I still use 64 bit double, but first I will use use linked array instead of linked list to save pointer usage, second I will dump the factored part of the LU immediately to disk and free those row and columns that are factored to save memory, and since we only working on a small portion of the matrix at a time, initially the matrix, say in a compressed row format, is in the disk, will bring them back in to memory a chunk at a time (maybe 4k-8k page).
I think most other matrix solver that tries to save memory are more interested in fill-in reduction or use a iterative method. they do not specifically target for a system with, say 1G RAM and 32G storage. On Wednesday, March 2, 2016 at 2:16:01 AM UTC-8, Harald Schilly wrote: > > > > On Wednesday, March 2, 2016 at 11:02:01 AM UTC+1, Margaret Hu wrote: >> >> I wonder writing a direct sparse LU matrix solver in C/C++ will be of >> interest to sageMath. >> > > Hello Margaret > > So, your solver idea is for 32 and 64 bit floating point number matrices? > Here at SageMath we have all kind of rings, so it's always good to state > this. Hence, I think your project idea is pure numerical mathematics and > although it's interesting, I'm not sure if we can mentor it. Have you tried > to find other organizations, that are more into this? > > also, how does your idea compare to SuperLU, UMFPACK/SuiteSparse, or other > approaches mentioned here? > http://crd-legacy.lbl.gov/~xiaoye/SuperLU/SparseDirectSurvey.pdf > > -- h > > > > -- You received this message because you are subscribed to the Google Groups "sage-gsoc" group. To unsubscribe from this group and stop receiving emails from it, send an email to [email protected]. To post to this group, send email to [email protected]. Visit this group at https://groups.google.com/group/sage-gsoc. For more options, visit https://groups.google.com/d/optout.
