JP2000148731A - Information processor and record medium - Google Patents

Abstract

PROBLEM TO BE SOLVED: To perform the efficient vector process of an array built-in function. SOLUTION: An array built-in function detecting means 1 detects the array built-in function when an executable program is executed. An array built-in function kind specifying means 3 specifies the kind of the detected array built-in function. A threshold generating means 4 generates the threshold corresponding to the specified kind. A vectorizable part specifying means 2 analyzes the array included in the detected array built-in function to specify a vectorizable part and refers to the threshold supplied from the threshold generating means 4 to supply the vectorizable part to a vector processing means 5 when its length is larger than the threshold or to a scalar processing means 7 when not. An execution environment specifying means 6 specifies the execution environment of the program and supplies it to the vector processing means 5. The vector processing means 5 performs the vector process of the array built-in function according to the processing procedure corresponding to the environment specified by the execution environment specifying means 6. The scalar processing means 7 performs the scalar processing of the array.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an information processing apparatus and a recording medium, and more particularly to an information processing apparatus capable of performing vector processing and a computer-readable recording medium storing a program to be executed by a computer capable of performing vector processing. About.

[0002]

2. Description of the Related Art For example, when performing scientific calculations, a plurality of data are often repeatedly processed. Therefore, there is a computer having a vector operation unit for speeding up such repetitive processing. are doing.

In such a computer, a vector,
That is, since the processing can be performed on the one-dimensional array data in a lump, it is substantially proportional to the vector length that can be processed at one time as compared with the conventional scalar processing that processes the data one by one. Processing can be speeded up.

A program (executable program) executed by a computer having a vector operation unit
Is generated by compiling a source program created in a high-level language such as Fortran or C with a compiler. At this time, the compiler
It is necessary to determine whether each source code included in the source program can be vectorized and, if vectorization is possible, generate an appropriate vector code.

FIG. 9 shows a source code including an "array built-in function" (portion to be vectorized) which is a built-in function that can use an array as an argument. 9 is a flowchart illustrating an example of a process executed when converting to a program. When this flowchart is started, the following processing is executed. [S1] The compiler acquires an array built-in function from a source program. [S2] The compiler determines whether the dimensions and the number of elements of the argument array included in the acquired array built-in function are unknown. If unknown, the process proceeds to step S3. Goes to step S4.

In other words, the compiler determines whether the shape (for example, dimension) of an array included in an array built-in function is unknown, or if an optional argument is used, and the content of the optional argument is unknown. Is determined, and if any of them is unknown, the process proceeds to step S3, and otherwise, the process proceeds to step S4. [S3] The compiler calls a library of array built-in functions and performs a process of converting the target function into a scalar process. [S4] The compiler performs a vectorization process by performing inline expansion of the target function. [S5] The compiler determines whether there is any unprocessed source code, and if so, returns to step S1 and repeats the same processing. Otherwise, the processing ends.

According to the above-described processing, the array included in the array built-in function can be appropriately converted into a vector code, so that the processing can be speeded up when the vectorization is performed. .

[0008]

However, it is often unknown at the time of compiling whether vectorization is possible or not. In such a case, the compiler cannot generate a vector code, so that a scalar can be used as before. There is a problem that the processing is executed and the hardware environment cannot be used effectively.

Therefore, at the time of compiling, a first object for vector processing and a second object for scalar processing
Object and a third object for selecting these two objects,
A method has been proposed to solve the above-mentioned problem by making the object select either the first object or the second object.

However, such a method has a problem that the size of the generated executable program is increased. For example, if the dimensions of the array to be generated are unknown, the first
, The size of the generated executable program increases.

Also, when the index of the array takes discrete values, there is another problem that vectorization is difficult. For example, A [3,3,3] which is a three-dimensional array
Assuming that the array is stored in the memory as shown in FIG.
0 (in this example, stored by the column priority method). That is, when the first-dimensional index (the leftmost value in parentheses) goes through the values from 1 to 3, the second-dimensional index (the central value in parentheses) is incremented by one. And the second dimension index is 1
When the value of “3” is cycled, the value of the third dimension is incremented by “1”. In this case, since each index can take all the values from 1 to 3, A [1,1,1] to A [1,1,1]
Elements can be stored in all the areas [3, 3, 3] (in FIG. 10, a mark “「 ”indicating that the elements can be stored is attached to all the areas).

In some cases, it is possible to specify the storage interval of elements when declaring an array. For example, as shown in FIG. 10, a three-dimensional array A [1:
3: 2, 1: 3: 2, 3], the first dimension and the second dimension are represented by a “triple expression” (1: 3: 2).
This indicates that the index increases from 1 to 3 every two. As a result, the possible values of the first and second dimension indices are two types, “1” and “3”.

Therefore, A [1: 3: 2, 1: 3: 2,
In the case of [3], the elements of the array are only in the area where the first dimension index is either “1” or “3” and the second dimension index is either “1” or “3”. Can be stored. Therefore, this sequence A [1: 3:
In the case of [2: 1: 3: 2,3], elements are stored in discrete areas as shown in FIG. 10 (in FIG. 10, marks “○” are discretely marked).

When elements are stored in discrete areas in this way, problems may occur in vectorization processing. That is, when the vector operation is performed by the vector operation unit, the operation target is specified by the head address of the memory storing the vector (one-dimensional array) to be operated and the vector length (data length). At this time, if the elements are arranged at predetermined intervals (for example, every other element), the interval is designated as "stride" (in the above example, "1" is used as the stride). By doing so, vector operations can be performed.

However, in the example shown in FIG. 10, since the arrangement intervals between the elements are not constant, there is a problem that it is difficult to target such an array for the vector operation.

The present invention has been made in view of such a point, and an object of the present invention is to provide an information processing apparatus capable of efficiently performing vector processing of an array built-in function.

[0017]

According to the present invention, in order to solve the above-mentioned problems, in an information processing apparatus capable of performing vector processing shown in FIG. 1, an array built-in function included in a program to be processed is used. Array built-in function detecting means 1 to be detected at the time of execution, and vectorizable part specifying means for analyzing an array included in the array built-in function detected by the array built-in function detecting means 1 and specifying a vectorizable part 2 and a vector processing means 5 for performing vector processing on the vectorizable part specified by the vectorizable part specifying means 2.

Here, the array built-in function detecting means 1
Detects array built-in functions included in the program to be processed at execution time. The vectorizable part specifying means 2 analyzes the sequence included in the array built-in function detected by the array built-in function detecting means 1 and specifies a vectorizable part. The vector processing means 5
The vectorizable portion specified by the vectorizable portion specifying means 2 is subjected to vector processing.

[0019]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram illustrating a configuration example of an embodiment of the present invention. In this figure, when an executable program to be processed is executed, an array-embedded function detection unit 1 detects an array-embedded function from the executable program.

The array built-in function type specifying means 3 specifies the type of the array built-in function detected by the array built-in function detecting means 1. The threshold value generating means 4 generates a threshold value corresponding to the type of the function specified by the array built-in function type specifying means 3.

The vectorizable part specifying means 2 analyzes an argument array included in the array built-in function detected by the array built-in function detecting means 1 to specify a vectorizable part, and a threshold value generating means 4. Is supplied to the vector processing means 5 if the length of the vectorizable portion is larger than the threshold value, and to the scalar processing means 7 otherwise.

The execution environment specifying means 6 specifies the execution environment of the program and supplies it to the vector processing means 5. The vector processing means 5 performs vector processing on the array reading function according to a processing procedure according to the environment specified by the execution environment specifying means 6.

The scalar processing means 7 reads a library corresponding to the array read function and performs scalar processing on the array read function. Next, the operation of the above embodiment will be described.

FIG. 2 is a flowchart for explaining the overall operation of the above embodiment. When this flowchart is started, the following processing is executed. [S10] The array built-in function detecting means 1 inputs an executable program. [S11] The vectorizable part specifying means 2 extracts an array built-in function included in the executable program and specifies a vectorizable part of the array elements included in the array built-in function. I do.

The details of this process will be described later with reference to FIG. [S12] The threshold value generating means 4 generates a threshold value corresponding to the function type specified by the array built-in function type specifying means 3.

The details of this process will be described later with reference to FIG. [S13] The vectorizable portion specifying means 2 compares the vector length of the vectorizable portion specified in step S11 with a threshold value. If the vector length is larger than the threshold value, the process proceeds to step S14. In this case, the process proceeds to step S16. [S14] The vector processing unit 5 executes the execution environment specifying unit 6
Select the vectorization process corresponding to the execution environment specified by.

The details of this process will be described later with reference to FIG. [S15] The vector processing means 5 executes vector processing. [S16] The scalar processing means 7 executes scalar processing. [S17] The array-embedded function detection means 1 determines whether or not the execution of the executable program has been completed. If the execution has been completed, the processing ends (END). Otherwise, the process proceeds to step S11. Return and repeat the same process.

Next, with reference to FIG. 3, the details of the "vectorizable portion specifying process" shown in FIG. 2 will be described. In the processing shown in FIG. 3, when the index of the array takes discrete values, a vectorizable portion of the elements stored in the array is specified. When this flowchart is started, the following processing is executed. [S20] The vectorizable part specifying unit 2 acquires the array built-in function detected by the array built-in function detecting unit 1. [S21] The vectorizable part specifying means 2 determines whether or not the dimension n of the array included in the array built-in function is one-dimensional. If the dimension n is one-dimensional, the process proceeds to step S22; In the case of, the process proceeds to step S24. [S22] The vectorizable portion specifying unit 2 sets the value of SIZE [1] to the variable VL indicating the vector length.

Here, SIZE [x] is the x-th of the array.
Indicates the number of elements of the dimension element. For example, the sequence A [4,
SIZE [1] = 4 for [5,1], and
SIZE [2] = 5, SIZE [3] = 1. [S23] The vectorizable part specifying means 2 sets the variable AD indicating the dimension of the array to "1". [S24] The vectorizable part specifying means 2 sets the variable WO
The value of TERM [1] is substituted for RK.

Here, TERM [x] is the x-th of the array.
It shows the distance on the memory when the index of the dimension increases by one. For example, a three-dimensional array A [x, y,
z], TERM [1] is A [x, y, z] and A
Shows the distance between [x + 1, y, z]. Also, TERM
[2] indicates the distance between A [x, y, z] and A [x, y + 1, z].

Therefore, in step S24, the variable WOR
For K, a distance in the memory (a value corresponding to the data amount of the element) with respect to a case where the index of the first dimension of the array is increased by 1 is substituted. [S25] The vectorizable part specifying means 2 initializes the variable i to a value of 1. [S26] The vectorizable part specifying means 2 sets the variable WO
The value obtained by multiplying RK and SIZE [i] is stored in a variable WORK.

Assuming that the processing is the first time, the variable WORK stores a distance (a value corresponding to the data amount of each element) when the index of the first dimension is increased by one. Since SIZE [1] stores the number of elements in the one-dimensional direction of the array, their product is 2
This corresponds to the distance in the dimensional direction.

For example, taking A [3,3,3] shown in FIG. 10 as an example, in the first processing, the variable WORK contains A
The distance between [1,1,1] and A [2,1,1] is stored, and the value of SIZE [1] is “3”. [1,1] and A [1,2,
1] (= TERM [2]). [S27] The vectorizable part specifying unit 2 executes the TERM
It is determined whether or not the value of [i + 1] is equal to the value of WORK. As a result, if they are equal, the process proceeds to step S28; otherwise, the process proceeds to step S30.

That is, the value of the variable WORK is set to TERM [i + when the elements are stored consecutively (when the index in that dimension has no discrete value).
1], the process proceeds to step S28 in that case, and otherwise (if the dimension index has discrete values), the process proceeds to step S30.
Proceed to. [S28] If the value of the variable i is smaller than the value (n-1) obtained by subtracting "1" from the dimension n of the array, the vectorizable portion specifying means 2 proceeds to step S29. Goes to step S32. [S29] The vectorizable portion specifying means 2 increments the value of the variable i by "1" and returns to step S26. [S30] The vectorizable part specifying unit 2 applies SIZE [1] × SIZE to the variable VL indicating the vector length.
[2] × ... × SIZE [i] is substituted.

For example, when TERM [i + 1] ≠ WORK is determined in the second (i = 2) processing, the vector length VL is 1 because the third-dimensional index has discrete values. SIZE [1] × SIZE [2] which is the product of the dimension and the number of elements in the second dimension is adopted.

As an example, in the case of A [3,3,1: 3: 2], as shown in FIG. 4, elements are stored only in the area where the third dimension index is "1" or "3". And the number of consecutive elements stored is
“9” (= 3 × 3). Therefore, this value “9” is
The vector length is VL. [S31] The vectorizable portion specifying means 2 stores the value (ni) for the variable AD indicating the dimension of the array.

In the example of FIG. 4, since the loop is released when i = 2, (ni) = (3-2) = 1, and the dimension of the array is "1". That is, a group of elements stored continuously in FIG. 4 is handled as one vector (one-dimensional array). [S32] The vectorizable part specifying unit 2 applies SIZE [1] × SIZE to the variable VL indicating the vector length.
The value of [2] ×... × SIZE [n] is substituted.

That is, when the indices of all the dimensions are continuous, the VL indicating the vector length includes SIZE
The value of [1] × SIZE [2] ×... × SIZE [n] is substituted. For example, in the case of A [3,3,3],
The value “27” (= 3 × 3 × 3) is substituted for the variable VL. [S33] The vectorizable portion specifying means 2 stores the value “1” for the variable AD indicating the dimension of the array.

That is, when the indices of all the dimensions are continuous, all the elements are treated as one vector, so that the dimension is set to "1". Through the above processing, it is possible to specify a part that can be subjected to vector processing among the elements included in the array.

Next, details of the "threshold generation process" shown in FIG. 2 will be described with reference to FIG. When this flowchart is started, the following processing is executed. [S40] The array built-in function type specifying unit 3 specifies the type of the built-in function detected by the array built-in function detecting unit 1, and notifies the threshold generating unit 4 of the type. [S41] The threshold value generating means 4 acquires a threshold value corresponding to the type of the function notified from the array built-in function type specifying means 3 from a table (not shown). Then, the process returns to the original process.

FIG. 6 is an example of a table referred to by the threshold value generating means 4. In this example, the threshold value is "10" when the function executes the summation operation of the array, and the threshold value is "5" in the case of the maximum value search, array element shift, or array copy. , “3” and “2”.
As such a threshold, an appropriate numerical value is set for each of the array built-in functions executable by the information processing apparatus.

Next, with reference to FIG. 7, the details of the “vectorization processing selection processing” shown in FIG. 2 will be described. When this flowchart is started, the following processing is executed. [S50] The execution environment specifying means 6 specifies the hardware environment and notifies the vector processing means 5.

As the hardware environment, for example, a corresponding hardware environment is selected from hardware types (for example, hardware A, hardware B, etc.) classified according to the processing capability of the vector operation unit and the like. You. [S51] The vector processing unit 5 executes the execution environment specifying unit 6
Selects the vectorization processing (program) corresponding to the hardware environment notified from the above, and returns to the original processing (return).

For example, the vector processing means 5 has a correspondence table as shown in FIG. 8, and when the execution environment specifying means 6 notifies that its hardware is hardware A, , A corresponding vectorization process A is selected. Note that the vectorization processes A to D are processes optimal for hardware A to D, respectively. Then, returning to the original processing, the vector processing is executed.

According to the above processing, when an executable program including an array built-in function is executed, a vectorizable part of the elements of the array included in the array built-in function is specified and the processing is performed. It is determined whether or not the vectorization processing is valid according to the type.If it is determined that the vectorization processing is valid, a vector processing suitable for the hardware environment under execution is selected and vectorization is performed. , Vector processing is executed, so that vector processing can be performed even for an array whose form is not determined at the time of compilation, and hardware resources can be effectively used.

If it is determined that the vector processing is not valid, the scalar processing is executed. Therefore, the optimum processing method is selected according to the form of the array, the hardware environment, or the type of the function. It is possible to do.

The above processing functions can be realized by a computer. In this case, the processing contents of the functions that the information processing apparatus should have are described in a program recorded on a computer-readable recording medium, and the above processing is realized by the computer by executing the program on the computer. Is done. Examples of the computer-readable recording medium include a magnetic recording device and a semiconductor memory.

When distributing in the market, a CD-ROM
(Compact Disk Read Only Memory) or a program stored in a portable recording medium such as a floppy disk and distributed, or stored in a storage device of a computer connected via a network and transferred to another computer via the network You can also. When the program is executed by the computer, the program may be stored in a hard disk device or the like in the computer, loaded into the main memory and executed.

[0049]

As described above, according to the present invention, in an information processing apparatus capable of vector processing, an array built-in function detecting means for detecting an array built-in function included in a program to be processed at the time of execution, Vectorizable part specifying means for analyzing an array included in the array built-in function detected by the array built-in function detecting means and specifying a vectorizable part, and vectorization specified by the vectorizable part specifying means And vector processing means for performing vector processing on possible parts, so that it becomes possible to execute vector processing even for an array whose form is uncertain at the time of compilation.
Hardware resources can be used effectively.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration example of an embodiment of the present invention.

FIG. 2 is a flowchart for explaining the overall operation of the embodiment shown in FIG. 1;

FIG. 3 is a flowchart illustrating details of a “vectorizable portion specifying process” illustrated in FIG. 2;

FIG. 4 is a diagram illustrating an example of a storage mode of elements in an array whose index has discrete values.

FIG. 5 is a flowchart illustrating details of a “threshold generation process” illustrated in FIG. 2;

FIG. 6 is an example of a table referred to by a threshold generation unit.

FIG. 7 is a flowchart illustrating details of a “vectorization process selection process” illustrated in FIG. 2;

FIG. 8 is an example of a table referred to by a vector processing unit.

FIG. 9 is a flowchart illustrating an example of a conventional array vectorization determination process.

FIG. 10 is a diagram illustrating an example of an array in which indices have continuous values, and an example of a storage mode of elements in an array having discrete values.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Array built-in function detecting means 2 Vectorizable part specifying means 3 Array built-in function type specifying means 4 Threshold value generating means 5 Vector processing means 6 Execution environment specifying means 7 Scalar processing means

Claims (5)

[Claims] 1. An information processing apparatus capable of performing vector processing, comprising: an array built-in function detecting unit that detects an array built-in function included in a program to be processed at the time of execution; Vectorizable part specifying means for analyzing an array included in the array built-in function and specifying a vectorizable part, and a vector for vector-processing the vectorizable part specified by the vectorizable part specifying means An information processing apparatus comprising: processing means. 2. A method according to claim 1, wherein said vectorizable portion specifying means specifies, as a vectorizable portion, the elements stored in the array that collectively include portions that are continuous or at regular intervals. The information processing apparatus according to claim 1, wherein 3. An array built-in function type specifying unit that specifies a type of the array built-in function; a threshold generation unit that generates a threshold corresponding to the type of the function specified by the array built-in function type specifying unit; 2. The information processing apparatus according to claim 1, further comprising: scalar processing means for performing scalar processing when the vector length of the vectorizable part specified by the convertible part specifying means is smaller than the threshold. . 4. An execution environment specifying means for specifying an execution environment of the program, wherein the vector processing means executes vector processing according to a processing procedure corresponding to the environment specified by the execution environment specification means. The information processing apparatus according to claim 1, wherein: 5. A computer-readable recording medium on which a program to be executed by a computer capable of vector processing is recorded, wherein the computer detects an array built-in function included in the program to be processed at the time of execution. Built-in function detection means, Vectorizable part specifying means for analyzing an array included in the array built-in function detected by the array built-in function detection means, and specifying a vectorizable part, The vectorizable part specifying means And a vector processing means for performing vector processing on the vectorizable portion specified by the computer.