Cholesky Out-of-core pfdpotrf.f Utilization

Open discussion regarding features, bugs, issues, vendors, etc.

Cholesky Out-of-core pfdpotrf.f Utilization

Postby Sidio47 » Wed Jul 06, 2011 2:16 pm


I am interested in implementations of out-of-core solvers, specifically the Cholesky factorization of a dense symmetric positive definite matrix. I have been searching and have come across Dr. D'Azevedo's and Dr. Dongarra's implementation of the ScaLAPACK prototype routine pfdpotrf.f.

In my work, I will be handling a potentially very large and dense matrices which I will need to complete out-of-core factorizations as well as multiplications with an in-core matrix to form another in-core matrix.

Currently, I have a code set up to handle everything via ScaLAPACK distributed in-core matrices for my application. I came across the LAWNs paper "The Design and Implementation of the Parallel Out-of-core ScaLAPACK LU, QR and Cholesky Factorization Routines". From what I can tell, these prototype routines would at least be a good starting point for me to begin implementation of out-of-core routines.

I have been able to download the source code of the prototype routines and install the library correctly after a few slight modifications. I have had mixed success with the testing routines in that a few of the default configurations in the input file fail while others seem to be fine.

With that being said, what would be most helpful though are examples that show how to take an out-of-core matrix in a file, within some loop structure read in a part of the matrix, complete the block factorization and then store this out-of-core once more. Is there any one who might still have some old examples of this? The testddriver.f routine seems to only implement the default random matrix generator so I don't have exposure to how the dlaread and dlawrite routines truly work.

Thanks in advance,
Posts: 3
Joined: Wed Jul 06, 2011 2:05 pm

Return to User Discussion

Who is online

Users browsing this forum: Yahoo [Bot] and 1 guest