- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Link Copied
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Could you please submit your test case as an attachment to your post? Looks like code your posted has been broken by the text editor.
Could you also specify what "info" value did you obtain on exit of dsyevr() function?
Best regards,
Artem
- 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
- 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
dsyevr calls dsytrd, which is threaded.
Did you try other eigensolvers such as dsyev, dsyevd, dsyevx?
What is the typical execution time for less size problem, say 10000, 20000, 40000 ?
Michael.
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Michael.
- 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
- 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 wrote that is not paralleled RRR algorithm, which is included in dsyevr: http://software.intel.com/en-us/forums/showthread.php?t=73653(Unfortunately, now my page is not available). Intel MKL - very good package, but for your problem, he is not well suited. Most likely an internal error occurs because the RRR algorithm. Most time is spent on bringing to tridiagonal form for BLAS level 2. The rest you can quickly calculate, given that you do not need all the eigenvectors. Once again I say that dsyevr because RRR algorithm inaccurate and unreliable.
- 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
- Mark as New
- Bookmark
- Subscribe
- Mute
- Subscribe to RSS Feed
- Permalink
- Report Inappropriate Content
Propose to act as follows:
1. dsytrd
2. dstedc (To program this feature, it took 10 years.)This algorithm requires a lot of RAM, but fast and reliable.
3. dormtr (Apply this feature only to those eigenvectors tridiagonal matrix that you want).
- Subscribe to RSS Feed
- Mark Topic as New
- Mark Topic as Read
- Float this Topic for Current User
- Bookmark
- Subscribe
- Printer Friendly Page