Errata to:

``Generalised vectorisation for sparse matrix--vector multiplication'',
by A. N. Yzelman, pre-print version.

Section V.2.C, page 6: `The shared memory machine [..] contains two Intel Xeon E5-2690 v2 processors [..]. Each socket connects four 16 GB DDR3 modules operating [..] at a theoretical maximum bandwidth of 51.2 GB/s.': while technically correct, the whole machine has a theoretical maximum bandwidth of 102.4 GB/s. This is 2.4x less than the NVidia K20X GPU, not 4.9x, as reported on page 7.