JP2539070B2 - Compiler processor - Google Patents

Description

The present invention relates to optimization processing of a compiler of a programming language having a pointer function in a computer, and a pointer optimization processing method that efficiently processes detection of a pointer optimization target and that has good transfer efficiency. For the purpose, two words that refer to the same content in the same area through the pointer variable from a plurality of words that form a program expressed in a predetermined language having a function of pointing data by the pointer variable, When performing a predetermined optimization process, two statements that refer to a variable having the same variable name in the structure pointed to by the same pointer variable are detected, and another statement in an intermediate execution order of the two statements is detected. In the case where it is detected that there is a process of defining a variable of a structure pointed to by the pointer variable, an area of the variable defined by the other statement,
The overlap with the variable area referred to in the two statements is identified by comparing the relative position and the area length in the structure to which both variables belong to, and the overlap of the two areas is identified. If not, the optimization is performed on the two votes.

[Industrial applications]

The present invention relates to a processing method for optimizing pointers in a computer programming language compiler having a pointer function.

[Conventional technology]

In the case of a computer programming language, such as the well-known C language, which has a pointer function for pointing data, a compiler that processes a source program written in that language translates the source program. So-called optimization processing is performed to improve the performance of the generated target program. As the optimization of the pointer, which is a part of such optimization processing, there is an optimization that shortens the execution time of the program by converting a data reference via the pointer into a form of directly referring to the data.

FIG. 2 shows an example of the configuration of a computer that performs the above-mentioned processing. For example, a C language compiler 1 executed by the processing device.
Inputs a source program 3 written in C language, analyzes and translates the source program 3 to generate a target program 4 to be executed by a required processing device.

Therefore, in the compiler 1, the control unit 5 performs initial settings for translation processing and sequentially reads the program sentences of the source program 3, and the analysis processing unit 6 analyzes the lexical phrase, the syntax, and the expression of the program sentence, The intermediate language program 7 is generated by converting it into a corresponding language of the intermediate language.

The intermediate language program 7 is a program expressed in an intermediate language defined so that the subsequent translation processing can be efficiently executed, and is composed of a plurality of words, and each word corresponds to various program statements and the like. It is configured to represent various commands and the like in a predetermined format.

Next, the optimization processing unit 8 optimizes the intermediate language program 7. As is well known in the optimization, optimizations such as putting out an invariant expression outside the loop or changing multiplication in the loop to addition for a loop program specified by a DO statement, and detecting common expressions Then, the optimization processing is performed by removing the common expression such as converting into a simple reference.

With respect to the intermediate language program thus optimized, the object generator 9 allocates a storage area and performs conversion processing, and outputs the target program 4.

By the removal of the common expression of the optimization processing unit 8 and the like,
Although the pointer optimization as described above is also performed, a dynamic analysis method and a static analysis method are known as processing methods for determining a target to which the pointer optimization can be applied. The processing of both analysis methods will be described by taking the case of processing the program illustrated in FIG. 5 as an example.

In FIG. 5 (a), referring to the line number displayed for explanation on the left end, the "struct strl" of the line is defined by the definition of the data structure in braces following it. Indicates that the structure type name is strl, and the structure consists of the elements shown in braces on the ~ line. That is, the element is float type data with name f1 and char type with name c
It consists of 1 data and int type data with the name i1. V1 outside the last curly brace is the name of the structure data.

The ~ line also defines another structure in the same manner as described above. It should be noted that the line in the definition of the element is an array of char type and the name is c1 and indicates that the number of array elements is 5.

The line is "* p", which means "the data pointed to by the pointer variable p", and that it is the data of the structure type str1, that is, the pointer variable p is a pointer that points to the data of the str1 type structure. Declare that there is, and similarly declare the row for the pointer variable q.

,,, lines are all assignment statements, and for example, the line defines the variable p as "address value of variable v1" represented by "&v1". Further, the display in the form of “p → f1” such as a line represents “the element f1 of the structure pointed to by the pointer variable p” and the like.

The same applies to FIG. 5 (b), but the line "void * p"
Indicates that the type of data pointed to by the pointer variable p is not limited. In addition, "(struct str1 *)" that modifies "p" between lines indicates that the data pointed to by p in this case is the structure type str1.

For example, in the case of processing by the dynamic analysis method in order to identify the row that refers to the common expression in FIG. 5 (a) or (b) and whether the row can be optimized, the execution of the vote is simulated. The common expression is determined by recognizing that the contents referred to by the common expression are not changed from line to line in the same manner as in FIG. 5 (a) and (b). In the case of, both votes are recognized as having a common expression.

On the other hand, in the case of the static analysis method, the data that each pointer variable may point to is checked for the entire program to be processed, and another pointer variable that has a duplication in the data that may point to If the variable indicated by is defined in the middle of both words, it is not a common expression. Therefore, in the case of (a), it is specified that the pointer variables p and q point to different types of data in the line, so the definition in the line does not matter,
And lines are recognized as having a common expression.

However, in the example of FIG. 5B, the type of data pointed to by the pointer variable p is not limited in the static relationship of the entire program, and p and q have the same address value of the variable v2 depending on the line. Since the variable pointed to by q is defined in the middle row between and, it is possible that the variable f1 referenced in and may be different. Does not recognize.

As described above, when the common expression is recognized, the fifth
In Figure (a) or (b), for example,
It can be converted to a simple reference of the form "s = r" for optimization.

[Problems to be Solved by the Invention]

As can be inferred from the above explanation of the dynamic analysis method and the static analysis method, the dynamic analysis method can recognize the optimization target with relatively no omission, but the processing that follows the execution for each word Requires a relatively long processing time, and since a bit vector for displaying the possibility of pointing with a pointer is provided for each processing unit, there is a problem that the work area also becomes large. On the other hand, static analysis In the method, the processing time is relatively short and the work area is small, but there is a problem that there are relatively many omissions of recognition that optimization is possible.

It is an object of the present invention to detect a pointer optimization target, have a high processing efficiency, and have a proper detection efficiency.

[Means for solving the problem]

 FIG. 1 is a flow chart of processing showing the configuration of the present invention.

The figure shows the structure of a pointer optimization processing method, in which the same contents of the same area are selected from a plurality of words constituting a program expressed in a predetermined language having a function of instructing data by a pointer variable. When performing a predetermined optimization process with respect to the two words referred to via, the two words referring to the variable having the same variable name in the structure pointed to by the same pointer variable are detected in processing step 10,
When it is detected in the processing step 11 that there is processing for defining the variable of the structure pointed to by the pointer variable in another statement in the middle execution order of the two statements, the other step is executed in the processing step 12. Comparing the area of the variable defined in the above statement with the area of the variable referenced in the two statements, the relative position and the area length in the structure to which the two variables belong, and comparing the two variables. If the two areas do not overlap, the optimization is performed on the two words in processing step 13.

[Action]

With this processing method, regarding the relationship between the common expression candidates and other definition words, instead of checking the variables that the pointer may point to, the presence or absence of overlap of the area pointed to by the pointer is checked. A comparatively short processing time, because it is determined by comparing the relative position and length of the variable region in the designated structure to determine whether or not the optimization process is applicable.
And small working areas can be used to detect optimizable targets relatively leak-free.

〔Example〕

FIG. 1 is a part of the processing by the optimization processing unit 8 in FIG. 2, for example, in order to recognize the common expression of the optimization target candidates, the processing is an alternative to the above dynamic or static analysis method, When the intermediate language program 7 is scanned and two words that refer to the variable having the same variable name of the structure pointed to by the same pointer variable are detected in the processing step 10, the intermediate execution of the two words is executed in the processing step 11. If there is a process that may define the data pointed to by the pointer variable in other words in order, and if there is no corresponding process, both parties refer to a common expression and processing step 13 Optimize.

If an intermediate vocabulary that may be applicable is detected, the overlap of the variable area defined in the other vocabulary and the variable area referred to in the subsequent vocabulary in step 12 is performed. The relative position and the region length in the structure to which the variable belongs are identified by comparing the two variables, and only when there is no overlap, the above-mentioned optimization is performed on these two statements in processing step 13. If it is determined in processing step 12 that there is an overlap, it is not the target of the optimization scheduled here.

FIG. 3 is a flow chart of the processing showing a detailed processing example of the processing step 12 of FIG. 1. First, in processing step 21, a statement defining a variable in the middle of both reference words of optimization candidates, It is determined whether the data pointed to by the pointer is the element of the structure with respect to the speech referred to thereafter, and if it is not the element of the structure, it is identified in processing step 22 whether the data type is the same, If they are different, they are not overlapped. It should be noted that if at least one of the types of data has a structure of 1 byte, it is always determined as "overlap".

If the data pointed to by the pointer is an element of the structure, the process branches to processing step 23, and if the pointer variables are the same, it is determined in processing step 24 whether the pointer variable points to a structure of the same type. (The type of the structure to be pointed to can be identified by a declaration such as the line in the example of FIG. 5B, or a modification in the parentheses on the right side of the line).

If the structures are different here, the relative positions of the different structures will not be determined at this stage of processing, so
The data pointed to by the pointer is determined to "overlap". However, if the same structure is used, the relative position where the elements pointed by both pointers are placed can be obtained from the declaration of the structure, so in processing step 25 the positions of both elements are compared as follows. The presence or absence of overlap is determined by.

That is, regarding the structure pointed to by the pointer variable,
The displacement of the referenced variable in the structure and the length of the variable's region are compared to those of the defined variable. As a result, if the displacement value of the smaller displacement is α, the length is l, and the other displacement value is β, and α + l ≦ β, then it is determined that “no overlap”, and if not, “overlap”. To do.

FIG. 4 is an example of comparison between “p → I1” and “p → I2” in the program of (a), and the structure pointed to by the pointer variable p has the structure shown in (b). Both variables are judged to be “not overlapping”.

When the pointer variables are different in processing step 23, the processing is divided in processing step 26 depending on whether or not the predetermined processing designation is taken and the safety side is judged. Here, the determination on the safety side is a process of determining “overlap” in more cases when there is no particular regulation in the program. On the other hand, the judgment that is not on the safe side is based on the assumption that "the pointer used to point to a type different from the declaration does not point to the same area as another pointer (such a program is not written)". Is a process in which the condition of the “overlap” determination is narrowed.

Therefore, in the case of the safety side, it is identified in the processing step 27 whether the structures are the same, and if the structures are the same, the processing step 25 is performed.
Then, as described above, the relative position and the length in the structure are compared to determine the overlap, and if they are not the same structure, the processing step 28
Then, the types of both elements pointed to by the pointers are identified, and if the types are the same, "overlap" is determined, and only if the types are different, "no overlap" is determined. However, if it is determined that the structure is not on the safe side, the determination is made in processing step 29, and if the structures are the same, the same processing is performed as described above, but if the structures are different, it is immediately determined that “no overlap”.

〔The invention's effect〕

As is clear from the above description, according to the present invention, in the optimization process of the compiler of the programming language having the pointer function in the computer, the detection of the pointer optimization target can be efficiently processed, so that the processing cost is low, There is a remarkable industrial effect that it is possible to generate a target program with excellent execution efficiency.

[Brief description of drawings]

FIG. 1 is a flow chart of processing showing the configuration of the present invention, FIG. 2 is a block diagram showing a configuration example of a computer, FIG. 3 is a flow chart of overlap determination processing of the present invention, and FIG. 4 is the position and length of data. And FIG. 5 is a diagram for explaining an example of the program. In the figure, 1 is a compiler, 3 is a source program, 4 is a target program, 5 is a control unit, 6 is an analysis processing unit, 7 is an intermediate language program, 8 is an optimization processing unit, 9 is an object generation unit,
10 to 13 and 21 to 29 indicate processing steps.

Claims (1)

(57) [Claims] 1. A compiler processing apparatus for interpreting a source program represented by a predetermined language having a function of pointing structure data by a pointer variable, and generating an object program of a format to be executed by a computer. Of the plurality of instructions forming the source program, two instructions for operating the same area in the same structure by using a pointer to the structure are used, and the latter of the two instructions is converted into a high-speed direct instruction. It is determined whether or not there is an instruction that is selected as a target of the replacement optimization processing and that changes a partial area of the structure pointed by the pointer variable, between the two instructions. A part of the structure based on the start position and length of the area to be changed by the instruction that changes a part of the structure, and the start position and length of the area operated by the two instructions An optimization process for determining whether or not the area to be changed of the instruction to be changed and the area to be operated by the two instructions overlap, and only when the areas do not overlap, the latter instruction of the two instructions is replaced with a high-speed direct instruction. A compiler processing device characterized by controlling so as to perform the following.