windows-server-2003 / quick_floyd_warshall Goto Github PK

A highly-tuned C++ implementation of the Floyd-Warshall algorithm for the APSP(All Pairs Shortest Path) problem.
This library is targeted to modern(roughly after 2010) x86-64 machines.
A major part of the speedup is based on the paper below:
J. S. Park, M. Penner and V. K. Prasanna, "Optimizing graph algorithms for improved cache performance," in IEEE Transactions on Parallel and Distributed Systems, vol. 15, no. 9, pp. 769-782, Sept. 2004, doi: 10.1109/TPDS.2004.44.

Include quick_floyd_warshall/qfw.h and write

quick_floyd_warshall::floyd_warshall<InstSet::AVX2, int64_t, 0>::run(n, matrix, matrix);

when n is the number of vertices in the graph and matrix is a n * n array of int64_t where matrix[i * n + j] is the weight of the edge connecting vertex i and vertex j.
If there is no edge, you should put quick_floyd_warshall::floyd_warshall<InstSet::AVX2, int64_t, 0>::INF instead.
You can change AVX2 to DEFAULT(no SIMD), SSE4_2 (SSE4.2), or AVX512 according to what SIMD instruction set you want to use.
You have to add the corresponding target options(-msse4.2, -mavx2, -mavx512f, and/or -mavx512bw) when compiling your code.
You can also change int64_t to int16_t or int32_t. Choose this type so that (n - 1) * max{abs(weight)} is less than half of the maximum value in the type.
The last template parameter is unroll_type and should be an integer between 0 and 3. This affects the performance, but it depends on other parameters and the environment which one is the fastet.
Negative edge cost is allowed, but negative cycle is not yet supported.
For more detailed specification, see document.md.

Results of benchmarks on my PC as a reference(conditions below)

• Intel Core i5-10600k
• Arch Linux live
• GCC 12.1.0
• -O3 -funroll-loops
• symmetric = false
• average of 5 runs

You can use the code under the terms of the GNU General Public License GPL v3 or under the terms of any later revisions of the GPL. Refer to the provided LICENSE file for further information.