Hi Simon, what would be a nice piece of work. Good luck and have fun ;-)
Please test the following if your implementation is ready: Try to use your Strassen implementation only with 1 step of recursion (in the next step only the naive algorithm). And a second version with only 3 steps of recursion. This will bound the memory requirements and give a speed up. This is a known heuristic (to stop the recursion early). But I don't know if it is better than the parameters for the algorithm found by the benchmark. By the way: With "Winograd's method" I mean the one from Section 2.3 (page 10) from http://arxiv.org/abs/1106.1347 . But please note that the survey article mentioned above is the first draft. It is not complete and the implementations are not optimized. Best wishes, Ivo. Am 18.07.2011 um 15:36 schrieb Simon King: > Hi Ivo, > > On 18 Jul., 13:32, hedtke <hed...@me.com> wrote: >> With the in-memory variant I mean: "Boyer, Dumans, Pernet and Zhou: Memory >> efficient scheduling of Strassen-Winograd's matrix multiplication algorithm. >> International Symposium on Symbolic and Algebraic Computation 2009." > > Of course, I can not speak for LinBox. But the stuff in > sage.matrix.strassen could really be a bit more memory efficient: It > uses more than 5 times the temporary memory that is needed even in the > least memory efficient multiplication algorithm from the paper you > cited. > > So, I'm now trying to improve the performance of sage.matrix.strassen. > > Cheers, > Simon > > -- > To post to this group, send an email to sage-devel@googlegroups.com > To unsubscribe from this group, send an email to > sage-devel+unsubscr...@googlegroups.com > For more options, visit this group at > http://groups.google.com/group/sage-devel > URL: http://www.sagemath.org -- To post to this group, send an email to sage-devel@googlegroups.com To unsubscribe from this group, send an email to sage-devel+unsubscr...@googlegroups.com For more options, visit this group at http://groups.google.com/group/sage-devel URL: http://www.sagemath.org
