next up previous contents
Next: E.1 Modified Compressed Sparse Up: Dissertation Stephan Wagner Previous: D. Calculation of Additional


E. Matrix Storage Formats

Since the linear equation systems are arising from discretized partial differential equations, only a few elements of the $ n \times n$ entries of the system matrices are unequal to zero. In Table 5.1 the share of nonzero entries is between 3.24$ \,$ $ {\,}^0\!/_{00}$  and 0.04$ \,$ $ {\,}^0\!/_{00}$. In order to reduce the memory consumption required for storing such sparse matrices, specific matrix storage formats are employed.

The assembly and in-house solver module use the storage format called MCSR [178], which stands for Modified Compressed Sparse Row. An analogous concept is the MCSC format, which stands for Modified Compressed Sparse Column. The order of the latter format is not row-oriented, but column-oriented, which significantly speeds up deleting of a complete column. The PARDISO solver module requires the matrix in a format called CSR, which stands for Compressed Sparse Row. The SAMG package requires a format with the same name, but a different variant of it.



Subsections
next up previous contents
Next: E.1 Modified Compressed Sparse Up: Dissertation Stephan Wagner Previous: D. Calculation of Additional

S. Wagner: Small-Signal Device and Circuit Simulation