in case you still need a complex singular value decomposition driver, I
have attached our current one. You will find the four arithmetics since
there have been some patches on the real drivers as well.
The SVD algorithm is based on the bidiagonal QR algorithm, not the MRRR
one. (The MRRR variant is still on going work at the Berkeley side (LBL
With best regards,
On Wed, 12 Oct 2005, Osni Marques wrote:
It shouldn't be difficult to write PZGESVD; writing the corresponding
testers is likely to require more time. In fact, I had started doing the
conversion of PDGESVD into PZGESVD a while back but then the Sca/LAPACK
team decided to change the focus of the corresponding development
effort, so we put PZGESVD on hold. Another reason is that we have plans
for a complete new parallel SVD based on the MRRR algorithm, borrowing
the techniques implemented in LAPACK's DSTEGR.
On Wed, 12 Oct 2005, Julien Langou wrote:
- you are right PZGESVD is missing from ScaLAPACK at the moment.
Is there something I'm missing that makes it very hard to write PZGESVD?
- PZGESVD is not terribly hard to write, you are correct, it is just a
question of time.... if you got this time please feel free to write a
version and send it to us. (The bottleneck for us is that adding a
driver means write the testing driver, write the timing driver, and test
all this on various platforms.)
On Tue, 11 Oct 2005, Chip Coldwell wrote:
I noticed that ScaLAPACK seems to be missing a parallel complex
singular value decomposition. I need one for my application, so I
thought I would try to write one. But before doing that, I want to
make sure that there isn't a mathematical/algorithmic reason why
somebody else hasn't already done it. Is there something I'm missing
that makes it very hard to write PZGESVD?
Scalapackers mailing list
-------------- next part --------------
A non-text attachment was scrubbed...
Size: 102400 bytes