Execution of cheev freezes?

Execution of cheev freezes?

I am using the cheev routine in two places in my program to compute the eigenvalues and eigenvectors of a complex Hermitian matrix. In both cases the routine is called within a loop, so it runs more than one times (quite a few, actually). While the program proceeds normally through the first point where cheev is called, it just freezes up when it reaches the second point. I get no error messages, no crash information, no "Segmentation Fault" or anything, and there is definitely no memory leak as verified by running top (memory consumption is more or less constant). Even if I leave the program running for 20 minutes or so, it still refuses to proceed beyond this routine call. It's as if it falls into an endless loop or something, but I can't figure out how that is possible since there is no explicit loop structure from my part at that point of the program. Any ideas or suggestions?

Andreas.

P.S. Is perhaps the existence of swap memory an absolute prerequisite for running this routine? I am only asking this because my 1 GB of swap has mysteriously recently disappeared and I don't know what to make of it. Swap is not just consumed or anything; it's just not there (top reports "0k total" swap memory!!!)

publicaciones de 4 / 0 nuevos
Último envío
Para obtener más información sobre las optimizaciones del compilador, consulte el aviso sobre la optimización.

Hi, Andreas.

CHEEVcould fall into an endless loop if NaNs appeared on entry to the routine. Which sizes of the problemdo you solve the first and the second time? Which MKL version?

Michael.

Andreas,

you can also try to turn off MKL memory manager by setting environment variable MKL_DISABLE_FAST_MM to any value.

Michael.

Hi Michael,

in both cases the size of the matrix is 2x2 for the runs I have tried... what exactly do you mean by NaN? Is it the result of a division by zero? In that case, wouldn't C++ complain at runtime? My version of MKL is 7.2.1. I will rerun the program and print the contents of the matrix and see what I get... the MKL_DISABLE_FAST_MM must be set to some random numerical value?

Thanks,
Andreas.

EDIT: It turns out all elements of the matrix are nan's... guess it's time for some further debugging to see what has happened...

Message Edited by aste@ait.edu.gr on 07-27-2005 11:23 PM

Inicie sesión para dejar un comentario.