53 results about "Vertex number" patented technology

The invention discloses a design method for a heterogeneous reconfigurable diagram calculation accelerator system on the basis of an FPGA (Field Programmable Gate Array). A whole accelerator system comprises a PC (Personal Computer) heterogeneous module and an FPGA heterogeneous module. The design method comprises the following specific steps that: firstly, loading a driving module required for starting the FPGA, and starting equipment, including the PCIe (Peripheral Component Interface Express) DMA (Direct Memory Access) and the like of the FPGA; according to the vertex number and the edge number of diagram data which needs to be processed, selecting a diagraph calculation accelerator engine; preprocessing the diagram data after the accelerator engine is selected; transmitting the processed diagram data to the onboard DDR (Double Data Rate) of an FPGA development board through the PCIe DMA; starting the accelerator to begin to read graph data from the appointed address of the onboardDDR; distributing the graph data to different processing units to be processed and calculated by the controller; after each processing unit calculates and processes the data, sending a result to a calculation result collection module; and writing the result back to the onboard DDR by the collection module, and reading the result from the onboard DDR by the PC after the whole diagram data finishesbeing processed. The method has the characteristics of being high in performance, energy efficiency, power consumption and the like.

The invention discloses laser-fuze near-field return wave power calculation based on wave beam decomposition and partial irradiation. The laser-fuze near-field return wave power calculation comprises the following steps: firstly, establishing a fuze coordinate system and a laser-fuze launching site model, adopting 3D Max to carry out geometric modeling, then, adopting a triangular grid mode to carry out surface grid division on a complex target surface, and extracting vertex numbers and vertex coordinate information of triangle face-units of a geometric model; carrying out wave beam decomposition on a laser wave beam at a cross section of the laser wave beam, and dividing grids at equal intervals on the cross section of the laser wave beam, approximately considering that strength on each grid is uniform when the grids are small enough, approximating the strength of the whole grid by utilizing strength of the center point of each grid, and adopting a transmission distance to determine longitudinal attenuation in a longitudinal direction; decomposing an original laser wave beam into a plurality of small wave beams, calculating return wave power of each small wave beam, and finally, superposing, thereby obtaining overall return wave power of fuze. The laser-fuze near-field return wave power calculation has the beneficial effects that calculation for a target near-field return wave signal is high in precision, and reaction speed is high.

The invention relates to a controller placement method for a software-defined network. The method includes the steps that firstly, network topology G is roughened through the maximum matching concept of graphs in the graph theory, G is roughened into G1, G1 is roughened into G2, G2 is roughened into G3, ..., and Gm-1 is roughened into Gm until the number of vertex |Vm| of Gm is smaller than c*k, wherein c is generally equal to 15; secondly, a final roughened network Gm obtained after the network topology G is roughened in the first step is divided; finally, after the network topology G is divided into k regions, placement of controllers is conducted, wherein the k regions are division regions of switches. By means of the controller placement method, control of the multiple controllers over one network topological structure by the multiple controllers is achieved, and the limitation generated when a single controller controls the network can be avoided; meanwhile, time consumption of stream establishment in an SDN network can be shortened, and the multiple controllers can easily cope with all streams when a large number of streams are generated in the network.

The invention discloses a graph division method and device and a computer readable storage medium. The method comprises the following steps: determining a graph division rule for proportionally distributing graph data to each computing node in advance according to the computing performance of each computing node of the parallel computing system and a preset measurement mechanism; in a graph division process, determining a data reading range of each loading thread according to a graph vertex number and a loading thread number, simultaneously calling a plurality of loading threads to read graphdata to be divided in parallel by taking a vertex as a unit, and transmitting the graph vertex data read by each loading thread to a corresponding division thread. Each division thread determines a sub-graph to which each piece of graph vertex data transmitted into the corresponding thread belongs according to a graph division rule based on the graph data divided at the current moment, and finallydivides each piece of graph vertex data to the corresponding calculation node based on a corresponding relationship between the sub-graph and the calculation node. The large graph can be efficientlyand accurately divided into a series of sub-graphs with the vertex number or the connecting edge number meeting any proportional relation.

The invention provides a method of deleting a redundant triangle on a mesh. The method comprises the following steps: 1) triangles on the mesh are traversed, and all triangles are merged to differentpolygons; 2) in each polygon, deletion of overlapped edges is carried out, and the polygon is simplified; 3) triangulation is carried out on the simplified polygon, and a new mesh is generated. Through using the algorithm, a mesh with a large number of triangles and vertexes can be changed to a mesh with a small number of triangles and vertexes, and the mesh appearance and main vertexes remain unchanged. Through deleting redundant triangles on the mesh, the number of triangles and vertexes on the mesh is reduced, the computational ability and bandwidth requirements of a graphics processing unit (GPU) in rendering are reduced, and the rendering efficiency is further improved.

The invention discloses laser-fuze near-field return wave power calculation based on wave beam decomposition and partial irradiation. The laser-fuze near-field return wave power calculation comprises the following steps: firstly, establishing a fuze coordinate system and a laser-fuze launching site model, adopting 3D Max to carry out geometric modeling, then, adopting a triangular grid mode to carry out surface grid division on a complex target surface, and extracting vertex numbers and vertex coordinate information of triangle face-units of a geometric model; carrying out wave beam decomposition on a laser wave beam at a cross section of the laser wave beam, and dividing grids at equal intervals on the cross section of the laser wave beam, approximately considering that strength on each grid is uniform when the grids are small enough, approximating the strength of the whole grid by utilizing strength of the center point of each grid, and adopting a transmission distance to determine longitudinal attenuation in a longitudinal direction; decomposing an original laser wave beam into a plurality of small wave beams, calculating return wave power of each small wave beam, and finally, superposing, thereby obtaining overall return wave power of fuze. The laser-fuze near-field return wave power calculation has the beneficial effects that calculation for a target near-field return wave signal is high in precision, and reaction speed is high.

The invention discloses a simplification method of polygon models of an image, comprising the following steps of: obtaining a left boundary pixel point set and a right boundary pixel point set through scanning a known binary image; obtaining all vertexes of convex polygon models of the image through searching the left boundary pixel point set and the right boundary pixel point set; replacing two vertexes of the shortest edge by an intersection point of extension lines of two adjacent edges of the shortest edge in the convex polygon models so as to achieve the purpose of deleting the shortest edge; after deleting the shortest edges for many times, obtaining a polygon model with same vertex number as necessary reference point number for image correction; and taking the vertexes of the obtained polygon model as the reference points for image correction. The simplification method of polygon models, disclosed by the invention, has simplicity and rapidness, and necessary reference points for image correction can be accurately obtained.

The embodiment of the invention provides a graph rendering method. The method comprises the following steps: acquiring vertex data of N vertexes to be rendered; if the load capacity of the GPU is larger than a preset value, M to-be-rendered vertexes are obtained from the N to-be-rendered vertexes according to the vertex data of the N to-be-rendered vertexes, and the M to-be-rendered vertexes are the vertexes visible in the visual angle range of a user; m is a positive integer smaller than N; and transmitting the vertex data of the M vertexes to be rendered to the GPU, so that the GPU performs graphic rendering. According to the method and the device, under the condition that the load capacity of the GPU is relatively large, the invisible vertexes in the visual angle range of the user are removed, so that the number of the vertexes needing to be rendered on the GPU side is reduced, and the load of the GPU is reduced.

The invention provides a graph data storage method and device.The method comprises the steps that vertex data and edge data are determined according to a graph structure of a target business, the vertex data comprise vertex reference identifiers of all vertexes, and the edge data comprise edge reference identifiers of all edges; the vertex reference identifiers of the vertexes and the edge reference identifiers of the edges are stored in the fixed-length data segments respectively, the lengths of the fixed-length data segments storing the vertex reference identifiers of the vertexes are the same, and the lengths of the fixed-length data segments storing the edge reference identifiers of the edges are the same; and forming identifiers of the vertexes according to the storage positions of the vertexes in the file, and establishing an index tree of a graph structure by combining the identifiers of the vertexes. According to the method and the device, the vertex reference identifier and the edge reference identifier are stored in the fixed-length data segment, the dynamic multiplexing of the storage space is realized, and the identifier of each vertex is formed through the storage position of the vertex, so that each vertex has ID invariance, and the positioning efficiency of the data is improved.

The invention discloses a graph vertex parallel recoding method based on a large synchronization model in a network system. The method comprises the steps that each task transmits a vertex Id of an input graph to an out-degree vertex, counts the number of respective input vertexes, and writes the number into a total synchronization directory; each task updates an original vertex Id according to the number of input vertexes of each task recorded by the total synchronization directory in a progressive increase manner, establishes new and old Id mapping at the same time, then establishes an inputvertex set of each vertex according to a received message, and finally performs reverse sending according to the input vertex set by taking the new Id as a message value; and each vertex is mapped according to the new and old vertexes Id, and the received new vertexes Id are summarized into a new output edge set. According to the method, the problems of storage resource waste, low-efficiency calculation and the like caused by vertex irregular coding can be avoided. Meanwhile, the structural relationship of the original graph is not influenced, and the method has wide practical value and application prospect in the technical field of graph calculation.

The invention provides a database search method for obtaining a maximally complete subgraph from an m-part graph. The search method is applied to a face recognition database. The method forms a k-order maximally complete subgraph by establishing an undirected graph model. subgraph, and store the undirected graph G through the adjacency linked list, take the positive sequence edge and the largest vertex as the basic quantity, and use the pruning method to calculate the number of vertices of the undirected graph G and the calculation amount T of the k-order complete subgraph k Perform calculations to obtain the space complexity and time complexity of the maximally complete subgraph, complete the search for the maximally complete subgraph, and use the search results in face recognition data. By looking for frequent itemsets, the Access to the database improves recognition efficiency.

The invention discloses a power grid sub-graph construction method based on region division, a topology analysis method, a power grid sub-graph construction device based on region division, and a topology analysis device. The power grid sub-graph construction method comprises the following steps: acquiring equipment components and attribute information thereof in a power grid system for establishment as vertex data; dividing the vertex data into a plurality of region vertex sets according to a geographic region attribute of the equipment; acquiring a binary relationship between each two vertices in each region vertex set, and constructing a binary relationship set of each region vertex set according to the binary relationship; constructing a first edge set according to the connection relationship of the equipment components in the power grid system; constructing a second edge set corresponding to the region vertex sets according to the correspondence between the binary relationship setand the first edge set; and constructing a region sub-graph of the power grid system according to the region vertex sets and the second edge set. By implementing the invention to effectively divide the power grid data, the divided sub-graphs are relatively balanced in scale, thereby realizing distributed storage and parallel processing of the power grid data, and improving the efficiency of topology analysis.

The invention proposes a workload-driven distributed graph data segmentation and replication method. In the present invention, the data in the graph data set are respectively stored in the server cluster through the hash table method; combined with the user's workload information, the active vertex set is constructed in the vertex data set combined with the load determination of the source server, and then each active vertex is combined with the server cluster According to the score function score, construct the target server score list, determine the target migration server through the target server load judgment; combine the active vertex set and the corresponding target migration server, judge whether the active vertex is a high-read vertex through the threshold, if the active vertex is not high The read vertex further dynamically adjusts the copy data of the active vertex through the maximum replication factor. The invention utilizes the characteristics of dynamic change of the workload to dynamically adjust the position of the graph vertex data, thereby improving the query efficiency and ensuring low delay and high throughput of the query.