JPH03266084A - Back search processing method for wiring processing - Google Patents

Abstract

PURPOSE:To accelerate a processing by executing back search while limiting a back search direction to a certain fixed direction by utilizing a fact that a route is generally a little bent in wiring. CONSTITUTION:Respective grids 15 to be the units of a wiring area 14 are allocated to respective processors PE and under the control of a controller 10, the advancing direction of the back search is set in a processing process 11 by investigating 4 directions when the starting the back search and changing the advancing direction of the back search. In a processing process 12, the back search is executed while limiting the advancing direction of the back search to the fixed direction, while is currently set, until the advancing direction is changed. Namely, since the search is executed concerning the 4 directions only when wiring is bent, and the search is executed in a linear part concerning the same direction as the former search direction, processing time can be shortened.

Description

[Detailed description of the invention]

[Summary] Search for printed circuit boards and LSI wiring patterns. The present invention relates to a backward search processing method for wiring processing that is executed at high speed using a SIMD type parallel computer composed of a large number of processors, and the purpose is to speed up the backward search processing in the maze method. When searching for a path in the reverse direction from the target to the source determined by the forward search by allocating each grid, which is a unit of the wiring area, to each processor in the processor array, the following information is used: When the direction changes, check the four directions and set the direction of backward search. The configuration is such that the backward search is continued by limiting the traveling direction of the backward search to the currently set constant direction until the traveling direction changes. [Industrial Field of Application] The present invention relates to a backward search processing method for wiring processing that searches for wiring patterns of printed circuit boards or LSIs at high speed using a SIMD type parallel computer composed of a large number of processors. As the density of wiring targets has increased, the scale of data to be handled has increased, so there is a demand for faster wiring processing. On the other hand, attempts have been made to speed up the processing through parallel processing. In an attempt to parallelize the maze method, which is one of the basic algorithms for wiring processing, a grid, which is a unit of the wiring area, is assigned to one processor (hereinafter referred to as PE) in a processor array connected in a grid pattern. , a method has been devised in which PEs on the wavefront in forward search operate in parallel. However, regarding the backward search of the maze method. Since a sequential processing method is still used, increasing the speed of the backward search portion is an important issue for increasing the speed of the wiring system as a whole. [Prior art] Figure 4 is a diagram for explaining backward search in the maze method.
FIG. 5 is an explanatory diagram of the backward search operation, and FIG. 6 shows the processing flow of the prior art. When performing the maze method, during the backward search portion, the direction of the arrow marked during the forward search is used as a clue from the PE in charge of the target during the forward search. The route is sequentially traced until reaching the PR in charge of the source (. An example of backward search will be explained in detail according to Figure 4. The label is shown, and S is the source. T is the target grid. The shaded area is the grid on the wiring prohibited area. The backward search starts from the target T at the position (8-chi). , since the direction in which the wave propagated to the target T is to the right, the next grid to the right (9-chi) will be the grid to be traced next. Next, referring to the grid attached to (9-chi), You can see that the arrow points to the right, so 5 trace the grid further to the right. In this way, until you reach source S 1
By sequentially tracing each grid one by one, as shown in Figure 4 (
A route as shown in b) can be obtained. The backward search as described above is performed in a first-order manner. It is assumed that the state of the wiring area is as shown in FIG. 5(a). One frame corresponds to the IPE of a processor array in a parallel computer, and for this IPE, 1
A grid is assigned. T represents the target PE, S represents the source PE, and all PEs are as shown in FIG. 5(b). It has a source flag (a), an arrow flag (b), and a search flag (C). Here, the source flag is set to "1" for the PE in charge of the source, and "0" for the other PEs. The arrow flag consists of four focus flags in the up, down, left and right directions, with the direction the arrow is pointing at being "1" and the remaining three directions being 0°. The search flag is a flag that indicates that the PE is on the route when searching backwards from the target to the source.The initial value is ``1'' only for the PE that indicates the target, and 0 for all other PEs. be. Eventually, the search flags of all PHs on the route become 1'. As shown in FIG. 5(c), the PEI search flag is “1”.
”, and when the arrow flag points to the left, if one backward search is performed, as shown in Figure 5 (d),
The search flag of PE2 on the left changes from "0" to "1". This process is conventionally performed as shown in Figure 6. Since it is executed on an IMD type parallel computer, all PEs , the processes shown in Figure 6 are executed simultaneously in parallel. First, as described above, the flags of each PE are initialized. Then, in order to start backward search from the target, the search flag of the target PE is set to "1". ". Next, in process 2, the search flag and the up arrow flag are ANDed, and this is set as the temporal flag of the upper PE. Then, the temporal flag of the upper PE and the upper P
It is logically ORed with the search flag of E, and it is set as the new search flag of the upper PE. Flag setting processing is similarly performed for the lower side, left side, and right side. After this, it is determined whether the search flag of the source PH has become "1", and if it has become "1". End backward search. If “O”, return to processing 1
Continue the next backward search. [Problems to be Solved by the Invention] In the conventional technology, each time a backward search is performed, it is necessary to determine the direction to proceed in the four directions (up, down, left, and right) in which the backward search can be performed, so the time required for this determination is reduced. A problem arose in that the materials were wasted. For example, the process (■) shown in Figure 6 requires 8 operations, and the process (■) requires 1 operation.
Since the calculations are performed twice, the time required to search backward by one grid is 9 calculation times. Therefore, it takes 9n calculation times to search backward by n grids. The present invention aims to solve the above-mentioned problems and aims to speed up the processing of backward search in the maze method using a parallel computer. (Means for Solving the Problems) Fig. 1 is an explanatory diagram of the principle of the present invention. In Fig. 1, 10 is a controller of a parallel computer;
1 and 12 are processing steps according to the present invention. 13 is a professional sensor array, 14 is a printed circuit board or LSI
15 represents a grid serving as a unit of the wiring area. The present invention includes a processor array 13 in which a plurality of processors PE are connected in a lattice, and a controller 10 that controls the processor array 13.
It is executed using a SIMD type parallel computer equipped with. Each processor PE in the processor array 13 is allocated with each grid 15 which is a unit of the wiring area 14.
A reverse route from the target determined by the forward search to the source is searched under the control of the controller 10. This backward search is processed in a first-order manner. In process step 11, at the start of the backward search and when the backward search direction changes, four directions are examined to set the backward search direction. In process step 12, the backward search is performed with the traveling direction of the backward search limited to the currently set constant direction until the traveling direction changes. If the direction of backward search does not change, process step 12 is repeated; if the direction of movement has changed, process step 11 is repeated. Set the new direction of travel again. After performing the above processing and searching for a route to the source, the processing ends. [Operation] The present invention takes advantage of the fact that once the backward search direction is determined, it is difficult to change, that is, there are generally few bends in the route in wiring, and the backward search direction is fixed to a certain fixed direction and the backward search is performed. This is intended to increase speed. In the conventional technology, a search is performed in four directions every time the grid is searched backwards, whereas in the present invention, a search in four directions is performed only when the direction of travel changes, that is, only when the wiring is bent. Since the search is performed in the same direction as the previous search direction in the straight line portion of the wiring, the processing time can be significantly reduced. (Embodiment) Fig. 2 shows an example of a hardware configuration used in an embodiment of the present invention, and Fig. 3 shows a processing flow of an embodiment of the present invention. The processor (PE) array 13 in FIG.
Controlled by I signal. Each PE. It has an internal memory that holds the data necessary for wiring processing, and an arithmetic unit that performs calculations on that data.
Executes the same instruction given by the m signal in parallel. The collection operation circuit 20 is a circuit that performs logical operations such as AND, OR, ADD, MIN, and MAX on the output values of all PEs on the PE array 13, and is disclosed in Japanese Patent Application No. 63-2, for example.
This circuit is as shown in No. 34545. The collection calculation circuit 20 obtains calculation results for the output values of all PEs, thereby making it possible to execute calculations that require synchronization and consensus among all PEs. The global memory 21 is a memory that holds values resulting from the collection calculation circuit 20 and values to be broadcast to all PEs. In implementing this embodiment, the second
The IPE of the PE array 13 shown in the figure is assigned to one grid, which is a unit of wiring area. Then, the controller 10 performs a backward search based on the results obtained in the forward search using a PE control signal generated according to a program. In backward search, you follow the direction of the arrow from the target in the opposite direction and search until you reach the source.
PE when performing backward search. As in the conventional case, there are three types of flags: a source flag (a), an arrow flag (b), and a search flag (C) shown in FIG. 5(b). Then, each PE has controller 1
Under the control of 0, backward search is performed according to the flowchart shown in FIG. Initialize the flag, set the target PE search flag to “1”
The process of setting the angle to °° is the same as in the prior art. Here, when starting the backward search, first the traveling direction is set. If the direction of travel is undetermined, it is necessary to try searching backward in four directions. However, once the direction of travel is determined, processes 1 to 2 are repeated only in that direction until the direction of travel changes. If it is detected in the process (2) that the traveling direction has changed, then in the process (2) it is determined whether the search flag of the source PH has become "1". If it is not "l", return to the process. Check the direction of backward search in four directions. Since one direction is always determined, set the direction of travel and repeat the process (1) to (2) in the same way. As a result of the process ■, the search flag of the source PH is set to “1”.
If it is found that the path from the target to the source has been determined, the backward search ends. Next, a specific example of operation by the process shown in FIG. 3 will be explained. It is assumed that the values held by each PE are as shown in FIG. 5(a). In the process (■) shown in FIG. 3, the direction of travel is set. This is done by logically ANDing the search flag and each arrow flag (four directions, up, down, left and right), and then using the logical sum in the collection calculation circuit 20 shown in FIG. It is found by taking . This process is performed for 4 directions, and the collection calculation result is °“1
” is determined as the backward search direction. Fifth
In the case of figure (a), the first direction is to the left. Next, in the process (2) shown in FIG. 3, the search flag and the left arrow flag are ANDed. As a result, the left temporal flag is set to “1°.” In process ①, it is determined whether the direction of travel has changed. This is determined by inverting the left search flag (the value is “
1”) and a temporal flag for the direction of travel (value is “1゛)”
The determination is made based on the logical AND value (the value is "'1") obtained by calculating the logical sum in the collection arithmetic circuit 20. If this value is "1", it means that the direction has not changed, and if it is "0", it means that the direction has changed. If the traveling direction has not changed, proceed to 1 process ■. In this process ■ , the search flag (value is "1") and the left temporal flag (value is "1") are logically summed to obtain a new search flag value (value is "1°"). The above process is repeated until the search flag of the source PE becomes "1". The time required to search backwards by 21 grids to perform four operations in process ■, one operation in process ■, one operation in process ■, one operation in process ■, and one operation in process ■ shown in Figure 3. In the case of this embodiment, it takes 8 computation times, including process (2). n
In order to search backwards from the grid, if the direction of movement does not change, it is sufficient to repeat Process 1, Process 2, and Process 2, so the total time excluding Process 2 is (4+3n+1) calculation time. The calculation time according to the conventional technology as shown in Fig. 6 is 9n.
Therefore, when compared with this, it can be seen that the speed can be significantly increased.

【Effect of the invention】

As explained above, according to the present invention, in the backward search of the maze method, the fact that there are few bent wires is utilized, and -
By repeatedly processing the determined arrow direction several times, it is possible to significantly reduce the number of times the four search directions are examined, and speeding up is possible. FIG. 1 is a diagram explaining the principle of the present invention. FIG. 2 shows an example of a hardware configuration used in an embodiment of the present invention. FIG. 3 is a processing flow of an embodiment of the present invention. FIG. 4 is a diagram for explaining backward search in the maze method. FIG. 5 is an explanatory diagram of backward search operation. FIG. 6 shows a processing flow of the prior art. In the figure, 10 is a controller, 11 and 12 are processing steps, 13 is a processor array, 14 is a wiring area, 15 is a grid, and PE is a processor.

Claims (1)

[Claims] A processor array (13) in which a plurality of processors (PEs) are connected in a grid, and a controller (13) that controls the processor array (13).
In the backward search processing method for wiring processing using the maze method, which is executed using a SIMD type parallel computer equipped with It is assigned to each processor (PE), and when searching for a reverse route from the target determined by the forward search to the source, it examines four directions at the start of the backward search and when the direction of backward search changes. to set the direction of backward search (
11), Continue backward search by limiting the direction of travel to the currently set direction until the direction of travel changes (12)
) A backward search processing method for wiring processing.