search

dbeurle / neon

A finite element code
Other
12 stars 8 forks source link

Reordering before iterative linear solve #148

Closed dbeurle closed 5 years ago

dbeurle commented 5 years ago

Performing a bandwidth reduction ordering of the sparse matrix before a linear solve has been shown to improve cache locality in the expensive spmv kernel. This PR implements a fast Cuthill-McKee algorithm to reorder the sparse matrix.

Closes #135.