maxflow-cuda

Implementation of the maximum network flow problem in CUDA. Uses the parallel push-relabel algorithm, as illustrated in this paper .

The program depicts the values of height, excess flow and the residual flow for each node at each iteration, along with the maximum flow obtained. It also runs a serial version of the push relabel algorithm and verifies the output.

Execution :

To build the project, run :
make

To run the algorithm :
./maxflow <number of vertices> <number of edges> <source vertex> <sink vertex>

In file "edgelist.txt", edge input to be given in the form :
<end of edge 0> <end of edge 0> <capacity of edge 0>
<end of edge 1> <end of edge 1> <capacity of edge 1>
<end of edge 2> <end of edge 2> <capacity of edge 2>
...

To clean, run :
make clean

Contents

1. Makefile
2. include -
   • serial_graph.h - for the serial implementation
   • parallel_graph.cuh - for the parallel implementation

3.src -

• graph_s.cpp - contains functions used by the serial check
• io_par.cu - contains the input function and the print function used after each iteration
• preflow.cu - contains the preflow/Init routine
• push_relabel.cu - contains the host push relabel routine
• push_relabel_kernel.cu - contains the kernel invoked as part of the push_relabel routine
• global_relabel.cu - contains the heuristic host global relabel routine

References

1. Efficient CUDA Algorithms for the Maximum Network Flow Problem
2. http://www.cse.iitm.ac.in/~rupesh/teaching/gpu/jan20/
3. https://www.geeksforgeeks.org/push-relabel-algorithm-set-2-implementation/
4. Parallel implementation of flow and matching algorithms - Agnieszka Łupińska, Jagiellonian University, Kraków
5. https://www.geeksforgeeks.org/ford-fulkerson-algorithm-for-maximum-flow-problem/