HypergraphsSuccintClosureCUDA

The succint closure on hypergraphs, with multi-threading on CUDA and openMP

Calcolo della chiusura transitiva succinta dato un ipergrafo H, in maniera parallela ove se ne disponga le risorse.

• Parallelismo CPU: per ogni vertice del ipergrafo si invoca BFS; a seguire li merge dei vettori ottenuti dalla BFS.
• Parallelismo GPU: la BFS per trovare tutti i nipoti del nodo.

Calculation of the succinct transitive closure given a hypergraph H, in a parallel way where there is the Hardware.

• CPU parallelism: BFS is invoked for each vertex of the hypergraph; then it merge the vectors obtained from the BFS.
• GPU parallelism: find all the grandchildren of the node.

Installation:

it needs the NVIDIA toolkit to compile, and an NVIDIA graphics card to run.
Copy all file in a folder, if you are using a windows PC run compiler.bat to compile various combinations of executables, otherwise from the console run the commands as follows.

Hypergraph generator

 \nvcc .\generator.cpp -o generator.exe

Succinct transitive closure of hypergraphs

 nvcc -rdc=true -lineinfo -std=c++17 -Xcompiler -openmp .\progetto.cu -o progetto.exe

 Flags:
  -D HIDE : hide the output of graph#
  -D DEBUG : show information on runtime
  -D FILE_OUT : export graph to file
  -D MAX_THREADS : max cuda threads
  -D MAX_BLOCKS_A : max cuda blocks BFS
  -D MAX_BLOCKS_AI : max cuda blocks BFS inside
  -D MAX_BLOCKS_B : max cuda blocks succintion
  -D NTHR : number of cpu threads
  -D TIME : enable time control
  -D NO_INPUT : remove enter click
  -D NTAB : hide the succinted graph outupt
  -D NO_DOUBLE : to use only external parallelization BFS CUDA
  -D DYNAMIC : Automatic blocks number (Experimental)
  -D CPU : custom BFS runs only on CPU, the other parts run on GPU yet

Work:

At begin generate the hypergraphs with generator and then lunch progetto

  generator.exe -v <number of vertices> -e <number of edges> -s <number of supersets>

  progetto.exe "grafo.txt"