = d", and in the block BK2, an instruction "a = c". Is followed by the instruction "a = d". Then, the processing moves from the blocks BK1 and BK2 to the block BK3.
[0030]
When optimization A is first performed on this source program, as shown in FIG. 9B, in the blocks BK1 and BK2, the earlier instruction of the two instructions that are defined twice for a is deleted. In each of the blocks BK1 and BK2, the instruction for “a” is only “a = d”. Thereafter, when the optimization B is performed, the instruction “a = d” common to the blocks BK1 and BK2 moves to the block BK3 as shown in FIG. 9C.
[0031]
On the other hand, when the optimization B is first performed on the source program of FIG. 9A, the instruction “a = d” common to the blocks BK1 and BK2 moves to the block BK3 as shown in FIG. . Thereafter, even if the optimization A is executed, as shown in FIG. 10B, the program does not change because neither of the blocks BK1 and BK2 has a double definition of a.
[0032]
Thus, for the source program shown in FIG. 9A, optimization is performed in the order of optimization A → optimization B, rather than in the order of optimization B → optimization A. Thereby, optimization can be performed efficiently.
Hereinafter, optimization factors for determining the execution order of the optimization method will be described. FIG. 11A shows the same source program as FIG. 7A, and FIG. 11B shows a method of calculating an optimization factor in the source program. Usually, it is necessary to prepare an optimization factor for each of a number of optimization methods to be applied, but here, for simplicity of explanation, the optimization method is based on the elimination of the double definition shown in FIG. It is assumed that there are only optimization A and optimization B by deleting an instruction that moves across basic blocks shown in FIG. 6, and that the optimization factor is only X (+ X and -X). The initial value of the optimization factor X is set to 0.
[0033]
The optimization factor X is calculated (aggregated) under the following conditions for the optimizations A and B.
1. If the definition of the same variable continues in the same block, the number of processes is reduced by one due to the optimization A, so that “X = X + 1” is set.
2. If there is an instruction that moves over the basic block, the number of instructions is reduced by the number of instructions due to the optimization B, so that “X = X−the number of instructions” is set.
[0034]
Then, according to the value of X, the execution order of the optimization method is selected as shown in Table 1 below.
[0035]
[Table 1]

[0036]
When the optimization factor X of the source program shown in FIG. 11A is calculated according to this condition, the calculation result of the optimization factor X becomes -3 as shown in FIG. 11B. Therefore, in this case, processing is performed in the order of optimization B → optimization A. Then, as shown in FIG. 8, optimization is efficiently performed.
On the other hand, when the optimization factor of the source program shown in FIG. 12A is calculated, as shown in FIG. 12B, the calculation result of the optimization factor X becomes 0, and in the order of optimization A → optimization B, Perform processing. In this case, the optimization is performed as shown in FIG. 9, so that the optimization is performed efficiently.
[0037]
As described above, in the present embodiment, the number of instructions to be processed is totaled as an optimization factor for each optimization method, and the optimization method and the execution order are automatically selected based on the result to perform optimization. Therefore, the efficiency of the optimization is high, and the program size can be reduced, the processing speed can be further improved, and the memory usage can be further reduced.
In the above-described embodiment, the optimization method by elimination of duplicate definitions and the optimization by elimination of instructions that move across basic blocks have been described. The optimization method is not limited to this, and it goes without saying that other optimization methods can be applied. In this case, the number of instructions to be processed is totaled for each optimization method, and the value is used as an optimization factor.
[0038]
【The invention's effect】
As described above, according to the program conversion device of the present invention, the optimization selecting unit counts the number of instructions to be processed for each optimization method as an optimization factor, and optimizes the optimization method to be applied. Since the execution order is selected and the optimization processing unit executes the optimization processing in the selected optimization method and the execution order, efficient optimization can be performed, and the program size is reduced and the processing speed is reduced. Optimization according to the source program, such as improvement of memory and reduction of memory usage, is performed.
[0039]
Further, according to the program conversion method of the present invention, the number of instructions to be processed for each optimization method is counted as an optimization factor from a source program, and the optimization method and the optimization method are performed in accordance with the value of the optimization factor. Since the execution order is determined, efficient optimization according to the source program is always performed. As a result, effects such as a reduction in program size, an improvement in processing speed, and a reduction in memory usage can be obtained.
[Brief description of the drawings]
FIG. 1 is a schematic diagram showing a program conversion device according to an embodiment of the present invention.
FIG. 2 is a schematic diagram illustrating an example of an optimization method and an execution order selection process thereof.
FIG. 3 is a schematic diagram showing another example of an optimization method and a process of selecting an execution order thereof.
FIG. 4 is a schematic diagram showing more specific processing of the selection processing example shown in FIG. 2;
FIG. 5 is a diagram showing optimization by deleting a double definition.
FIG. 6 is a diagram illustrating optimization by deleting an instruction that moves across basic blocks.
FIG. 7 is a diagram illustrating an example of optimization when optimization is performed in the order of optimization A → optimization B;
FIG. 8 is a diagram illustrating an example of optimization when optimization is performed in the order of optimization B → optimization A.
FIG. 9 is a diagram illustrating another example of optimization when optimization is performed in the order of optimization A → optimization B.
FIG. 10 is a diagram illustrating another example of optimization when optimization is performed in the order of optimization B → optimization A.
FIG. 11 is a diagram illustrating an example of aggregation of optimization factors.
FIG. 12 is a diagram illustrating another example of the aggregation of optimization factors.
FIG. 13 is a schematic diagram showing a conventional program conversion device.
[Explanation of symbols]
Reference Signs List 11 Input unit 12 Lexical analysis unit 13 Syntax analysis unit 14 Information analysis unit 14a Data dependency analysis unit 14b Optimization factor information collection unit 15 Optimization selection unit 16 Optimization processing unit 17 Output unit

Claims (4)

An information analysis unit that aggregates the number of instructions to be optimized as an optimization factor for each of a plurality of types of optimization methods prepared from a source program;
An optimization selection unit that selects an optimization method to be adopted based on the optimization factor and an execution order thereof,
An optimization processing unit that executes an optimization method selected in an execution order selected by the optimization selection unit. The information analysis unit performs the totalization of the optimization factors for each processing block having no branch between the input and the output, and the optimization selection unit performs the processing according to the totalization result of the optimization factors of the information analysis unit. The optimization method and the execution order are selected for each block, and the optimization processing unit executes the optimization method for each processing block according to a selection result of the optimization selection unit. Item 2. The program conversion device according to Item 1. From the source program, for each of a plurality of types of optimization methods, a step of counting the number of instructions to be optimized as an optimization factor;
Selecting an optimization method to be adopted according to the optimization factor and an execution order thereof,
Executing the selected optimization method in the selected execution order. 4. The program conversion method according to claim 3, wherein the aggregation of the optimization factors and the selection and execution of the optimization method are performed for each processing block having no branch between the input and the output.

Images