Systolic array for schur complement computation

Abstract

The computation of the Schur complement in the domain decomposition method often forms a bottleneck to the problem of solving the large sparse linear systems which occur in the Finite Element Method. A systolic array with ns + n(n + 1)/2 processing elements is designed to compute the Schur complement for a bordered block diagonal matrix. It is shown that the systolic array can attain an efficiency as high as 77.8%.

Keywords:

• Systolic array
• Schur complement
• domain decomposition
• speedup
• efficiency

C.R. CATEGORIES::

• G.1.3.
• C.1.2