Skip to main content
6 events
when toggle format what by license comment
Mar 15, 2014 at 3:06 history edited Bill Bradley CC BY-SA 3.0
Extended numerical experiments and accelerated rejection algorithm
Jan 31, 2014 at 20:13 comment added Stephan Müller @Bill Its pure engineering, its good to hold as much as possible in CPU's cache. Depending on implementation either row- or column-access is good in its memory access pattern while the other is unfavorable. However by transposing one factor, you can work cache optimal. This sounds as a minor detail, but from a performance point of view its not.
Jan 31, 2014 at 15:59 comment added Bill Bradley @Stephan Müller How can you speed up the multiplication by transposing them? That sounds interesting.
Jan 30, 2014 at 21:58 comment added marshall arxiv.org/pdf/math/0511636v1.pdf has another approach for solving these sorts of problem exactly for small matrices.
Jan 30, 2014 at 18:21 comment added Stephan Müller I like the simultaneous singularity check of $M_1, ..., M_k$. You could also use some of them transposed for even faster multiplication.
Jan 30, 2014 at 15:16 history answered Bill Bradley CC BY-SA 3.0