JPH05257707A - Compiling system - Google Patents

Abstract

PURPOSE:To directly vectorize an operation expression directly describing arrangement without degrading performance as a compiler by efficiently converting intermediate work/arrangement information into an intermediate word. CONSTITUTION:A compiler 1 is composed of a syntax analysis part 2, intermediate part (provided with a vector transformation part and an optimizing part) 3 and code generation part 4. An intermediate word 6 remaining the form of arrangement operation for the operation expression of arrangement is generated, it is judged whether the operand of the intermediate word 6 expresses the entire arrangement or not and when the entire arrangement is expressed, information indicating a vector length, initial value and increased component value required for generating a vector instruction is added to the intermediate word 6 based on information indicating the size of arrangement for the operation expression of arrangement. Then, the intermediate word 6 is converted into an intermediate word 7 capable of generating the vector instruction, and the vector instruction is generated by the converted intermediate word 7. When the entire arrangement is not expressed the intermediate word is deformed into a loop repeating intermediate word, and the vector instruction is generated by the deformed intermediate word.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a compiling method for generating a target program for efficiently executing a program described by an array operation function on a vector processor system.

[0002]

2. Description of the Related Art Conventionally, a "compiler" has been used as a vectorizing method for executing a loop iteration type array operation in a source program on a vector processor system.
The method described in (1981, Ikuo Nakata Sangyo Tosho, pp. 258 to 263) is known.

[0003]

However, the vectorization method described in the above document does not consider the case of vectorizing an arithmetic expression that directly describes an array. In the case of vectorization, it must be transformed into an intermediate word in the form of repeating the operation of array elements and then vectorized, and in order to perform vectorization from the transformed state,
Sequence data flow analysis and subscript analysis must be performed. Therefore, there is a problem that the performance as a compiler deteriorates.

An object of the present invention is to provide a compiling method capable of directly vectorizing an arithmetic expression directly describing an array without degrading the performance as a compiler.

[0005]

In order to solve the above object, the present invention generates an intermediate word in which an array operation expression is left in the form of array operation, and whether the operand of the intermediate word represents the entire array. If it represents the entire array, the information indicating the vector length, the initial value, and the increment value necessary for generating the vector instruction based on the information indicating the size of the array of the array operation expression is added. It is added to an intermediate word and transformed into an intermediate word capable of generating a vector instruction, and this transformed intermediate word generates a vector instruction. , The vector command is generated by this modified intermediate language.

[0006]

According to the above-mentioned means, an intermediate word in which the array operation expression is left in the form of array operation is generated, it is judged whether or not the operand of the intermediate word represents the entire array, and the entire array is represented. If so, information indicating the vector length, initial value, and increment value necessary to generate a vector instruction based on the information indicating the array size of the array operation expression is added to the intermediate word to generate a vector instruction. Because it is transformed into an intermediate language that is possible,
Array data flow analysis and subscript analysis to check the availability of vectorization of the array is not required, and the dependency of the array is not required. Directly written arithmetic expressions can be vectorized.

[0007]

An embodiment of the present invention will be described below with reference to the drawings.

FIG. 1 is an overall configuration diagram showing an embodiment in which the present invention is applied to a FORTRAN compiler. The compiler 1 includes a syntax analysis unit 2, an intermediate unit 3, and a code generation unit 4.
It is composed of and.

The syntactic analysis unit 2 inputs the FORTRAN source program 5 and converts it into intermediate language / sequence information 6. In addition, the intermediate unit 3 including the vector conversion unit and the optimization unit
Inputs the intermediate word / array information 6 and converts it into the intermediate word 7 after vector conversion. Furthermore, the code generator 4
Generates an object program 8 based on the intermediate language 7.

The present invention relates to the intermediate part 3 and efficiently converts the intermediate word / array information 6 into the intermediate word 7.

The detailed operation will be described below with reference to an example of the FORTRAN program.

FIG. 2 shows a FORTRAN source program 5
However, it is assumed that the program unit 9 described in this example has a conventional expression format in which each array element is calculated by loop iteration. In contrast,
It is assumed that the program unit 10 uses the array operation function to represent the same array operation as the program unit 9.

When such a FORTRAN source program 5 is input to the syntactic analysis unit 2, the syntactic analysis unit 2 generates an intermediate word 17 represented by loop iteration as shown in FIG. To do. Similarly,
For the program unit 10, an intermediate word 19 which retains the form of array operation as shown in FIG. 4 is generated.

Therefore, the intermediate part 3 has such an intermediate language 1
Vector conversion processing is performed on 7 and 19.

FIG. 5 is a flow chart showing the procedure of the vector conversion process. First, it is checked whether or not there is an array operation for the intermediate words 17 and 19 input from the intermediate word / array information 6 (step S50). , It is checked whether vectorization is possible (step S51). In this case, whether or not the loop iterative calculation program can be vectorized by the dependency analysis of variables and arrays is described in the above-mentioned document "Compiler", and therefore the description thereof is omitted here.

When it is determined that vectorization is possible when the intermediate word 17 which is not an array operation like the program unit 9 in the example of FIG. 2 is examined, it is possible to generate a vector instruction as shown in FIG. It is transformed into word 18 (step S5)
2).

On the other hand, since the intermediate word 19 described in the program unit 10 by the array operation has the array operation, the operand (A = B +) is next calculated.
It is determined whether or not all of C) represent the entire array (step S53).

In this determination, if part of the operands of the array operation does not represent the entire array, it is transformed into the loop repeat type intermediate word 17 (step S54).

However, when all the operands represent the entire array, based on the information of the array size (100 × 100 = 10000) indicated by the array information,
Calculate vector length, initial value, and increment value (step S5)
5). Then, the information of the vector length, the initial value and the increment value obtained by the calculation is added to the intermediate word 19 and transformed into the intermediate word 18 capable of generating a vector instruction as shown in FIG.

Therefore, the program unit 10 is transformed into the intermediate language 18 capable of directly generating a vector instruction, and therefore, the dependency of the array data flow analysis and the subscript analysis array for checking the vectorization of the array is performed. It is possible to directly vectorize an arithmetic expression that directly describes an array without degrading the processing performance of the compiler because it does not require checking of relations and requires less transformation of intermediate language.

[0021]

As described above, according to the present invention, an intermediate word in which the array operation expression is left in the form of array operation is generated, and it is determined whether or not the operand of the intermediate word represents the entire array. , If the entire array is represented, information indicating the vector length, initial value, and increment value required to generate a vector instruction based on the information indicating the array size of the array operation expression is added to the intermediate word. Then, it is transformed into an intermediate language capable of generating a vector instruction, and a vector instruction is generated by this transformed intermediate language, and if it does not represent the entire array, it is transformed into a loop repeating intermediate language and this transformed Since the vector instruction is generated by the intermediate language, there is an effect that the arithmetic expression directly describing the array can be directly vectorized without degrading the performance as a compiler.

[Brief description of drawings]

FIG. 1 is an overall configuration diagram showing an embodiment of a FORTRAN compiler to which the present invention is applied.

FIG. 2 is an explanatory diagram showing an example of a FORTRAN source program in the embodiment.

FIG. 3 is an explanatory diagram showing an intermediate language expression of repetitive type array calculation in an example.

FIG. 4 is an explanatory diagram showing an intermediate language expression of an array operation in the embodiment.

FIG. 5 is a flowchart showing a vector conversion process of a vector conversion unit in the embodiment.

FIG. 6 is an explanatory diagram showing an intermediate language expression after vector conversion is performed in the example.

[Explanation of symbols]

1 ... Compiler, 2 ... Syntax analysis unit, 3 ... Intermediate unit, 4 ... Code generation unit, 5 ... FORTRAN source program, 6
... intermediate language / sequence information, 7 ... intermediate language, 8 ... object program, 9 ... repeated array calculation program part of FORTRAN source program, 109 ... array operation program part of FORTRAN source program, 17 ... repeated array calculation Intermediate language, 18 ... Vectorized intermediate language,
19 ... Intermediate language of array operation.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Tetsuo Nishida 6-81 Onoe-cho, Naka-ku, Yokohama-shi, Kanagawa Hitachi Software Engineering Co., Ltd. In-house

Claims (1)

[Claims] 1. In a compile method for translating a high-level programming language having an array operation function into a machine language including vector instructions, an array operation expression is generated as an intermediate word in which an array operation form is left, and the operand of the intermediate word is It is determined whether or not it represents the entire array, and if it represents the entire array, the vector length, initial value, necessary to generate a vector instruction based on the information indicating the size of the array of the array operation expression, Information indicating an increment value is added to the intermediate word, transformed into an intermediate word capable of generating a vector instruction, and the transformed intermediate word generates a vector instruction. If the whole array is not represented, loop iteration is performed. A compile method characterized by transforming into a type intermediate language and generating vector instructions by this transformed intermediate language.