Page 1 of 1

Think there is a bug in Zgeev

PostPosted: Fri May 21, 2010 12:44 pm
by wk-cof
My task is to perform the iterative diagonalization of a complex-valued square matrix. Occasionally we try to diagonalize matrices that are "nearly diagonal": Most but not all of the off diagonal elements are either 0 or nearly 0.

For the nearly diagonal matrices, BLAS version of Zgeev works fine. CULA, however, gives me a "segmentation fault" error while calling culaZgeev routine.

I attached both BLAS and CULA versions of my program with a sample input matrix.

In both CULA and BLAS:
diag.c - calls BLAS or CULA routine
diagtest.c - wrapper. sets up variables and calls diag.

omegai.dat - imaginary part of input matrix
omegar.dat - real part of input matrix

Makefile - (make diagtest)

In BLAS version:
*.f - fortran dependencies

Re:Think there is a bug in Zgeev

PostPosted: Fri May 21, 2010 2:14 pm
by kyle
It doesn't look like the files were successfully attached. If you'd like, you can e-mail them to me: spagnoli@emphotonics.com

Re:Think there is a bug in Zgeev

PostPosted: Fri May 21, 2010 3:55 pm
by kyle
I can confirm that this is a reproducible bug. The segmentation fault has since been fixed and will most likely be available in CULA 2.0 (final).