- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
It seems that we have to call the MKL DSS functions in order in the same executable. After calling the factor command, is there a way to store the factored matrix in the same format as the matrix we send to it? I want to write this factored matrix to a file, then call solve on it later on. Also, can we send MKL DSS a zero-indexed matrix (so the top-left element is the (0,0) element)? Thanks alot.
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
It seems that we have to call the MKL DSS functions in order in the same executable. After calling the factor command, is there a way to store the factored matrix in the same format as the matrix we send to it? I want to write this factored matrix to a file, then call solve on it later on. Also, can we send MKL DSS a zero-indexed matrix (so the top-left element is the (0,0) element)? Thanks alot.
Also tried but found PARDISO (DSS) does not support this. We cannot extract a factorized matrix.
A workaround may be tostorethe internal data structure (void *pt[64]) or call pardiso with multiple right hand side.
I believe PARDISO only supports 1-based indexing.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
I think unless the author providesa function like GetFactorizedMatrix(), there's no way toextract this information. Other packages such as UMFPACK, CHOLMOD, and TAUCS all allowus to extract a factorized matrix, but I do not know why PARDISO (=DSS)does not support this,I guess maybe on purpose?
If there's a way to store a factorized matrix in any format, Ialso really want to know about it.
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page