JP3027388B2 - Program code generation method - Google Patents

Description

Description: BACKGROUND OF THE INVENTION The present invention relates to a program code generation method, and more particularly, to a program for automatically generating a numerical calculation program for performing a numerical simulation of a physical phenomenon in a format suitable for a vector computer. It relates to a code generation method.

[Conventional technology]

2. Description of the Related Art Conventionally, a numerical calculation program is used to perform a numerical simulation of a physical phenomenon in a region where the physical phenomenon occurs. In a program code generation method for automatically generating a numerical calculation program of this type, a shape of an area for analyzing a physical phenomenon, grid division information for setting the grid points to be analyzed in the area by dividing the area into a grid,
A partial differential equation expressing the physical phenomena in the region and a region analysis condition consisting of the boundary conditions of the region are input, and the region is divided into a grid and grid points are set as analysis points in the region. Generates a program code for performing calculation of a discrete expression for performing analysis on grid points at all grid points. in this case,
The region to be analyzed is divided into small regions, and a calculation loop for controlling the calculation order is generated from the index range of the small region. If the region to be analyzed has an index range in two or more directions, multiple discrete formula calculation loops for controlling the calculation order are generated corresponding to the number of dimensions in each direction of each index range. , A program code for calculating the discrete expression of each grid point is automatically generated.

[Problems to be solved by the invention]

Incidentally, the above-described conventional method of generating a program code has, for example, the following problems. That is, the first problem is that when a calculation loop for a discrete expression calculation is automatically generated for each small area divided based on information on the shape and boundary conditions of the area, the index range in each direction of the small area index range is changed. Multiple calculation loops for the number of dimensions are generated to control the calculation order. Therefore, the maximum loop length in the multiple calculation loops is the number of divisions of one side of the small area. Therefore, in the numerical calculation program generated in this way, the loop length that can be vectorized and processed is shorter than the scale of the problem, which hinders the speeding up of the processing by the vector computer.

The second problem is that, from information of discrete equations corresponding to each small area, a program that integrates small areas having the same discrete equation and performs calculation can be automatically generated. Since a program for a multiple calculation loop is generated, it is necessary to match the number of dimensions of the index for controlling the calculation order and the index range in each direction. In this case, integration is permitted only when the area shape after integration is rectangular. Therefore, if the region shape is not rectangular, a plurality of sets of the same discrete calculation loop are generated, and the generated program code is redundant.

The present invention has been made to solve the above problems.

An object of the present invention is to unify a discrete calculation loop,
It is an object of the present invention to provide a program code generating method for increasing a calculation loop length and generating an efficient program code for a vector computer.

[Means for solving the problem]

In order to achieve the above object, a program code generating method according to the present invention comprises: a shape of an area for analyzing a physical phenomenon; grid division information for setting the grid points to be analyzed in the area by dividing the area into a grid; Input the partial differential equation expressing the physical phenomena in the area and the area analysis conditions consisting of the boundary conditions of the area, divide the area into small areas with the same area analysis conditions, and Create an index table that specifies the order of calculation of the discrete expression, the program code of a single calculation loop that uses the index table to control the flow of the calculation order that performs the calculation of the discrete expression for analyzing each grid point It is characterized by generating.

[Action]

According to this, in the method of generating a program code, first, the shape of an area for analyzing a physical phenomenon, grid division information for dividing the area into a grid and setting grid points to be analyzed points in the area, Input the partial differential equations expressing the physical phenomena of and the region analysis conditions consisting of the boundary conditions of the regions. According to the area analysis conditions, the area to be analyzed is divided into small areas having the same analysis conditions. Next, an index table is created that specifies the calculation order of the discrete equations for the grid points of the divided small areas. Then, a program code of a single calculation loop for controlling a flow of a calculation order for performing a calculation of a discrete expression for performing an analysis on each grid point by using the index table is generated. Thus, in the discrete expression calculation loop, the order of calculating the discrete expression of each small area is always instructed by referring to the index table, so that the calculation order of each grid point is independent of the number of dimensions of the small area. Can control. For this reason, a discrete calculation loop can always be generated by a single calculation loop. Therefore, the generated program code has a longer calculation loop length, less control for changing the calculation order, and is a program code for numerical calculation suitable for a vector computer by pipeline control.

That is, what is characteristic here is that, for the analysis target area, an index range in each direction of the region is determined, and an index table indicating the calculation order for calculating the discrete expression of the area is created. To generate a program code for performing calculations in a single calculation loop by controlling the calculation order.

In addition, since the index table is created for each of the divided small areas including the grid points having the same discrete formula, a plurality of small areas having the same discrete formula are collected by integrating the index tables, and a single calculation loop is formed. Program code can be generated. As a result, the length of the calculation loop can be further increased, and a program code suitable for a vector computer can be generated by pipeline control.

As described above, since the index table indicating the order of calculating the discrete expressions of the small area is created and the calculation loop for controlling the calculation order of the discrete expressions using the index table is generated, the index does not depend on the number of dimensions of the small area. Thus, it is possible to always generate a program code in a single calculation loop. Thereby, the loop length of the calculation loop is conventionally the number of divisions of the side of the region, but becomes the number of grid points of the small region, so the loop length becomes longer, and the control of the change of the calculation order is reduced,
The execution speed of a program generated on a vector computer by pipeline control can be increased.

In addition, in the small area where the same discrete expression holds, the index table of each small area is put together to create one index table, and a discrete loop is used to generate a program code. A program code that does not make a program redundant can be generated without generating a plurality of calculation loops.

〔Example〕

Hereinafter, an embodiment of the present invention will be specifically described with reference to the drawings.

FIG. 1 is a block diagram showing a configuration of a main part of a program code generation system for implementing a program code generation method according to an embodiment of the present invention.

In FIG. 1, reference numeral 10 denotes a program code generation system. A program code generation system 10 receives input data 11 of an area analysis condition including an area shape of an analysis area, grid division information, partial differential equations, and boundary conditions, generates a discrete equation calculation program 12, and outputs the generated program code. It is an automatic generation processing system.

The program code generation system 10 includes a region division processing unit.
13, an index table creation processing unit 14, and a program generation processing unit 18. The area division processing unit 13
The region is divided into small regions that can be analyzed by the same discrete formula based on the region shape of the analysis region, the lattice division information, the partial differential equation, and the region analysis conditions of the boundary condition, and the region division information 19 is generated.
In the index table creation processing unit 14, the area division information 19
, An index table indicating the order in which the discrete expression of each grid point is calculated is generated. Also, the program generation processing unit 18
Generates a discrete expression calculation program 12 from the index table generated by the index table creation processing unit 14 and the input data 11 of the area analysis conditions.

The index table generation processing unit 14 includes: an index table generation processing unit 15 that generates an index table for each small area; a determination processing unit 16 that detects a plurality of small areas having the same discrete formula; When the small area is detected, an index table integration processing unit 17 that combines the index tables of the small area into one is configured.

The operation of the processing unit configured as described above will be described using a specific example.

2a, 2b, and 2c are diagrams showing examples of input data of the area analysis condition.

For the area where the physical phenomenon to be analyzed is occurring, input the area shape, grid division information indicating how to divide the area, partial differential equations representing the physical phenomenon, and boundary conditions. Input as data. In this example, as shown in FIG. 2a, a region 20 where a fluid is flowing from left to right corresponds to a region shape 21 of a rectangular ABCD to be analyzed.
And grid division information such as grid points 24 and grid lines 25 are input. The discrete partial differential equations representing the physical phenomena in this region include, for example, the velocity u of the fluid in the x direction, the velocity v of the fluid in the y direction, the pressure P, and the Reynolds number Re. Navier showing the motion phenomenon of viscous fluid
The Stokes equation is input as the partial differential equation 22 of the input data, as shown in FIG. 2b. The boundary condition in this case is, as shown in FIG.
In AB, a boundary condition between the velocity u = 1.0 in the x direction of the fluid and the velocity v = 0.0 in the y direction is given. On the sides AD and BC, the velocities u = 0.0 and y in the x direction are given, respectively.
A boundary condition with a velocity in the direction v = 0.0 is given. Area ABC
On the side DC of D, respective changes in the velocity u in the x direction u and the velocity v in the y direction (∂u / ∂x = 0.0, ∂v / ∂
x = 0.0).

Based on the input data of the area analysis conditions given in this way, the area division processing unit 13 divides the area into small areas that can be analyzed by the same discrete formula, and generates area division information.

FIG. 3 is a diagram showing a division example in which the analysis target region is divided into small regions according to region analysis conditions. The analysis target area here is divided into nine small areas as shown in FIG. That is, the small area 3 that captures the boundary conditions on the boundary AB
9, A small area 33 that captures the boundary conditions on the boundary BC, a small area 35 that captures the boundary conditions on the boundary DC, a small area 37 that captures the boundary conditions on the boundary AD, and a boundary condition on the boundary between the boundary AB and the boundary BC. The small area 32 to take in, the small area 34 to take in the boundary condition on the boundary between the boundary BC and the boundary CD, the small area 36 to take in the boundary condition on the boundary between the boundary CD and the boundary AD, the boundary AB and the boundary A
This is divided into small areas 38 for taking in boundary conditions on the boundary of D and small areas 31 not taking in any boundary conditions. As the area division information 19, the index range of these nine small areas and boundary information indicating in which direction the boundaries are in contact are stored as table data.

If the table data of the area division information 19 has an index range in two or more directions for each small area, an index table that stores a combination of indexes in each direction of all grid points existing in the small area is Generated.

FIG. 4 is a diagram showing an example of an index table generated corresponding to the small area 31 (shaded area) not taking in the boundary conditions shown in FIG. Since this small area 31 has index ranges in two directions of the x direction and the y direction, the fourth area
As shown in the figure, an index table 41 is stored in which the combinations of the respective directional indexes of all the grid points existing in both areas are stored by the x-direction index I and the y-direction index J. Here, this index table
The number of combinations of each direction index stored in 41 is set as the loop range of the discrete calculation loop, and the program code of the single discrete calculation loop is automatically generated.

FIG. 5 is a diagram showing an example of a program code generated as a single discrete-type calculation loop for controlling the calculation order using an index table. In the program code of FIG. 5, reference numeral 51 denotes a program code of a DO loop constituting a discrete formula calculation loop, and 52 denotes a process of detecting an index in each direction from an index table ITBL storing index data in each direction. This is the program code that performs 53 is a program code of a discrete equation of the partial differential equation. A program code generated by a well-known automatic program generation method is inserted into a part of the program code 53 of the discrete equation of the partial differential equation. This content is not shown here in detail because it is not particularly relevant here.

As shown in this example, the area to be analyzed is divided into small areas having the same boundary conditions, and an index indicating the index of each grid point in each direction in the small area for each small area A table is generated, and a program code that forms a single calculation loop is automatically generated using the index table. This makes it possible to generate a program code as a single calculation loop for a grid point with a discrete calculation loop for a small area having an index range in a plurality of directions, thereby reducing the processing for changing the calculation order and reducing the length of one calculation loop. Can be lengthened. Therefore, the execution performance of the generation program on the vector computer is improved.

Next, an application example in a case where the region shape to be analyzed is not a rectangular shape will be described.

By the way, the shape of the region where the physical phenomenon occurs is the sixth shape.
In the case of the shape as shown in the figure, the boundary condition of the region is given on each side AB, BC, CD, DE, EF, AF of the region. In this case, the analysis area 60 is divided into 14 small areas (61, 62,..., 74) as shown in FIG. 7 in the processing of the area division processing unit 13 based on the boundary conditions of each side of the area. You. The division information of each of the divided regions is the region division information.
As 19, table data is created and stored. this
Of the 14 small areas, small area 61 and small area 62 (shaded area)
Has an index range in both the x-direction and the y-direction.
For the index table 81 as shown in FIG. 8a.
Is created, and for the small area 61, an index table 82 as shown in FIG. 8b is created. Here, the boundary condition of the boundary where the two small regions 61 and 62 are in contact is determined from each region information of the table data of each small region of the region division information 19. In this case, since the small area 61 and the small area 62 are determined not to be in contact with any of the boundaries of the sides AB, BC, CD, DE, EF, and AF, the corresponding discrete expressions are determined to be equal, and the index table is determined. The integration processing 16 is performed. In the index table integration process 16, the index table created for the small area 62 and the small area 61 is created.
The index table 81 and the index table 82 are integrated to generate one integrated index table 91 as shown in FIG. Then, by using the integrated index table 91, for example, a program code by one single calculation loop as shown in FIG. 5 is generated.

In this way, it is possible to integrate the index tables of a plurality of small areas having the same discrete formula and to combine the discrete formula calculation loop into one calculation loop. in this case,
An index table indicating the order of calculating the discrete expressions of the small area is summarized, and a calculation loop for controlling the calculation order of the discrete expressions is generated using the combined index table, so that it does not depend on the number of dimensions of the small area, It is always possible to generate a program code in a single calculation loop. Thereby, the loop length of the calculation loop is conventionally the number of divisions of the side of the region, but becomes the number of grid points of the small region, so the loop length becomes longer, and the control of the change of the calculation order is reduced. In addition, the execution speed of a program generated on a vector computer by pipeline control can be increased. In addition, a small area in which the same discrete expression is satisfied by the boundary condition is obtained by combining the index tables of the respective small areas into one.
Since one index table is created, and one discrete loop is used to generate a program code, a program code that does not make the program redundant can be generated without generating a plurality of calculation loops of the same discrete formula.

As described above, the present invention has been specifically described based on the embodiments. However, it is needless to say that the present invention is not limited to the above-described embodiments, and can be variously modified without departing from the gist thereof.

〔The invention's effect〕

As described above, according to the present invention, for the analysis target region, the index range in each direction of the region is obtained,
An index table indicating the calculation order for calculating the discrete expression of the region is created, and a program code for performing the calculation in a single calculation loop is generated by controlling the calculation order using the index table. Since the index table is created for each of the divided small regions including the grid points having the same discrete formula, a plurality of small regions having the same discrete formula are integrated by integrating the index tables, and the program code is formed as a single calculation loop. Generate Thereby, the calculation loop length can be further increased,
Program code suitable for a vector computer can be generated by pipeline control.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of a main part of a program code generation system for implementing a program code generation method according to an embodiment of the present invention. FIGS. 2a, 2b, and 2c show area analysis conditions. FIG. 3 is a diagram showing an example of division of an analysis target region into small regions according to region analysis conditions, and FIG. 4 is a diagram showing a small region which does not take in the boundary conditions shown in FIG. FIG. 5 is a diagram showing an example of an index table generated corresponding to a region (hatched portion). FIG. 5 is an example of a program code generated as a single discrete formula calculation loop for controlling the calculation order using the index table. FIG. 6 is a diagram showing another example of a region shape input as a region analysis condition. FIG. 7 is a diagram showing an example of dividing the region shape of FIG. 6 into small regions according to the region analysis condition. Figure, Figure 8a and Figure 8 Fig. b is a diagram showing an example of an index table generated for each of the two small regions in Fig. 7, and Fig. 9 is a diagram showing an example of an integrated index table in which the small regions are integrated. . In the figure, 10: a program code generation system, 11: input data, 12: a discrete expression calculation program, 13: a region division processing unit, 14: an index table creation processing unit, 15 ...
... Index table generation processing unit, 16 ... Judgment processing unit,
17: Index table integration processing unit, 18: Program generation processing unit, 19: Area division information, 31 to 39, 61 to 74
… Divided small area, 41,81,82 …… index table,
91 …… Integrated index table.

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-62-208166 (JP, A) JP-A-62-212771 (JP, A) JP-A-1-145723 (JP, A) JP-A-1 307826 (JP, A) "Communications of ACM" Vol. 29, No. 12 (1986-12) p. 1196-1198 (58) Field surveyed (Int. Cl. ⁷ , DB name) G06F 9/06, 9/44

Claims (2)

(57) [Claims] 1. A shape of an area for analyzing a physical phenomenon, grid division information for dividing the area into a grid and setting grid points to be analyzed points in the area, and a partial differential equation for expressing a physical phenomenon in the area And an input means for inputting an area analysis condition comprising a boundary condition of the area, and an index table storage means, and a program code for performing calculation of a discrete expression for analyzing grid points in the area for all grid points A program code generation method in a program code generation system that generates a region, based on a region analysis condition input by the input unit, dividing the region into small regions having the same region analysis condition; Generating an index table for designating the calculation order of the discrete expression for the grid points of the divided small areas, and storing the index table in the index table storage means; And a third step of detecting whether or not there is another small area in which calculation for analyzing a grid point can be performed with the same discrete expression for each divided small area. A fourth step of ordering each index table in the index table storage means for the small area to be calculated and integrating them into one index table; and a discrete expression for analyzing each grid point using the integrated index table. Generating a program code for a single calculation loop that controls the flow of the calculation sequence for calculating
And a program code generating method. 2. A shape of an area for analyzing a physical phenomenon, grid division information for setting the grid point to be an analysis point in the area by dividing the area into a grid, and a partial differential equation for expressing a physical phenomenon in the area. And a program code generation system for generating a program code for performing calculation of a discrete expression for analyzing grid points in the area for all grid points based on an area analysis condition including a boundary condition of the area. , An area dividing means for dividing an area into small areas having the same area analysis condition, and an index table in a calculation order for calculating an index range in each direction of the divided small area and calculating a discrete expression for analyzing a grid point Index table generation processing means, and a single calculation loop for controlling a flow of a calculation order for calculating a discrete expression for performing an analysis for each grid point by using the index table A program generation processing means for generating a program code, wherein the index table generation processing means further includes means for detecting a small area in which the discrete expression to be analyzed is the same, and a small area in which the discrete expression to be analyzed is the same. A program code generation system, comprising: index integrating means for integrating index tables.