CUDA adn Graph Algorithms Graph theory algorithms and data structures using CUDA

Accelerated Computing CUDA CUDA Programming and Performance
1

Hi there,
We are working on algorithms utilizing Graph theory to develop metabolic and gene network pathways.

Although there are Java and C++ api’s such as the boost graph library, I was wondering if utilizing CUDA we could utilize some existing graph libraries.

Thanks, kenan.

2

Hi kenan,

I develop graph library on CUDA, Thrust Graph Library.

http://code.google.com/p/thrust-graph

TGL supports API’s like BGL, such as, vertex_iterator, edges( g), add_edge( u, v, g), and property_map.

But, the documents are still undeveloped…

thanks, drkkojima

3

Hi drkkojima,

that’s great. can i ask what led to writing this library. is it possible to use boost graph library algorithms/data structures under cuda.

thanks.

kenan.

4

Sorry for the late replies.

Of course! You can ask to me.

TGL supports only 2 BGL algorithms, breadth_first_search and generate_random_graph.

The internal data structure, vertices and edges, is device_vector of Thrust Library.

It is very simple.

If you want to implement some algorithms on GPU, you can access to vertices and edges from apis such as vertices(g, thrust::call_from_device_tag()), edges(g, thrust::call_from_device_tag()), and adjacenct_vertices( u, g, thrust::call_from_device_tag()) in kernel codes.

If you want to access vertices and edges from host code, the apis are verticrs(g) or vertices( gm thrust::call_from_host_tag()) so on.

TGL supplies sample codes, ex. breath_first_search.

5

Sorry for the late replies.

Of course! You can ask to me.

TGL supports only 2 BGL algorithms, breadth_first_search and generate_random_graph.

The internal data structure, vertices and edges, is device_vector of Thrust Library.

It is very simple.

If you want to implement some algorithms on GPU, you can access to vertices and edges from apis such as vertices(g, thrust::call_from_device_tag()), edges(g, thrust::call_from_device_tag()), and adjacenct_vertices( u, g, thrust::call_from_device_tag()) in kernel codes.

If you want to access vertices and edges from host code, the apis are verticrs(g) or vertices( gm thrust::call_from_host_tag()) so on.

TGL supplies sample codes, ex. breath_first_search.