implementation of GpSM program by Ha-Nguyen to solve subgraph isomorphism problem on GPU
注意：所有数据集的节点编号都必须从0开始，所有标签必须从1开始， 否则会有问题
we target at a one-to-one mapping at a time, the query graph is small(vertices less than 100), while the data graph can be very large. (but can be placed in GPU's global memory)
目前的子图同构采用的是普通的子图的形式，而不是导出子图的形式(induced subgraph)
论文中默认处理的是无向的不带边标签的图，对于有向图或带边标签的图，均需要做修改(比如论文中用到的生成树) 或者先视为无向图用算法完成匹配，再对匹配的每条结果用真实完整的限制条件进行验证
默认图中不存在这种情况: A->B and B->A, 否则在视为无向图处理时，将出现平行边
Fast Subgraph Matching on Large Graphs using Graphics Processors, DASFAA 2015 (CCF Database B class)
filter and verify framework
edge as the basic unit
joins candidate edges in parallel to form partial solutions during the verification phase
To solve the problem of the considerable amount of intermediate results for joining operations, we adopt the pruning technique of "Relational consistency algorithms and their application in finding subgraph and graph isomorphisms", further enhancing it by ignoring low-connectivity vertices which have little or no effect of decreasing intermediate results during filtering.
When a directed graph with edge labels is viewed as an undirected graph without edge labels, many parallel edges may arise as prblems. (We believe there are no parallel edges in the original directed graph with edge labels)
TODO: better to consider this situation to ensure the correctness