A recently developed library of C functions used for assembling and directly solving large, symmetric and sparse systems of linear equations as needed for linear and nonlinear applications of the FE-method including eigenmode analysis is described. For the computationally most demanding parts of the solution process, i.e. first of all for the LU-factorisation of the global stiffness matrix, but also for forward reduction, back substitution and eigenmode analysis, new algorithms based on parallel programming by the "thread" approach on shared-memory multiprocessors have been developed. Especially the parallel efficiency of the so-called "Parallel Column Cholewski" or "PCC" LU-factorisation algorithm, which is generally considered to be difficult to achieve, has proved to be excellent. In fact, this algorithm is based on a relatively simple and - to our knowledge - new idea which greatly enhances parallel efficiency. Some of the timing results obtained on two shared memory parallel machines are given. The source listing of the library's header file is available via Internet. There the functionality of all externally callable functions as well as their in- and out-parameters are explained with all the details needed for actually using them in an FE-program.

purchase the full-text of this paper (price £20)

go to the previous paper
go to the next paper
return to the table of contents
return to the book description
purchase this book (price £75 +P&P)