JPH0250257A - Program optimizing processing method - Google Patents

Abstract

PURPOSE:To improve the execution efficiency, etc., of an object program in an optimizing processing for generating a vectoring object program from a source program of a computer by enabling the apply of a vector instruction, which obtains maximum and minimum values, to be applicable and increasing objects of vectrization. CONSTITUTION:A first block, for which definition concerning arrangement data and the reference of the definition continue in a processing step, is detected, and when there is such a block, a second block, which is a block to continue to the block and refers again the arrangement data in the processing step, is detected. When there is the first and second blocks, a register to hold the defining contents of these arrangement data over the first and second blocks is induced in the processing step and optimization is executed to replace the reference of the arrangement data to the reference of the register. After that, vectoring processing is started. Thus, a relation between the two blocks is related by the register, which is common for the blocks, and the vectrization can be executed to apply the vector instruction, with which the maximum and minimum values are obtained.

Description

[Detailed Description of the Invention] [Summary] Regarding optimization processing for generating a vectorized target program from a source program in a computer, it is possible to apply vector instructions to obtain maximum/minimum values, and to improve vectorization. Aiming at a program optimization processing method that can expand the target, a computer source program is analyzed, the innermost loop is generated with the subscript of the array data of the source program as a control variable, and it is divided into multiple blocks. In the process of generating a vectorized target program from the intermediate text, a first block that defines and references the array data, and a block following the first block that defines and references the array data. A second block to be referenced again is detected, a register is introduced to hold the definition contents of the array data across the first and second blocks, and the reference processing of the array data is optimized. .

[Industrial application field]

The present invention relates to a processing method for optimizing a program in a computer in order to generate a vectorization target program from a source program.

A vector processing device is a processing device that speeds up operations by sequentially processing operations on array data using a so-called pipeline control method. Vectorization is performed to create a target program composed of vector instructions so that a required portion of the program can be executed by a vector processing device.

[Prior art and problems to be solved by the invention] A program to be processed by a vector processing device is, for example, a FOR
In the source program written in the TRAN programming language, the well-known so-called ri.

Vectorization is performed by configuring vector data such that each element is array data in a loop identified by a Do variable, which is a control variable that controls repetition of this loop. or,
In the case of a program with a so-called multiple DO loop, in which a DO loop is further nested inside an O loop, if necessary, execution of vector data of the program is performed only in the innermost loop. Construct multiple loops so that there is a statement, and vectorize using the DO variable of the innermost loop.

A compiler that performs translation processing on a source program first processes the source program as a normal so-called scan data program, analyzes the source program to generate an intermediate text format program, and performs optimization processing on this program. After that, a blank or vectorized target program is generated. Here, the intermediate text is divided into blocks based on the structure of the program, and vectorization processing is also performed in principle on a block-by-block basis.

FIG. 2 is an example of a primitive program for a DO loop written in FORTRAN, and as is well known, the range of the loop is D.

4M of variable 1 specified as a control variable of the statement is increased by 1 from 1 to N, and execution of the loop range is repeated. Inside the loop, the result of the "expression" is assigned to the array data ^(1), and the decision statement rlF・-・-" is used to transfer the result to the variable M(
The initial value is set outside the loop)
and A(I), and if AM is smaller than ^(1), it silently assigns A(I), and if it is smaller, it does nothing and proceeds to the next iteration, and the first executable statement (
This example is a typical program for finding the maximum value of the values defined by the assignment statement rA(1)=expression. When determining the minimum value, the configuration is the same except that the determination conditions are different.

In generating intermediate text, a program including such a judgment sentence is divided into a block of judgment processing based on the judgment sentence and a block of processing selected according to the judgment result, as shown in FIG.

FIG. 3 is a diagram showing the conceptual content and block structure of the intermediate text for the purpose of explanation. By placing the result of the expression set in the register TRI introduced for work in A(1), A(T)
is defined, and necessary judgments are made with reference to its contents. One of the judgment results results in a branch to block 3, which has no processing, for example, and the other judgment result proceeds to the subsequent block 2.

As shown in block 4 in Figure 3, this block 1 is optimized by replacing the reference to A (1) with the reference to register ↑R1, and vectorization of the possible range is performed for each block. . However, since the blocks are divided, processing that spans blocks is not performed, and therefore it is not possible to detect that the program seeks the maximum or minimum value. It is not possible to vectorize the vector by using the vector instruction to find the vector.

An object of the present invention is to provide a program optimization processing method that can expand the scope of vectorization by making it possible to apply a vector instruction for determining the maximum value or minimum value in the above-mentioned cases.

[Means to solve the problem]

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

The figure shows optimization processing before vectorization processing of intermediate text in the innermost loop in a compiler or the like for generating a vectorized target program, and IQ~12 is a processing step.

[For production]

When analyzing a computer source program and generating a vectorized target program from intermediate text generated for the innermost loop with the subscript of the source program's array data as a control variable and divided into multiple blocks,
In processing step 10 of FIG. 1, the compiler detects the first block in which a definition related to array data and its reference continues, and if such a block exists, in processing step 11, the compiler detects the first block in which the definition regarding array data and its reference continues. A second block that exists and references the same array data again is detected.

If there are the first and second blocks, a register that holds the definition of the array data across the first and second blocks is introduced in processing step 12, and the array data is referenced in that register. Optimization is performed to replace the reference with , and then the vectorization process begins.

With the above processing method, two blocks are associated with each other by a register common to the blocks, and vectorization can be performed using a vector instruction to obtain a maximum value or a minimum value.

〔Example〕

In the optimization process for vectorization, the compiler
For example, by scanning the intermediate text in FIG. 3(b), in processing step 10 in FIG. 1, the array data to be vectorized with the control variable of the innermost loop as the index variable is
Find a block with a definition followed by a reference step, and if there is no such block, terminate the process.

Figure 3 (If there is a block that has consecutive definition and reference steps, such as block 4 in bl, processing step 1
1 scans the next block 2, detects that the same array data defined and referenced in the previous block is referenced again, and if there is no such reference, this process ends.

If there is a re-reference as described above, processing step 12
Then, a register (referred to as a common register) that is commonly used across blocks 4 and 2 is introduced, and a common register N is introduced, as in block 5 illustrated in FIG. 3 (C1).
By R2, the local register TRI is replaced in block 4, and as in block 6, the reference to the array data A(1) of block 2 is also replaced with the reference to register NR2, and the process ends.

In the vectorization process, vectorization is performed through blocks 5 and 6 that are related by NR2, and since there is a definition/reference/reference pattern for the same vector data, this The portion is processed to correspond to a vector instruction for finding the maximum or minimum value.

〔Effect of the invention〕

As is clear from the above description, according to the present invention, it is possible to apply vector instructions to obtain maximum/minimum values in optimization processing for generating a vectorized target program from a source program in a computer. As a result, the targets of vectorization can be expanded, which has a significant industrial effect of improving the execution efficiency of the target program.

[Brief explanation of the drawing]

FIG. 1 is a process flowchart showing the configuration of the present invention, FIG. 2 is a diagram showing an example of a source program, and FIG. 3 is an explanatory diagram of intermediate text. In the figure, 1 to 6 indicate blocks, and 10 to 12 indicate processing steps.・Representative Patent Attorney Sada Igeta-1 (Processing flowchart showing the configuration of the present invention.

Claims (1)

[Scope of Claims] An object that is vectorized from intermediate text that is generated by analyzing a computer source program and is divided into a plurality of blocks for the innermost loop in which the subscripts of the array data of the source program are used as control variables. In the process for generating a program, a first block that defines and references the array data; and a second block that follows the first block and that references the array data again. (11), and introduces a register that holds the defined contents of the array data across the first and second blocks to optimize reference processing of the array data (12). A program optimization processing method characterized by: