Skip to content

[localcolumnobject] add method axpy for quadrature point

Tobias Malkmus requested to merge feature/improve-matrix-assemble-efficiency into master

The assemble of matrices of huge systems has a suboptimal efficiency.

This patch introduces the method localMatrix.column(i).axpy( qp, u, du ). This implementation now makes use of vectorial expressions. Here some assamble timings for a system with dimRange=5 and a simple l2 projection:

Old version new Version
0.0817779 0.0580213
0.329994 0.223477
1.37288 0.869674
5.5101 3.54086

The test setting is modified in this commit to show the timings. It was performed on a starting grid of about 32 x 32 elements.

Merge request reports