RU2597480C1 - METHOD FOR PARALLEL PROCESSING OF THREE-DIMENSIONAL BODIES IN b-rep REPRESENTATION - Google Patents

Abstract

FIELD: computer engineering.

SUBSTANCE: invention relates to processing of three-dimensional bodies, particularly processing three-dimensional bodies using parallel computations. Method of processing three-dimensional bodies in B-Rep representation using parallel computations, the structure of B-Rep representation is converted into a graph of dependence between faces of the body, preliminary forming a dependence between the faces on the basis of information on connectivity of the faces through child elements, herewith dependences arrays are formed in the memory, at least, of identifiers of the faces and their vertices, as well as identifiers of vertices and faces each vertex belongs to; after that, if necessary, the graph is optimized excluding back-up dependences, and parallel computations are applied to processing of the faces in the obtained graph, processing each next face after processing of all the faces preceding it in the graph.

EFFECT: technical result is reduced time of processing three-dimensional bodies in B-Rep representation.

1 cl, 4 tbl, 7 dwg

Description

Technical field

The present invention relates to the field of processing three-dimensional bodies in a B-Rep representation, namely, processing three-dimensional bodies using parallel computing.

State of the art

The technical solutions known for the "Parameterizing a 3d modeled object for tessellation" [Patent Application No. US2014184599, z. December 28, 2012] and "Tessellation of a Parameterized 3D Modeled Object", patent application No. US2014184598, s. December 28, 2012]. Both technical solutions describe methods for processing three-dimensional models to solve the problems of meshing (triangulation).

Moreover, to solve the triangulation problem, there is no need to transform objects inside the body in the B-Rep representation, for example, of geometric objects (surfaces, curves, points) used by the corresponding topological objects (faces, edges, vertices). Accordingly, these technical solutions cannot be applied to cases where such transformations to the body in the B-Rep representation must be applied. The claimed invention addresses such cases.

SUMMARY OF THE INVENTION

The task to which the invention is directed is to accelerate the processing of three-dimensional bodies in a B-Rep representation through the use of parallel computing. B-Rep representation is described using the structure of elements connected among themselves (vertices, edges, faces, etc.). Due to the connected elements (for example, in a closed body, each edge is divided by two faces, and each vertex is separated by at least two edges), the effective use of parallel computing is very difficult (due to the strong dependence on the data).

This problem is solved by processing three-dimensional bodies in the B-Rep representation using parallel computing. The structure of B-Rep representations is transformed into a dependency graph between the faces of the body, previously forming arrays of dependencies between the faces based on information about the connectivity of the faces through child elements (vertices or edges of the body in the B-Rep representation). The relationship “dependencies” between the faces Fi and Fj denotes the requirement that the face Fj can only be processed after the face Fi. In this case, dependency arrays are formed in the memory of the computing device using at least the identifiers of the faces and their vertices, as well as the identifiers of the vertices and faces to which each vertex belongs. Then, if necessary, the graph is optimized, eliminating duplicate dependencies in it, and parallel calculations are applied to the processing of faces in the resulting graph, processing each subsequent face after processing all the faces preceding it in the graph.

Also, the graph of dependencies between the faces of the body can be represented in the form of tables.

Parallel computing is carried out with the simultaneous use of at least two computing devices. To increase the efficiency of the method in the presence of free computing power, the corresponding computing device analyzes the presence of faces that are not involved in processing at a given time, and upon receiving a positive response, begins processing them. A computing device is understood to mean any device capable of performing calculations. Examples of computing devices include a computer’s central processor, a central processor core, a hyper-thread of a processor’s core, cores and hyper-threads of a co-processor or graphics accelerator, nodes of a computing cluster, etc. At the same time, parallel computing can be used at any level of computer technology, for example, using vector processor instructions, multi-threaded programming, programming on systems with distributed memory (clusters), on systems using co-processors and graphics accelerators, as well as using arbitrary combinations of the above methods.

Memory refers to any kind of physical device or medium for storing data on a computing device, including registers and processor cache, dynamic random access memory (DRAM), random access static memory (SRAM), any kind of storage device (for example, HDD and SSD )

Technically, the result provided by the given set of features is to reduce the processing time of three-dimensional bodies in the B-Rep representation.

The described method can be used to convert three-dimensional bodies between different formats, restore, correct, and other changes to geometric data in three-dimensional bodies. It is also obvious to experts in the subject field that it can be applied to other use cases for processing three-dimensional bodies, for example, for solving visualization problems, building grids, etc.

Brief Description of the Drawings

The invention is illustrated by drawings.

Fig. 1 - A body having a B-Rep representation (11). B-Rep (Boundary representation) - a way of representing a body using a data structure containing information about the boundaries of the volume (vertices, edges, faces) and their connection with each other [http://wap.ism-06-2.ru/shpora .php? razdel = 5 & id = 402 &].

Fig. 2 and 3 - B-Rep representation and its internal structure. In fig. 2 and 3, the following notation is used:

V1, V2 - peaks;

E1, E2, E3 - ribs;

F1, F2, F3 - faces.

Fig. 4 is a model of a rectangular parallelepiped in a B-Rep representation. In fig. 4 the following notation is used:

0, 1, ..., 7 - peaks;

F0, F1, ..., F5 are faces.

Fig. 5 and 6 - dependency graphs of the faces of the body shown in Fig. 4, complete and optimized.

Fig. 7 is an example of a fragment of a complex dependency graph of the faces of a body.

DETAILED DESCRIPTION OF THE INVENTION

Body elements in the B-Rep representation (Fig. 2 and 3) are understood to mean faces, edges, and vertices. Elements of the B-Rep representation are ordered according to a hierarchical principle (Fig. 3), where the senior elements are composed of minor elements. For example, each face in the B-Rep view consists of edges, and each edge consists of vertices. Depending on the type of the three-dimensional body, the hierarchical principle can be applied to other elements of the B-Rep representation, such as surfaces and curves.

As an example of a body illustrating the application of the method, a rectangular parallelepiped was chosen (see Fig. 4). For the body in the B-Rep representation (Fig. 4), we create a set of elements that make up this representation and are arranged in a hierarchical manner. For example, for a body in the B-Rep representation (Fig. 4) we create an array of vertices (V0-V7), an array of faces (F0-F5) that make up this B-Rep representation (Fig. 4). The number and content of these arrays depends on the particular three-dimensional body.

Then at the same time we create 2 tables:

1. A table containing the numbers of all faces and the numbers of its vertices (see Table 1);

2. A table containing the numbers of all vertices and the numbers of faces to which this vertex belongs (see Table 2).

F0 V0 V1 V2 V3 F1 V4 V1 V5 V2 F2 V6 V5 V3 V2 F3 V7 V0 V4 V1 F4 V7 V6 V0 V3 F5 V7 V6 V5 V4

Table 1. Correspondence of the face and the numbers of its vertices

V0 F0 F3 F4 V1 F0 F1 F3 V2 F0 F1 F2 V3 F0 F2 F4 V4 F1 F3 F5 V5 F1 F2 F5 V6 F2 F4 F5 V7 F3 F4 F5

Table 2. Correspondence of the vertex and the number of faces to which this vertex belongs

Then, using the data from the tables (Table 1 and 2), we form a table of dependencies between the faces (Table 3). This table expresses the relationship between the “previous” face (left column) and the “subsequent” faces (right column). The terms “previous” and “subsequent” indicate that any “subsequent” face can be processed only after processing all “previous” faces.

F0 F4 F3 F2 F1 F1 F5 F3 F2 F2 F5 F4 F3 F5 F4 F4 F5 F5

Table 3. Table of dependencies between faces

Next, for the body in the B-Rep representation, we form a graph of the dependencies of the faces, in which the edges indicate the relationship of the relationship "previous" - "subsequent" face. In Fig. 5 it is clearly seen that such a graph contains redundant edges, for example, between the faces F0 and F4 there are three paths connecting them: F0-F2-F4, F0-F3-F4 and F0-F4. This is due to the fact that some faces are simultaneously associated with several “previous” faces. This redundancy can increase the processing time of the body, as additional checks will be required for the condition that all the “previous” faces have been processed before the start of processing each “subsequent” face. Accordingly, in order to speed up the processing process, it is possible to optimize the graph, that is, reduce the dependence of the face on the “previous” faces to a minimum. We optimize the graph, excluding for each face those "previous" faces that are "previous" for other "previous" faces (see Table 4).

F0 F1 F1 F3 F2 F2 F4 F3 F4 F4 F5 F5

Table 4. Table of dependencies between faces

In the example above, for F4, exclude F0 as “previous”, because F0 is “preceding” for F2 and F3, which themselves are “preceding” to F4. Based on the table, we form an optimized graph (Fig. 6). Then we begin processing in accordance with the optimized graph. After the processing of F1 is completed and when the “subsequent” faces F2 and F3 are processed (Fig. 6), it becomes possible to use parallel computations due to the independence of the processing of F2 and F3 from each other. To increase the efficiency of parallel computing, a method called “theft of tasks” (from the English “work stealing”) can be used.

This principle can be explained by the following example. For large complex B-Rep bodies, branching in the dependency graph (when many “follow-up” depends on one “previous” face) can occur arbitrarily. In fig. Figure 7 shows an example of a more complex dependency graph with many branches. The mandatory use of parallelism for each such branching would create the creation of an excessive number of slaves (for example, the number of computational threads is greater than the number of cores in the central processor). This would lead to a slowdown, not an acceleration of computation, which is obviously not effective. Instead of such “forced” parallelism, the method of “theft of tasks” is used. The concept of "theft of tasks" is that each WU, when processing its existing elements (for example, faces) is completed, tries to find other unprocessed elements. In this case, the taking away (“theft”) of elements from another computing device, on which it has already begun to work, can take place. In the claimed method, the implementation of this concept consists in the fact that when processing elements of the B-Rep representation (faces), each working WU analyzes the presence of elements that are not involved in processing at a given time, and upon receiving a positive response, it starts processing the corresponding elements.

Depending on the features of the B-Rep representation or computational algorithms, the implementation of the method may be based on information about the connectivity of the faces by edges rather than vertices.

Thus, this invention accelerates the process of processing three-dimensional bodies in the B-Rep representation through the use of parallel information processing.

Claims (1)

A method for processing three-dimensional bodies in a B-Rep representation using parallel computing, characterized in that the B-Rep structure is converted into a dependency graph between the faces of the body, previously forming dependency arrays between the faces based on information about the connectivity of the faces through the child elements, while the arrays dependencies are formed in memory at least from the identifiers of the faces and the vertices belonging to them, as well as the identifiers of the vertices and faces to which each vertex belongs, after which, if necessary, Qdim optimized graph, eliminating duplicate depending therefrom, and parallel computation processing applied to the faces of the resulting graph, each successive facet processing after processing all facets preceding it in the graph.

Images