tuteng0915 / cuda-floyd_warshall

CUDA implementation of the Blocked Floyd Warshall All pairs shortest path graph algorithm

Cuda Floyd Warshall implementation

CUDA implementation of the Blocked Floyd-Warshall All pairs shortest path graph algorithm based on article: "A Multi-Stage CUDA Kernel for Floyd-Warshall" (Ben Lund, Justin W. Smith)

Tested:

Hardware (MSI GP72 7RE): Processor: Intel(R) Core(TM) i7-7700HQ GPU: GTX 1050 Ti 2GB RAM RAM: 8GB RAM

Environment: System: Ubuntu 17.10 NVCC: 9.1 CC: 6.1

Performance results:

|V|	|E|	fw.cpp	fw-cuda.cu	Speedup	blocked-fw-cuda.cu	Speedup
1000	150000	0.724s	0.176s	4.215x	0.117s	6.188x
2000	400000	5.895s	0.642s	9.182x	0.213s	27.67x
5000	500000	84.91s	8.406s	10.10x	2.301s	36.90x
7500	1125000	279.0s	28.31s	9.855x	7.550s	36.95x
10000	2000000	665.9s	63.81s	10.43x	17.28s	38.53x
12500	3125000	1269s	125.5s	10.11x	34.03s	37.27x

Compile source:

1. Install make/nvcc
2. Update makefile with path to nvcc compiler
3. Run command: make all

Run tests:

1. Install Python 3.6
2. Go to project directory
3. Run command: python3 -m unittest discover -s test -v

---

Author: Mateusz Bojanowski