A GPU-based dynamic graph coloring method

Abstract

The invention provides a GPU-based dynamic graph coloring method. The method comprises the following steps: converting an original undirected graph into a directional graph; updating directional diagrams after multiple deletion and inserting operation is carried out in batches; storing outer adjacent information of the directional diagram in a compressed sparse row CSR mode, storing inner adjacent information of the directional diagram in a compressed sparse column CSC mode, and transmitting the outer adjacent information and the inner adjacent information to a GPU global memory; judging whether the queue is empty or lower than a threshold value, if not, partitioning a subgraph formed by the to-be-updated node by using a greedy coloring method, and transmitting a partitioning result to the GPU; if yes, processing by the CPU end; carrying out parallel RC-processing of nodes in each region according to partitioning results Hybrid coloring. According to the method, the parallel processing capability of the GPU can be fully utilized, hybrid partitioning design is carried out on the nodes to be processed, the concurrency is improved, and the correctness and the consistency of a coloring result are ensured.

Description

technical field [0001] The invention belongs to the field of large image data processing, and in particular relates to a GPU-based dynamic image coloring method. Background technique [0002] The graph coloring problem is a classic problem in graph theory. The practical significance of graph coloring is to mark two things with a certain relationship with different colors. In the graph, it means that any node with a common edge has a different color. Graph coloring is an important component in many graph applications and has been widely used in graph partitioning and computation scheduling applications. The first step in many graphics applications is to obtain independent sets of vertices for subsequent parallel calculations through graphics shading. The graph coloring problem is also an important combinatorial optimization problem, which widely exists in various fields such as management science, computer science, molecular physics, and biology. Therefore, exploring the g...

AI Technical Summary

Problems solved by technology

The RC_Local method can achieve dynamic graph coloring in incremental form, without repeating the coloring calculation of the entire graph when the graph is dynamically updated, and can obtain better coloring efficiency, but this method considers the color changes of nodes locally, which means There are two disadvantages: one is that the coloring quality is poor, that is, the coloring result will include more colors; the other is that the coloring result is inconsistent. For the same batch of updates, when the update order of the graph changes, different results will be obtained. coloring result

[0006] The existing graph coloring methods cannot solve the problem well under the dynamic graph mainly because of three reasons

The first reason is that most of the current graph coloring methods are based on static graph model coloring

In a frequently updated dynamic graph, when the graph is updated (addition and deletion of nodes or insertion and deletion of edges), the coloring result of the original graph may not meet the requirements of the coloring algorithm, which leads to dynamic graph coloring question

The simplest way to solve this problem is to recolor the updated graph. Obviously, compared to updating, the time cost of recoloring is too high. We need to propose a simpler method

The second reason is that the update of the graph is an infinite sequence that is continuously received, and the current method based on graph coloring is to perform calculations in sequence according to the order of the queue to be processed. Therefore, frequent updates are calculated in parallel, which affects the efficiency of coloring; Three reasons The traditional RC_Local is to reduce the increase in the color of the entire graph update by visiting the two-layer neighbors of the node, and the time complexity is O(dmax 2 ), has high efficiency and good coloring quality, but there are still some shortcomings in this method, mainly in two aspects: (1) When inserting or deleting edges, only the neighbors of nodes, that is, the local color state, are not considered. Considering the global angle (the whole graph), there is still the possibility of improving the reduction of the chromatic number; (2) the inconsistency of the coloring results

Although due to the sparsity of the graph, the number of colors required in dynamic graph coloring is not proven to be so large, but for the current high-efficiency coloring update based on node saturation, or the coloring update algorithm based on global structure How to quickly process graph updates and ensure the quality of graph coloring will also be a very big challenge

In general, the current methods based on dynamic graph coloring cannot solve the frequent updates on dynamic graphs quickly and with high quality, which urgently requires new methods and technologies to meet the challenges of dynamic graph coloring problems

Images