In the netlib documentation, xSYEVR appears to be the current best choice for eigenvalue/eigenvector problems on symmetric matrices, both in terms of speed and memory efficiency. Apparently it's not (yet?) implemented in magma, while the previous best argorithm (xSYEVD) is. Is this a choice? I.e. does it mean that xSYEVD is a much more efficient algorithm than xSYEVR when implemented on a GPU, and therefore one should modify her/his codes to use xSYEVD instead of xSYEVR? Or is it just a matter of time and we will see xSYEVR in magma?
Thanks in advance
Giacomo Mulas
xSYEVR in MAGMA?
-
Stan Tomov
- Posts: 283
- Joined: Fri Aug 21, 2009 10:39 pm
Re: xSYEVR in MAGMA?
Hello,
Yes, it is just a matter of time to add them. Actually, we have them in single and double complex arithmetic. We have the list of all functions here.
We generate the different precisions automatically and in this case there were some differences that we have to manually fix.
Stan
Yes, it is just a matter of time to add them. Actually, we have them in single and double complex arithmetic. We have the list of all functions here.
We generate the different precisions automatically and in this case there were some differences that we have to manually fix.
Stan