JPH06175857A - Program generation processor - Google Patents

Abstract

PURPOSE:To designating a parallel computer so as to generate a partial charge program for arranging data in the memory of respective processor elements (PE) and allocating the execution range of a program without transferring data between the PEs. CONSTITUTION:A division designation processing part 11 decides the allocation lower limit value and allocation upper limit value of an index to be allocated to each PE and stores those values in a divided index storage part 12 each time the division of respective indexes described in a source program 10 is designated. A partial charge program generation processing part 13 decides the attached character range of arrangement data to be allocated to the respective PEs and the repeating range of the program so as to satisfy the allocation lower and upper limit values of the designated index division each time the division of respective arrangement data in the source program 10 is designated and the division of each program is designated.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a program generation processing device for generating a program for a parallel computer in which a plurality of processor elements having distributed memories share the operation of array data.

A single program is shared and executed by a plurality of processor elements (hereinafter referred to as PEs), and each P
When E has its own distributed memory, if the array data accessed by the instruction execution unit of each PE is not located in its own distributed memory, the processing efficiency of the operation decreases because it refers to the memory of another PE. Therefore, it is desirable to arrange the necessary data in each own distributed memory as much as possible, but at the same time, it is desirable to prevent redundant allocation.

[0003]

2. Description of the Related Art FIG. 6 is a block diagram showing a configuration of a parallel computer system. The system includes a plurality of (n units) processor elements PE2-i (i = 1, 2, 3, ... N). Each PE consists of CPU21 and memory 22 as distributed memory.
And PEs are connected by an appropriate communication path 23.

When the PEs share the operations on the array data and execute the array data, the array data are arranged in the memory 22 of each PE2-i. If the array data necessary for the calculation is not stored in the memory 22 in PE2-i, the memory 22 of the PE other than itself storing the array data is transferred via the communication path 23 via the communication path. To access the sequence data.

Description will be given of a case where the operations concerning the array data A and B (declared by DIMENSION A and B in FIG. 7A) shown in the example of the iterative calculation program in FIG. 7 are equally shared by the three processors. To do.

The user of the parallel computer, in order to share the program of FIG. 7 (a) among the three processors (PE2-1-3), as shown in FIG. 7 (b), PE2-1-3 respectively Write a shared program to be executed, divide it evenly into each PE of the array data, and, corresponding to the divided array data,
Adjust the control value of the repetition of the DO statement that controls the subscript range of the processing target of array data.

These sharing programs are translated into PEs.
It is executed by the CPU 21 of 2-1 to 2-3, and in this case, as the array data, the data of the subscript range declared in each sharing program is arranged in each memory 22 of PE2-1 to PE2-1.

That is, PE2-1 has DIMENSION A (1: 100), DIME
NSION B (1: 100), PE2-2 DIMENSION A (101: 200),
DIMENSION B (101: 200), but PE2-3 has DIMENSION A (201:
299) and DIMENSION B (201: 299) are placed.

Here, as is clear from the respective sharing programs, for example, the array B referred to by the CPU 21 of PE2-2 is B
B (100) to B (199) for (I-1), and B (1 for B (I)
01) to B (200), and B (I + 1) to B (102) to B (201).

Therefore, in PE2-2, the array element B (I-
When referencing 1) and B (I + 1), it is necessary to refer to B (100) and B (201) other than arrays B (101) to B (200) arranged in the memory 22 of PE2-2. In that case, as described above, it is necessary to access the memory 22 of each of PE2-1 and PE2-3 via the communication path 23, which causes data transfer between processors and delays access. . Similarly, the need to access the memory of another PE also occurs in PE2-1 and PE2-3.

In this case, for example, as shown in FIG. 7C, if the array B is arranged in the memory of all PEs without being divided, the above-mentioned data transfer between PEs can be avoided, but no access is made at all. A large amount of redundant data that does not require is held in each PE, and the memory cannot be effectively used.

[0012]

When the array data is distributed in the memory of each processor and one program is shared and executed by a plurality of processors, B (100) described above in the example of FIG. 7B is used. ) And B (201), there is a problem that data transfer between processors occurs at the boundary of array data, and the processing efficiency of the parallel computer is reduced.

According to the present invention, it is easy to generate a shared program in which the array data is arranged in the memory of each PE and the execution range of the program is assigned so that the data transfer between the PEs does not occur in the parallel calculation. The object is a program generation processing device that can be specified.

[0014]

FIG. 1 is a block diagram showing the configuration of the present invention. The figure shows the configuration of the program generation processing device, which is composed of a plurality of PEs having a distributed memory, and for the array data divided and arranged in the distributed memory of each PE, the PEs share the operation of the array data. Is a program generation processing device for generating each shared program 14 for execution by a given original program 10,
It has a division designation processing unit 11, a division index storage unit 12, and a sharing program generation processing unit 13.

The division designation processing unit 11 determines, for each predetermined index division designation described in the original program 10, based on the number of PEs specified in the index division designation, the index value range, the number of left sleeves and the number of right sleeves. , When the index range is equally divided into the PEs without overlapping, a value smaller by the number of left sleeves than the index value of each lower limit is set as each allocation lower limit value, and is larger than the index value of each upper limit by the number of right sleeves. Using the value as each allocation upper limit value, the allocation lower limit value and the allocation upper limit value of the index allocated to each PE are determined, and the allocation lower limit value and the allocation upper limit value are stored in the divided index storage unit 12.

For each predetermined array data division designation of the original program 10, the sharing program generation processing unit 13 assigns to each PE of the designated sequence data according to the array data and index division designation designated in the array data division designation. The allocation subscript range is determined to be a value that satisfies the allocation lower limit value and the allocation upper limit value held in the split index storage unit 12 for the specified index division designation and does not exceed the subscript range of the array data.

Further, the shared program generation processing unit 13 determines, for each predetermined program division designation of the original program 10, each repeat control value of the designated program portion in accordance with the program division designated in the program division designation and the index division designation. A value that satisfies the allocation lower limit value and the allocation upper limit value held in the split index storage unit 12 for the specified index division designation in the allocation control value range to the PE, and does not exceed the control value range of the program. To decide.

The sharing program generation processing unit 13
For each of the above, the original program 10 is modified by the determined allocation subscript range and allocation control value range to generate each of the sharing programs 14.

[0019]

With the program generation processing device of the present invention, each P
In order to generate a shared program to be executed in E from the original program, it is possible to specify on the original program one or more index division designations and array data and program parts to which those index division designations are applied. become.

In each index division designation, an index value range, the number of PEs to be divided, and the number of left and right sleeves are designated with respect to an index for division in correspondence with a subscript of array data or a repetition control value of a program part. For example, a name for identifying each index division designation is given. In addition,
The number of left and right sleeves is 0 or a positive integer.

In each array data division designation, the array name of the array data and the index division designation name applied to the division are designated, and in the program division designation, the program part (for example, DO loop) to be shared in parallel and the index division designation are designated. Specify the name and.

The division designation processing unit 11 obtains the lower limit and the upper limit of the index to be assigned to each PE from the index division designation as the allocation lower limit value and the allocation upper limit value as described above, and divides them by, for example, the division designation name. The value is stored in the value storage unit 12.

Therefore, by assigning a positive integer to at least one of the number of left and right sleeves, the allocation range of the index determined by the allocation lower limit value and the allocation upper limit value of each PE can overlap with the allocation range of the adjacent PE. .

The sharing program generation processing unit 13 determines, for each array data division designation of the original program, an assigned subscript range of the array data to each PE according to the designated index division designation, and for each program division designation. According to the designated index division designation, the allocation control value range of the repetition control value of the designated program part to each PE is determined, the original program is modified by the determined allocation subscript range and the allocation control value range, and each PE is modified. To generate a sharing program.

As described above, it becomes possible to easily specify, on the original program, to divide the sequence data and the program by partially overlapping them. For example, in the case of the conventional example, the number of left and right sleeves is 1 for each. By designating and assigning the range of array data B to PE2-2 to B (100) to B (201), array elements B (100) and B (201) are PE2-1, PE2-3. , And data transfer between PEs can be eliminated.

[0026]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS As shown in FIG. 6, a memory distributed parallel computer has a plurality of PE2-i (i) having a CPU 21 and a memory 22.
= 1,2, ..., n), but these PE2-i are 1
When a single program is shared and executed by a plurality of units, the user writes instruction information for dividing the original program into shared programs into the original program at the source program level and inputs it to the program generation processing device of the present invention. To do.

The division instruction includes an index division designation,
There are array data division designation and program division designation. In the index division designation, the number of PEs to be shared, the section information of the index that is the division reference of the program instruction or data, and the number of left sleeves and right sleeves are specified. To do.

An example of the original program to which the division instruction information is added will be described with reference to FIG. The program shown in FIG. 3 (a) is the program shown in FIG. 7 to which the division instruction information according to the present invention is added.
PE2-i (i = 1,2,3) of each unit is planned to be shared and executed.

As division instruction information for evenly dividing and executing this program on a plurality of PEs, "PARTITION" for describing the index division designation as shown in the figure.
Statement "and the" LOCATE statement "that describes dividing the array data by applying the index partitioning specification, and the" LOCATE statement "that describes applying the index partitioning specification to partition the range of control values for repeated calculation. PARALLEL DO statement ".

The left of the equal sign (P, Q in the example in the figure) of the operand of the PARTITION statement is the name of the index split specification, and the right of the equal sign is the number of PEs (= 3) and the index section before split. (1 to 300) and the number of left and right sleeves. However, if the description of the number of left and right sleeves is omitted (in the case of Q in the figure), 0 is specified for both the number of left and right sleeves.

In the LOCATE statement, the name B of the array data and the name of the index division designation applied to the division are described in parentheses as shown in the figure, so that the subscript range of the designated array can be set as described later. Specifies to be assigned to each PE according to the designated index division designation.

The PARALLEL DO statement is written by replacing it with the necessary "DO statement" as shown in the figure. By specifying ": name" at the right end, the index division specification of "name" is specified, and the DO statement is displayed. The range of control values for controlling the repetition of P is set according to the specified index division specification as described below.
Specifies to be assigned to E.

The partition designation processing unit 11 in FIG. 1 is based on the PARTITION statement of index partition designation, and the i-th PE (PE2-
The allocation lower limit value x _i and the allocation upper limit value y _{i of} the index to be assigned to i) are obtained as follows and stored in the split index storage unit 12 together with the index split designation name.

Allocation lower limit value x _i = max (a, c _i −e) Allocation upper limit value y _i = min (b, d _i + f) However, in the above equation, the lower limit value of the entire range of the index = a, the upper limit value = b, the lower limit value of the index of PE2-i in the case of dividing into each PE without overlapping the-index = c _i, the upper limit value = d _i, the number of, left sleeve value = e, the right sleeve number of values = F 2, max (p, q) is a function that takes the larger value of p and q, and min (p, q) is a function that takes the smaller value of p and q.

Therefore, according to the index division designation P in the example of FIG. 3, the range of indexes assigned to the three PEs is P: (1: 101), (100: 201), (200: 300), and the index division is performed. In the designated Q, it becomes Q: (1: 100), (101: 200), (201: 300).

The sharing program generation processing unit 13 divides the subscript range of the array data with the LOCATE statement, and PARALLEL described later.
When the range of the repeat control value of the program is divided by the DO statement, the lower limit value sx _i and the upper limit value sy _i of the assigned subscript range or assigned control value range of each PE2-i are determined by the following formula.

Lower limit value sx _i = max (sa, x _i ) Upper limit value sy _i = min (sb, y _i ) However, in the above equation, the lower limit value of the subscript range or the control value range = sa, the upper limit value = s
b, the allocation lower limit value = x _i and the allocation upper limit value = y _i of PE2-i according to the designated index division specification.

Therefore, in the example of FIG. 3A, for example, the assigned subscript range of the array data B to PE2-3 is as follows: lower limit value sx ₃ = max (sa, x ₃ ) = max (1,200) = 200 upper limit value sy ₃ = min (sb, y ₃ ) = min (300, 300) = 300. By dividing array data A, B and DO statements in this way, a sharing program as shown in Fig. 3 (b) is generated. To do.

In this way, for example, in PE2-3, the array data B is not B (201) to B (300) but B (200) to B (300). Can easily be assigned so as to overlap with PE2-2, so that the state of accessing the memory 22 of a processor other than itself does not occur.

FIG. 4 shows an example of another program (a)
When the program shown in (4) is shared by 10 PEs and processed in parallel, the original program is added with the information of division designation as shown in (b), for example.

By inputting this original program,
The division designation processing unit 11 processes the index division designations P and Q as described above to determine the upper and lower limit values of the index allocation, and the sharing program generation processing unit 13 based on this determines the index division designation P for the array data A. , Sequence data B, C
The index division designation Q is applied to and the division of the array data is determined as shown in FIG. 4 (c).

Also, the index division designation Q,
The index division designation P is applied to the second DO statement to determine division of the repeating range as shown in FIG. 4 (d), and the division program shown in FIG. 5 is generated according to the division result.

FIG. 2 is a diagram showing an example of the processing flow of the present invention. First, the division designation processing unit 11 searches for the PARTITION statement of the original program in processing step 30, and if there is a PARTITION statement, in processing step 31, the index range specified in the PARTITION statement is divided evenly by the specified number of PEs. .

In the processing step 32, the range of the divided index values is corrected by the above expression with the value of the designated number of sleeves,
The allocation lower limit value and the allocation upper limit value of each PE are determined, and the allocation upper and lower limit values determined in processing step 33 are stored in the split index storage unit 10 together with the name.

After processing the above for all PARTITION statements of the original program, the sharing program generation processing unit 13 next
In processing step 34, search for the LOCATE statement of the original program, and
If there is a CATE statement, in process step 35, the corresponding allocation upper and lower limit values are read from the partitioned index storage unit 10 by the index partitioning designation name specified in the LOCATE statement.

In processing step 36, the upper and lower allocation limits of each PE read out are compared with the subscript range defined in the DIMENSION statement of the specified array data, and each PE is calculated by the above equation.
The assigned subscript range of each PE is determined, and in the processing step 37, the DIMENSION statement of the sharing program of each PE is generated according to the assigned subscript range.

When all the LOCATE statements of the original program have been processed, the sharing program generation processing unit 13 next searches the PARALLEL DO statement of the original program in processing step 38,
If there is a PARALLEL DO statement, the PARALEL DO
The corresponding allocation upper and lower limit values are read from the partitioned index storage unit 10 by the index partitioning specification name specified in the LEL DO statement.

At processing step 40, the read allocation upper and lower limit values of each PE are compared with the range of the repetition control value defined in the PARALLEL DO statement, and the allocation control value range of each PE is determined by the above formula. Then, in a processing step 41, a DO statement of the sharing program of each PE is generated according to the allocation control value range.

When all the PARALLEL DO statements of the original program are processed as described above, next, the sharing program generation processing unit 13
In process step 42, the unchanged part of the original program is copied to the assortment program to generate each of the assortment programs.

[0050]

As is apparent from the above description, according to the present invention, in a parallel computer having a plurality of processor elements, data transfer between processor elements is prevented from occurring in the memory of each processor element in parallel computation. There is a remarkable industrial effect that it becomes possible to easily specify to generate a shared program in which the arrangement of the array data and the execution range of the program are assigned.

[Brief description of drawings]

FIG. 1 is a block diagram showing the configuration of the present invention.

FIG. 2 is a flow chart of processing of the present invention.

FIG. 3 is a diagram for explaining division processing by a program example.

FIG. 4 is a diagram for explaining division processing by another program example.

FIG. 5 is a diagram explaining a sharing program.

FIG. 6 is a block diagram of a configuration example of a memory distributed parallel computer.

FIG. 7 is a diagram illustrating an example in which programs are shared in parallel.

[Explanation of symbols]

 2-1, 2-2, 2-n PE (processor element) 10 Original program 11 Partition designation processing unit 12 Partition index storage unit 13 Shared program generation processing unit 14 Shared program 21 CPU 22 Memory 23 Communication path 30 to 42 Processing steps

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Naoki Sueyasu 1015 Kamiodanaka, Nakahara-ku, Kawasaki-shi, Kanagawa Fujitsu Limited

Claims (1)

[Claims] 1. For array data composed of a plurality of processor elements having a distributed memory and divided and arranged in the distributed memory of each processor element, the processor elements share and execute the operation of the array data. Is a program generation processing device for generating each shared program of a given original program, the division designation processing unit (11), the division index storage unit (12),
A divisional program generation processing unit (13), and the division designation processing unit (11) is a processor for designating the index division designation for each predetermined index division designation described in the original program (10). Based on the number of elements, the range of index values, the number of left sleeves, and the number of right sleeves, a value smaller than the lower limit index value by the number of left sleeves when the index range is evenly divided into each processor element without duplication The allocation lower limit value and the allocation upper limit value of the index to be allocated to each processor element are determined by setting each allocation lower limit value and a value that is larger than the index value of each upper limit by the number of right sleeves as the allocation upper limit value. And the allocation upper limit value for the divided index storage unit
(12), and the shared program generation processing unit (13) stores the original program (1
For each predetermined array data division designation of (0), according to the array data and index division designation specified in the array data division designation, the assigned index range of the designated array data to each processor element is designated by the designated index division designation. Is determined to satisfy the allocation lower limit value and the allocation upper limit value held in the division index storage unit (12) and not exceed the subscript range of the array data, and specify each predetermined program division of the original program. Every
In accordance with the program portion and index division designation specified in the program division designation, the allocation control value range of the repeat control value of the designated program portion to each processor element is held in the division index storage unit for the designated index division designation. A value that satisfies the allocation lower limit value and the allocation upper limit value and does not exceed the range of the control value of the program is determined, and the value is determined according to the determined allocation subscript range and allocation control value range for each processor element. Original program
A program generation processing device, characterized in that (10) is modified to generate each of the sharing programs (14).