JP3034582B2 - Compile processing method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a compile processing method for executing compile processing for optimization to generate and output an object corresponding to a source program. The present invention relates to a compile processing method that enables execution of compile processing for optimization without describing an optimization control line.

In order to optimize the operation description of the source program, it is necessary to obtain the data dependency of the array of the operation description. Based on the data dependency of this array, it is possible to determine whether or not it is possible to perform a vector operation on the operation description, and whether or not it is possible to execute the operation description in parallel with a plurality of CPUs. It is because it becomes possible to judge. From now on, FORTRAN77 will analyze the data dependency of the array which has the operation description by configuring the source program to describe the optimization control line (OCL) which displays the data relation of the subscript of the referenced array. The method of supporting processing is adopted.

[Conventional technology]

Conventional compilers use the description contents of the optimization control line described in the source program to identify the data dependency of the array of the operation description to be optimized, and perform vector operation according to the specified data dependency. The processing is performed so as to determine whether the execution is possible or the parallel execution is possible.

For example, the source program shown in FIG.
When the condition of “N> 0” is satisfied, the vector operation becomes possible, and when the condition of “N <0” is satisfied, the vector operation becomes impossible.
When it is known from the data processing conditions and the like that the condition of “N> 0” is satisfied, the programmer who creates the source program determines that “N> 0” as shown in FIG. By writing the optimization control line * VOCL LOOP, N.GT.0 in the source program to indicate that there was something, the compiler was processing to generate the corresponding vector instructions.

The optimization control line described in this manner is used to specify the data dependency of the array of the operation description to be optimized.

[Problems to be solved by the invention]

However, in such a conventional technique, as can be seen from the example of FIG. 5, when there is a loop in a portion whose execution is controlled by the conditional expression, the same as the conditional expression is described as an optimization control line. There was a problem that it had to go. Thus, there is a problem that the load on the programmer increases. If the description of the optimization control line is forgotten, a scalar instruction is generated despite the fact that the vector operation can be performed, and the execution performance of the program is degraded. It is.

The present invention has been made in view of such circumstances,
If there is a loop in a part whose execution is controlled by a conditional expression, a new compile processing method that enables execution of compile processing for optimization without writing an optimization control line in the source program It is intended to be provided.

[Means for solving the problem]

 FIG. 1 is a diagram illustrating the principle of the present invention.

In the figure, reference numeral 1 denotes a data processing device equipped with the present invention, which compiles a source program into a corresponding object program; 2 denotes a source program compiled by the data processing device 1; Object program to be executed.

The data processing device 1 includes a program input unit 10 that receives an input of the source program 2, a program analysis unit 11 that performs syntax analysis of the source program 2 received by the program input unit 10 and generates a corresponding intermediate language, The optimizer 12 performs various optimization processes such as parallelization and vectorization on the intermediate language output from the analyzer 11, and the intermediate language sequence optimized by the optimizer 12 is supported. A code generation unit 13 for generating and outputting the object program 3;

The optimization unit 12 performs the optimization process by:
In addition to detecting the condition determination statement described in the source program 2, the intermediate language corresponding to the conditional expression description of the detected condition determination statement is matched with the intermediate language expression of the optimization control line described in the source program 2. A conditional expression recognizing unit 14 generated in a format, a data dependency recognizing unit 15 for recognizing a data dependency of an array of an operation description of the source program 2 according to the optimization control line information described in the source program 2, An instruction generating unit 16 that generates an optimized instruction according to the instruction form specified by the recognition result of the dependency relationship recognizing unit 15.

[Action]

In the present invention, the conditional expression recognition unit 14
When the condition determination statement described in (1) is detected, an intermediate language corresponding to the conditional expression description of the detected condition determination statement is generated in a format consistent with the intermediate language expression of the optimization control line.

That is, the conditional expression recognizing unit 14 detects a conditional judgment statement described in the source program 2 and fetches the conditional expression description of the conditional judgment statement, and the optimization control line describing the fetched conditional expression description. To generate an intermediate language.

In response to the generated intermediate language, the data dependency recognition unit 15 recognizes the data dependency of the array of the operation description having the condition determination statement described in the source program 2 so that the operation description becomes a vector operation. The instruction form of the optimization process is determined by determining whether the instruction can be executed or can be executed in parallel. Then, in response to this determination processing, the instruction generation unit 16 generates an optimizable instruction that can be generated.

As described above, according to the present invention, the conditional expression of the condition determination statement described in the source program 2 can be processed according to the processing of the conditional expression recognition unit 14 without writing the optimization control line. 15 is notified of the data information. From now on, if there is a loop in the part whose execution is controlled by the conditional expression, the source program 2
Since the optimization compilation process can be executed without writing the optimization control line, the load on the programmer is greatly reduced, and the execution performance of the program due to forgetting to write the optimization control line is reduced. Therefore, it is possible to prevent the occurrence of deterioration.

〔Example〕

 Hereinafter, the present invention will be described in detail with reference to examples.

FIG. 2 shows an embodiment of a flowchart of a process executed by the conditional expression recognition unit 14 described in FIG. 1, and FIG. 3 shows a flowchart of a process executed by the data dependency recognition unit 15 described in FIG. 1 is illustrated.

Next, the operation processing of the present invention will be described in detail with reference to the flowcharts of FIGS.

As shown in the flowchart of FIG. 2, the conditional expression recognizing unit 14 firstly executes the source program 2 in step 1.
Executes the detection processing of the condition judgment statement described therein. Next, in step 2, a process of extracting a conditional expression of the detected condition determination statement is executed. Subsequently, in step 3, a process of converting the conditional expression into an optimization control line is executed. That is,
The data information of the conditional expression is converted into an intermediate language expression specified by the optimization control line. And finally,
In step 4, the data information of the converted conditional expression is incorporated into the intermediate language expression of the optimization control line, and the process ends. That is, the data information of the conditional expression is incorporated as if it were described in the optimization control line, and the processing is terminated.

In response to the processing of the conditional expression recognizing unit 14, the data dependency recognizing unit 15 first retrieves an array reference included in the operation description to be compiled in step 10, as shown in the flowchart of FIG. Execute Next, in step 11, it is determined whether or not the data dependency can be determined from the subscript of the extracted citation. If it is determined that the data dependency is not determined, the process proceeds to step 12, and a process of obtaining the data dependency by referring to the optimization control row is executed. At this time, the conditional expression is handled in the same manner as the description of the optimization control line according to the processing of the conditional expression recognition unit 14. Subsequently, the process proceeds to step 13, where it is determined whether or not the processing has been performed for all pairs of sequence references. On the other hand, if the data dependency is found by the determination in step 11, it is not necessary to refer to the optimization control line, so the process immediately proceeds to step 13 to determine whether or not the processing has been performed for all pairs of array references. .

If it is determined in step 13 that the processing has been executed for all pairs, the process proceeds to step 14 in which it is determined whether optimization can be applied to the operation description to be compiled. And terminate the processing. In other words, if it is described in the determination processing of the applicability of the vectorization processing, when the operation description is executed by the scalar operation, the vector operation is performed in the order of “definition” and “reference”. If it is executed, if it is the same order relationship of "define" and "reference", it is determined that vector operation is possible, while the opposite order relationship of "reference" and "define" If it is determined that vector operation is not possible, and if the operation description is executed by scalar operation, the vector operation If it is executed, it is determined that the vector operation is possible if the same order relationship of "reference" and "definition" is obtained, while the opposite order relationship of "define" and "reference" is performed. , The vector operation By going judged to be impossible, it is to determine whether it is possible to apply the vectorization processing.

In this way, when the applicability of the optimization is determined according to the processing of the data dependency relationship recognizing unit 15, the instruction generating unit 16 described with reference to FIG. Execute the generation process.

In this way, by using the present invention, when there is a loop in a part whose execution is controlled by a conditional expression, the same thing as the conditional expression can be described as an optimization control line,
As shown in FIG. 4, by knowing the dependency of the data in either direction, information such as whether or not a vector instruction can be generated (in the example shown in FIG. 4, both can be vectorized) Can be obtained.

〔The invention's effect〕

As described above, according to the present invention, a data dependency can be obtained for a conditional expression of a condition determination statement described in a source program without writing an optimization control line. From now on, when there is a loop in the part whose execution is controlled by the conditional expression, it is possible to execute the optimization compilation process without writing the optimization control line in the source program, so that the load on the programmer can be increased. Is greatly reduced, and deterioration of the execution performance of the program due to forgetting to write the optimization control line can be prevented.

[Brief description of the drawings]

FIG. 1 is a block diagram showing the principle of the present invention, FIG. 2 is a flowchart of a process executed by a conditional expression recognition unit, FIG. 3 is a flowchart of a process executed by a data dependency recognition unit, and FIG. FIG. 5 is an explanatory diagram of the prior art. In the figure, 1 is a data processing device, 2 is a source program, 3
Is an object program, 10 is a program input section, 11
Is a program analyzer, 12 is an optimizer, 13 is a code generator, 14 is a conditional expression recognizer, 15 is a data dependency recognizer, 16
Is an instruction generation unit.

──────────────────────────────────────────────────続 き Continued on front page (58) Field surveyed (Int.Cl. ⁷ , DB name) G06F 9/45 G06F 17/16

Claims (1)

(57) [Claims] In a compile processing method for compiling a source program, a condition judgment statement described in the source program is detected,
Detecting means for retrieving the conditional expression description of the condition determination statement; and, in accordance with the conditional expression description retrieved by the detecting means, the data dependency of the array of the operation description controlled by the condition determination statement detected by the detecting means. Optimizing control line generating means for generating an intermediate language of an optimizing control line describing information necessary for recognition; and the detecting means according to the intermediate language of the optimizing control line generated by the optimizing control line generating means. Recognition means for recognizing data dependence of an array of an operation description under control of the detected condition determination statement, and instruction generation means for generating an optimization instruction corresponding to an instruction form specified by a recognition result of the recognition means And a compiling method.