Performance issues with bellman_ford_shortest_paths
Hi, I just did a timing experiment on bellman_ford_shortese_paths. It appears to be 200 times slower than LEDA. I am using BGL 1.33.1. Has anyone experienced this? Thanks, Paul
Paul De La Musica
Hi,
I just did a timing experiment on bellman_ford_shortese_paths. It appears to be 200 times slower than LEDA. I am using BGL 1.33.1. Has anyone experienced this?
Thanks,
Paul
The same happened on Release 1.38.0. The graph has negative cycles. In fact, the bellman_ford_shortese_paths() is used here for negative cycle detection. Is there a better work-around? Thanks, Paul
paul wrote:
Paul De La Musica
writes: Hi,
I just did a timing experiment on bellman_ford_shortese_paths. It appears to be 200 times slower than LEDA. I am using BGL 1.33.1. Has anyone experienced this?
Thanks,
Paul
The same happened on Release 1.38.0. The graph has negative cycles. In fact, the bellman_ford_shortese_paths() is used here for negative cycle detection.
Is there a better work-around?
Did you compile with full optimizations? - Volodya
paul wrote:
Vladimir Prus
writes: Did you compile with full optimizations?
- Volodya
Hi Volodya,
Yes. I compiled in "Release" mode on MSVC 7.1 with the "maximize speed" option.
Then, you probably get to produce a self-contained example that reproduces this problem, and file a ticket at svn.boost.org. 200x performance difference sounds very strange. - Volodya
participants (3)
-
paul -
Paul De La Musica
-
Vladimir Prus