Applications solved on an unstructured mesh

Hi Everyone,

Does NVDIA CUDA SDK Sample contain sample code for an application that is solved on an unstructured grid?
Unlike structured grids, unstructured grids require a list of the connectivity which specifies the way a given set of vertices make up individual elements. Each node in the grid has irregular connectivity with its neighboring nodes.


Based on my limited knowledge:

Regularity in grids result in “dense supernodes” in sparse matrices which can be solved by BLAS3 operations – thereby speeding up the whole process on CUDA.

Otherwise, without this regularity, I believe that it is difficult to accelerate using CUDA.

Others may be able to tell you more,

There aren’t any unstructured grid examples in the SDK. There is a lot of very interesting work published on using CUDA with unstructured grids. Graham Markall from Imperial College has some great material on his research with CUDA accelerated FEM for CFD applications, for example.

There is also a project by Prof Giles at Oxford University :

The OP2 project aims to develop an open-source framework for the execution of untructured grid applications on clusters of GPUs or multi-core CPUs. Although OP2 is designed to look like a conventional library, the implementation uses source-source translation to generate the appropriate back-end code for the different target platforms.