JP3246043B2 - Compiler unit - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an apparatus for translating a source program of a computer to generate a target program, and in particular, to automate the selection of specified items such as optional items (options) relating to optimization of the target program. Compiler device that can be used.

[0002]

2. Description of the Related Art As is well known, when a computer program is described in a so-called high-level programming language, a compiler for the high-level programming language is used as a source program. The program is translated into a target program for a specific computer, and if necessary, is processed into a single program that can be directly executed by a computer through a combined editing process of linking a plurality of target programs.

Here, as is well known, in translation processing, optimization techniques are applied in order to improve the execution efficiency of a target program (required memory (main memory) amount during execution, execution time, etc.). With the advancement of compiler technology, a number of optimization items are provided, and can be selected as so-called compile options.

Examples of such optimization include, for example, in an object program of a system having a subroutine-like function as a function and using a stack, "when an operation returns from a function, an argument is immediately popped" or "arithmetic". When performing an operation, copy the operand on the memory to a register in advance to perform inter-register operation "," When performing an arithmetic operation, copy the address of the operand on the memory in advance to the register ", There are "saving a frame pointer in a register", "expanding a function in a program", and others.

For these optimizations, for example, considering a simple example of “optimizing a function in a program”, if the function is expanded, the overhead of function call at the time of execution can be eliminated. In general, there is a trade-off in execution efficiency, such as high efficiency, but of course a large program.

Therefore, for each optimization item, it is necessary to specify whether or not to perform optimization in consideration of such a trade-off and the situation of a program to be created.
Conventionally, there is no other choice but to specify each optimization item appropriately based on the idea recognition of such a situation. Even more, it is not possible to consider the mutual influence of various optimizations, etc. Has not been used effectively.

[0007] In addition to the compile options described above, there are a plurality of compilers having different characteristics depending on the programming language, and there are cases where the compiler can be selected as a matter that can be selected in generating the target program. When linking, the order in which modules are connected may affect execution efficiency.

[0008] Since these are also appropriately selected based on the idea recognition, it cannot be said that an appropriate purpose program is obtained by effectively utilizing those functions.
SUMMARY OF THE INVENTION It is an object of the present invention to provide a compiler apparatus for automatically selecting and compiling designated items that can be selected in generating a target program so as to obtain an appropriate target program.

[0009]

FIG. 1 is a block diagram showing the configuration of the present invention. The figure shows the configuration of a compiler device, which translates a source program described in a predetermined programming language into a target program, controls the translation process, and has a plurality of designated items that may affect the result of the translation. Is a device capable of selecting a group generation unit 1
, A decoding unit 2, a translation unit 3, an evaluation unit 4, and a genetic operation unit 5.

In the first invention, the population generating unit 1 associates each locus of a chromosome of the genetic algorithm with each of the predetermined specified items among the specified items, and specifies the specified contents of each of the specified items. Is encoded as a chromosome having the locus corresponding to all of the predetermined designated items, and a predetermined number of codes of the chromosome are generated as a group of the genetic algorithm.

The decoding unit 2 sequentially decodes the codes of the chromosomes of the population into the designated items for the population generated by the population generation unit 1 and the population as a result of the operation performed by the genetic operation unit 5. Hand over to translator 3.

The translating unit 3 translates a given source program in accordance with the designated item for each chromosome passed from the decoding unit 2 and generates each target program corresponding to each chromosome. Execute.

The evaluator 4 executes, for each of the chromosomes, the normality of the translation processing result of the translator 3 and, when the target program is generated, executes the target program under predetermined conditions. A predetermined evaluation value is generated based on the size and execution status of the program.

The genetic operation unit 5 performs, based on the evaluation value given by the evaluation unit 4 corresponding to each chromosome of the population, to the population that has been decoded by the decoding unit 2, The group is updated by executing a predetermined operation of the algorithm, and the group resulting from the update is passed to the decoding unit 2
Is repeated a predetermined number of times,
When detected, execution of the predetermined operation of the genetic algorithm
End line .

In the second invention, in the case where the source program is divided into a plurality of modules, the group generation unit 1 provides a locus corresponding to each of the designated items individually for each of the modules. The code of each chromosome is generated, and the decoding unit 2 sequentially decodes the code of each chromosome of the group generated by the group generation unit to the designated item,
The data is divided into modules and passed to the translation unit 3.

The translation unit 3 executes the translation process by individually applying the designated items for each module for each chromosome passed from the decoding unit 2 to each module. Third
In the present invention, the predetermined specification items in the first and second inventions include at least a plurality of specification items regarding whether or not the optimization processing of the target program can be applied.

According to a fourth aspect, in the first to third aspects, the translation unit 3 has a plurality of compilers for performing different translation processes, and the predetermined specification item includes at least one of the compilers. To include one of the specified items.

In a fifth aspect based on the first to fourth aspects, the source program is divided into a plurality of modules, and at least one of the predetermined designation items is generated from each of the modules. The evaluation unit 4 includes a designation item for designating the order of the module when linking a program, and the evaluation unit 4 links and executes the target program according to the link order as a result of decoding by the decoding unit 2.

[0019]

With the compiler device of the present invention, each selectable specification item such as a compile option is associated with a chromosome locus, and for example, a value of 0 or 1 represents whether or not the specification item is specified. Code as follows.

First, the group generation unit 1 appropriately generates a certain number of chromosome codes, and passes it to the decoding unit 2 as an initial group. The decoding unit 2 decodes the code for each chromosome into a set of designated items such as a compile option, and passes it to the translating unit 3.

The translation unit 3 executes a translation process for a given source program corresponding to each chromosome in accordance with the set of designated items. Generate a program. In this case, when the specification item includes the specification of the compiler, the compiler that executes the above translation processing is selected for each chromosome.

The evaluation unit 4 performs execution evaluation on the generated target program, assigns an evaluation score according to a predetermined place, and evaluates each evaluation value for all chromosomes of the population including the case where the target program cannot be generated. give.

The genetic operation unit 5 performs an operation of a genetic algorithm that reflects an evaluation value for a group,
The chromosomes constituting the group are updated, the resulting group is passed to the decoding unit 2, and the above processing is repeated by decoding by the decoding unit 2.

The operation of a genetic algorithm is well known (for example, see David E. Goldberg, "Genetic Algorithms in
Search, Optimization, and Machine Learning "ADDISON-
WESLEY PUBLISHING COMPANY, INC. 1989), group (populat
ion), based on the evaluation value assigned to the chromosome of the population, reproduction, crossover, and sudden transition
(mutation).

Here, the regeneration is an operation of discarding a chromosome with a low evaluation value among the chromosomes of the population and forming a population of copies of a good chromosome. The operation of exchanging the value of each digit) is an operation of changing the gene stochastically.

By repeating the above processing, genetic progress / selection is simulated by the genetic algorithm operation, and it is expected that the code of the chromosome of the population is updated so as to obtain a better evaluation value sequentially. So you can
By repeating the appropriate number of times, an appropriate option is automatically selected.

[0027]

FIG. 2 (a) shows an example in which the compile options as described above are encoded on a chromosome for implementing the present invention. FIG. 2 (a) shows the code. Sequentially indicates the following.

That is, the first digit is the first locus,
The function associated with the compile option specified by the flag "-fno-defer-pop" and the meaning of the flag when it is valid is "pop the argument immediately after returning from the function. Generate a program) ”and the value“ 1 ”is valid (yes)
Is invalid (no) by “0”.

Therefore, if the code is, for example, "11001", pop the argument, copy the operand to the register, and expand the function, but do not copy the memory address constant and save the frame pointer. Is specified.

In this case, as a result of decoding by the decoding unit 2, "-fno-defer-p
op-fforce-mem-inline-function "is passed to the translation unit 3.

FIG. 2 (b) shows an example in which a chromosome is encoded corresponding to a plurality of modules constituting a source program as in the second invention. Two modules with compile options as specified items
Corresponds to sga.c and io.c.

In this example, for example, the code is "1100101100"
If so, the translation unit 3 has "-fno-dfer-pop -fforce-mem -inline-functions" for the first module.
Is passed, and for the second module, "-fforce-mem
-fforce-addr "will be passed.

FIG. 3 (c) shows an example in which a system having a plurality of compilers and a compiler is automatically selected as in the fourth invention. In this example, two types of compilers gcc and pcc are used. 2 shows a case where the compiler specification is coded together with the same compile options as in the example of FIG.

In this example, for example, if the code is "110010", the translation unit 3 specifies "-fno-dfer-pop -fforce-mem -inline-functions" as a compile option.
Is passed, and the compiler name "pcc" is specified as the compiler specification
Is specified.

FIG. 3D shows an example in which the link order of a plurality of modules of the target program is automatically specified as in the fifth invention. In this example, two modules sga.c are used. 2 and io.c, where the specification of the link order is coded together with the same compile options as in the example of FIG.

In this code, if the sixth digit is "1", the decoding unit 2 sets the first to sga.c and the second to io.c.
From the module 3 to the evaluator 4 via the translator 3, a list of module names “sga.c io.c” is passed as designation information of the link order, and if “0”, the io.c is set first. `` Io.c
sga.c "is specified.

FIG. 4 is a flowchart of a process for explaining an example of the process of the present invention in detail. First, the group generation unit 1 of FIG. After the setting, the initial value of the group is generated in the processing step 11 as follows, for example.

That is, as an example, as described with reference to the chromosome code of FIG. 2 (a), when the number of individuals in a group is, for example, four, for example, by an appropriate random number generating mechanism,
By taking out four random numbers and using the lower five digits of each random number as a code of each chromosome as a code of each chromosome, a code group consisting of the codes of the four chromosomes as illustrated in FIG. A group based on the genetic algorithm.

Upon receiving this group, the decoding unit 2 decodes each chromosome in the processing step 12 as described above and passes it to the translating unit 3, so that the compiler function of the translating unit 3 is normally executed in the processing step 13. Compiles the source program, and if an error occurs in the generated target program or compile processing, the result is evaluated by the evaluation unit 4.
Pass to.

In the processing step 14, the evaluation unit 4 executes the target program under predetermined conditions specified in advance, and determines the necessary memory amount, execution time, and correctness of the execution result (for example, when the output data is (E.g., whether or not the data matches the data of the execution result) is collected.

When the necessary evaluation data has been collected, the evaluation unit 4 calculates an evaluation value from the data in a processing step 15 and holds it as an evaluation value F for the chromosome. The evaluation value F is, for example, a value obtained by the following equation.

F = execution time + α × memory amount + β + γ where α is a constant for giving an evaluation weight between the execution time and the memory amount, β is 0 when the compilation ends without error, and If there is, a variable that takes ∞ (infinity), γ is a variable that takes 0 if the execution result of the program is correct, and γ if it is not correct.

When the evaluation value is obtained, it is determined whether or not all chromosomes of the population have been processed in processing step 16, and if there is no processing, the process returns to processing step 12, and evaluation is performed for all chromosomes as described above. After obtaining the values, the evaluation unit 4 passes the population and the evaluation value to the genetic operation unit 5.

The genetic operation unit 5 discriminates whether or not the counter K of the number of repetitions has reached a predetermined value in processing step 17 and, if it has reached the result, in processing step 20 (for example, the data of the group and evaluation value at that time) ) Is output and the processing is terminated.

If K has reached the predetermined number of times or if it has not been identified, the genetic operation unit 5 increments K by 1 in processing step 18 and then processes the group and the evaluation value received from the evaluation unit 4 in the processing step. At 19, the group is processed according to the genetic algorithm.

For example, as a result of the group generation unit generating the group illustrated in FIG. 5A and evaluating this group by the above processing, the genetic operation unit 5 transmits the data of FIG. An example of a genetic manipulation as received will now be described.

The genetic operation unit 5 first performs a reproducing operation,
The data with the worst evaluation value is killed and the best data is duplicated. As a result, the second data in the group (b) is killed and the group in (c) where the first data is duplicated in the second. Get. By this operation, the average evaluation value of the group is naturally improved as shown in the figure.

Next, the genetic operation unit 5 performs
With two data randomly selected for the reproduced group, one of the bit strings that are divided into right and left at the boundary is exchanged between the two data with the position determined at random as the boundary.

That is, for the group (c), if the second and third data are selected and randomly determined so as to place a boundary between the third and fourth bits, the right of the second data By exchanging "11" of " 3 " with "00" on the right of the third data, the group shown in (d) is obtained .

Next, as a mutation operation, the genetic operation unit 5 inverts one bit selected at random from the data of the group (d), and for example, the bits marked with * in (d) are randomly selected. Then, the mutations result in the population shown in (e).

As described above, the genetic operation unit 5
When a new group as shown in (e) is obtained, the new group is decoded.
And the process for evaluating this group is started again as described above.

[0052]

As is apparent from the above description, according to the present invention, when compiling a source program of a computer, appropriate selections are automatically made for compile options for optimization and other selection items. There is a significant industrial effect that it is possible to activate the option.

[Brief description of the drawings]

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

FIG. 2 illustrates a chromosome code example (part 1)

FIG. 3 illustrates a chromosome code example (part 2)

FIG. 4 is a flowchart of the process of the present invention.

FIG. 5 is a diagram illustrating a processing example of the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Group generation part 2 Decoding part 3 Translation part 4 Evaluation part 5 Genetic operation part 10-20 Processing steps

Continuation of front page (56) References Beaty, "Genetic Algorithms and Instru- ment Scheduling", Proceedings of 24th annual international symposium on Microarchitecture, 1991. 206-211 S.C. Beaty et al., "Motion and framework for Usage Genetic Algorithms for Microcode Compaction", Proceedings of the nation's general insurance program. 117-124 Invention Technical Publication No. 93-8695 (58) Fields investigated (Int.Cl. ⁷ , DB name) G06F 9/45 G06N 3/00

Claims (5)

(57) [Claims] 1. A source program written in a predetermined programming language is translated into a target program, the translation process is controlled, and a plurality of designated items that may affect the result of the translation can be selected. An apparatus, comprising: a population generation unit, a decoding unit, a translation unit, an evaluation unit, and a genetic operation unit, wherein the population generation unit specifies each locus of a chromosome of a genetic algorithm by the designation. pre-defined specification each said of the items
Corresponding to the item , giving a value representing the specified content of each of the specified items, encoding as a chromosome having the locus corresponding to all the predetermined specified items ,
The code of the chromosome of a predetermined number to generate a population of the genetic algorithm, the decoding unit, the collection of the population generating unit has generated
Dan, and said with <br/> said population of results genetic manipulation portion is manipulated, sequentially the finger code of each of the chromosomes of the population
Passing the said translation unit decodes the constant fields, said translation unit in accordance with the designated entry for each of the chromosomes <br/> passed from the decode unit, given the primitive program
Each said the purpose of translating the beam, corresponding to each of the chromosome
Performs a process of generating a program, the evaluation unit, for each said chromosome, and health of the translation processing result of the translation section <br/>, the object program when the object program is generated by running under a predetermined condition, on the basis of the size and state of execution of the object program, it generates a predetermined evaluation value, the genetic manipulation unit, to the population of the decoding unit is currently decoding , based on the evaluation value assigned by the evaluation unit in response to each of said chromosome of the population, the running predetermined operation of the genetic algorithm
Updating the group, and updating the group of the update result to the decoding unit
Repeat this process a specified number of times, and
Is detected, the predetermined operation of the genetic algorithm is performed.
A compiler device configured to terminate execution of a work . In the case where wherein said source program is divided into a plurality of modules, said population generation unit, a locus to correspond to each of said designated item, individually provided for each of the module, the code of each chromosome The decoding unit, each of the group generated by the group generation unit
Decoding sequentially the designated item code of the chromosome, by dividing by the module passes the translation unit, said translation unit, by the module of each of the chromosomes <br/> passed from the decoding unit the specified items, each of the character of
Executing the translation processing is applied separately to Yuru, wherein
Item 7. The compiler device according to item 1. 3. The compiler apparatus according to claim 1, wherein the predetermined specification items include at least a plurality of specification items regarding whether or not the optimization processing of the target program can be applied. 4. The translation unit has a plurality of compilers that perform different translation processes, and the predetermined specification item includes at least a specification item for selecting one of the compilers. 4. The compiler device according to claim 1, 2 or 3. 5. The program according to claim 1, wherein the source program is divided into a plurality of modules, and the predetermined specification item specifies at least the order of the modules when linking a target program generated from each of the modules. It is executed by linking the object program according to the specified item, according to claim 1, claim 2, claim 3 or claim 4
Compiler device as described.