JP3156343B2 - Data dependency information processing method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a data dependency information processing method for retaining and using information indicating data dependency on a program in a compiler process in a computer.

[0002]

2. Description of the Related Art In the processing of a compiler for generating a target program from a source program, the source program is usually analyzed first, and an intermediate language defined in a manner suitable for the subsequent processing is used. Convert to intermediate text.

[0003] Next, the intermediate text is subjected to so-called optimization processing such as removal of a common expression, fusion and arrangement of loops, synchronization of memory access, etc., so as to be modified so as to become an efficient target program. Generate the target program from the optimized intermediate text.

[0004] In performing the above-mentioned optimization, an important condition as a constraint of the detection and optimization of the optimization target is a known so-called data dependency. Here, as is well known, for example, data that is an operand of each operation of an intermediate text has a predetermined temporal order relationship between the reference and the definition of the same operand from the viewpoint of program logic. This is the relationship between both operands when they must be drooped.

Accordingly, the compiler provides control information in which bit positions are assigned to names assigned to data, and based on the result of the program analysis, defines the relationship in which data having a certain name is defined by referring to data having a certain name. , The corresponding bit on / off.

Further, in the above information, the data dependency in consideration of the subscript cannot be identified for the array data. Therefore, when it is necessary to identify the data dependency in consideration of the subscript of the array data, the data of the operand array data is identified. The data definition and reference status are analyzed based on the subscripts and loop control information.

However, recognizing the data dependency as necessary as described above is inefficient in the processing efficiency of the compiler.
In addition, since the recognized data dependency is stored in correspondence with the corresponding operation on the intermediate text, if the positional relationship or the internal information of the intermediate text is changed in the optimization process, the same data dependency is stored. Regarding the relationship, there is a case where the program must be analyzed and recognized again, which further impairs the processing efficiency.

According to the present invention, information indicating a data dependency is
It is an object of the present invention to provide a data dependency information processing method that can be efficiently stored and used.

[0009]

FIG. 1 is a flow chart of the processing showing the configuration of the present invention. The figure shows the configuration of a data dependency information processing method in the processing of a compiler that generates an intermediate text in a predetermined language from a source program in a predetermined language and generates a target program from the intermediate text.

In this processing method, a data dependency identifier corresponding to each data dependency of a predetermined range type is provided in processing step 1 for each operand of the intermediate text.

At a processing step 3, a predetermined data dependency on the program logic of the intermediate text is analyzed, and at a processing step 4 to 9, a different numerical value for each of the operand groups having the data dependency is assigned to the operand. Is recorded in the data dependency identifier of the range type.

In the process of optimizing the intermediate text, the data dependency identifier between the operands is identified by the data dependency identifier.

[0013]

According to the processing method of the present invention, for each operand of the intermediate text, a data dependency identifier is provided for each required range type, for example, in a basic region, in a loop, between loops, and the like. Is analyzed, and a unique number for each relationship is recorded in the data dependency identifier of all the corresponding operands.

Therefore, when it is necessary to identify the data dependency during the optimization process, the data dependency identifier of the required operand can be immediately recognized as to whether or not the data dependency exists. Can be found by searching for an operand having a as a data dependency identifier.

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS A data dependency identifier is provided, for example, by enlarging the area of each operand of an intermediate text to provide a necessary number of data dependency identifier areas.

The number of data dependency identifiers provided for each operand is made equal to the number of necessary range types so that different data dependency identifiers are used for each type of program range required in the processing of the compiler. .

The range type includes, for example, in a basic region, in a loop, between loops, and the like, but other ranges may be provided. Here, the basic region refers to the minimum range of the optimization processing determined by the compiler for the convenience of the processing.

The data dependency in the loop refers to the case where the data dependency in one execution of the loop iteration is to be analyzed, and may include a plurality of basic regions. (In the example shown in FIG. 2B, when the range indicated by is the range of one in-loop analysis target, and may be basic regions, respectively). The term "between loops" refers to a data dependency existing between repetitions of a loop.

The numerical value recorded in the data dependency identifier is:
For example, "0" is invalid (data dependency is not analyzed)
And “1” indicates that there is no data dependency. If there is a data dependency, a value greater than or equal to "2" is used to give the group of operands having the dependency the same value unique within the range type.

FIG. 2 shows an example of setting the data dependency identifier determined as described above. FIG. 2A shows an example in a basic region, FIG. 2B shows an example in a loop, and FIG. It is an example.
The figure uses FORTRAN instead of intermediate text for illustration.
A program is shown in the form of a source program of a language, and a data dependency identifier provided corresponding to an operand is indicated by a rectangular frame attached to each variable.

The three numerical values in each frame represent each data dependency identifier in the order of the basic region, the loop, and the loop. In FIG. 2B, the data C (I) is captured as a relationship in the basic region.
It is not necessary to show the relationship in a loop as shown in the figure, but for convenience of processing, it may be shown in both of them.

The processing flow shown in FIG. 1 will be described in detail. In the initial setting of the processing step 1, an area of a data dependency identifier is obtained for each operand of the intermediate text.
Set them all to “0”.

In processing step 2, the range type to be analyzed is determined. Analyzing the intermediate text in processing step 3,
The data dependency identifier of the range type of the current processing target is “0”
For the operand of, an operand having a data dependency is analyzed and searched in a usual manner.

In this case, in the basic region and in the loop, the dependency from the top to the bottom in the program is obtained. As a result, if a data dependency is detected for the operand, the processing goes through processing step 4 to processing step 5
The number starting from “2” is incremented by one each time it is used, and is updated to obtain the current number value, and the data dependency identifiers of all operands of the detected data dependency are set in processing step 6
To record the current number value.

If it is determined that the operand has no data dependency, "1" is recorded in the data dependency identifier of the operand in processing step 7 after processing step 4.

After the above, if there is an operand identified as 0 in the data dependency identifier of the range type of the current processing target identified in the processing step 8, the processing returns to the processing step 3 and the above processing is performed.

When the processing is completed for a certain range type, it is identified in processing step 9; if there is an unprocessed range type, the process returns to processing step 2 and the above processing is repeated.

[0028]

As is apparent from the above description, according to the present invention, in the processing of the compiler in the computer, the information indicating the data dependency can be efficiently held and used. There is a remarkable industrial effect of improving the processing efficiency of the processing.

[Brief description of the drawings]

FIG. 1 is a flowchart of a process showing a configuration of the present invention.

FIG. 2 is a diagram showing a setting example of a data dependency identifier.

[Explanation of symbols]

 1-9 processing steps

Claims (1)

(57) [Claims] 1. A compiler for generating an intermediate text in a predetermined language from a source program in a predetermined language and generating a target program from the intermediate text, wherein a predetermined range corresponding to each operand of the intermediate text is provided. A data dependency identifier corresponding to each type of data dependency is provided (1), a predetermined data dependency on the program logic of the intermediate text is analyzed (3), and a group of the operands having the data dependency is analyzed. Is recorded in the data dependency identifier of the range type of the operand (2, 4 to
9), in the case of the intermediate text optimizing process, the data dependency identifier between the operands required is identified by the data dependency identifier. Processing method.