Re: [Lapackpp-devel] Updates
Status: Beta
Brought to you by:
cstim
From: Christian S. <sti...@tu...> - 2005-04-06 15:05:27
|
Hi Jack, yes, please test this as well. Oh, I just see that LaLUInverseIP doesn't check the size of the work variable and the documentation doesn't say anything about the expected size of the work variable. Maybe you could add code like this: // Check for minimum work size if ( work.size() < A.size(0) ) { int nb = LaEnvBlockSize("DGETRI", A); long int W = N*nb; work.resize(W); } Regards, Christian Jacob (Jack) Gryn schrieb: > I have tested them all except for the LUInverse with the external work > variable. If you like, I'll test that one too. > > Jack > > >>-----Original Message----- >>From: Christian Stimming [mailto:sti...@tu...] >>Sent: Wednesday, April 06, 2005 6:04 AM >>To: Jacob (Jack) Gryn >>Cc: lap...@li... >>Subject: Re: [Lapackpp-devel] Updates >> >>Dear Jack, >> >>yes, I've seen your code. It looks quite good and I didn't have any >>further questions (otherwise I'd have replied earlier). Have you >>verified the new functions that they really do what they are supposed >>to? If yes, then it's good. If you want to, I could make a 2.2.1 release >>with your functions added, but otherwise I'd wait until May for the next >>release. >> >>Regards, >> >>Christian >> >>Jacob (Jack) Gryn schrieb: >> >> >>>Incase you haven't noticed, I've updated the CVS with the following >>>additions: >>> >>> >>> >>>- Addition of LUInverse and Addition of LaEigSolve for >>>general-purpose matrices. >>> >>> >>> >>>There are two versions of the Inverse; one of which, you can re-use the >>>'work' variable. >>> >>> >>> >>>There are two versions of the LaEigSolve; the input is a square >>>LaGenMatDouble - since the output may have complex eigenvalues, one >>>version has output with an LaVectorComplex eigenvalue set, and the other >>>version has two LaVectorDouble eigenvalue sets (one real and one >>>imaginary; in case you only need one or wish to deal with real and >>>imaginary data separately, it won't allocate more memory). Both >>>versions give a LaGenMatDouble matrix for the set of eigenvectors. >>> >>> >>> >>>Jack >>> > > > > > |