LAPACK Archives

[Lapack] Regarding use of zggev


Hi Marta, 

Matlab is using optimized LAPACK for performing the linear algebra it needs. 

When you call eig(A,B) in Matlab, you are essentially using a wrapper on top of 
LAPACK ZGGEV. 

For large enough matrices, the cost of the wrapper is practically negligible, 
so Matlab?s performance are the same as the one of LAPACK. 

I do not think you are doing anything wrong with your Fortran programming. 

Cheers, 
Julien.


On Jun 24, 2014, at 1:24 AM, marta.trodahl@Domain.Removed wrote:


Hello!

I am working on a oceanographic problem in which I need to solve a 
generalized eigenvalue problem with complex matrices. I have written my code 
in Fortran, where I call the zggev solver inside 4 loops. Through 101x201 
wavenumbers as the two innermost loops, and I am attempting,in The two outer 
loops, to loop over every tenth gridpoint in my domain-which is the Arctic 
ocean.  I tested on a smaller domain 800x1000 gridpoints, but it took days to 
finish. I am wondering if you have any advice on how to optimize the calling 
of The solver inside many loops? My supervisor is solving The same problem, 
but in matlab using eig(), I wanted to use fortran to speed up The 
computations, but so far our programs are just as slow. I have tried to write 
my code with both a derived datatype in The module containing my subroutines, 
and also by just creating my matrices(which are done within these loops) by 
arguments sent in and out of the routines from the main program. The 
different approaches 
are still just as slow. I am a newbie in fortran programming,so it might be 
my poor programming skills causing a bottleneck somewhere, but if there's a 
chance you have any tips with the solver I would very much appreciate it!

Have a great day!

Kind regards,
Marta Trodahl


Sendt fra min iPhone
_______________________________________________
Lapack mailing list
Lapack@Domain.Removed
http://lists.eecs.utk.edu/mailman/listinfo/lapack


<Prev in Thread] Current Thread [Next in Thread>


For additional information you may use the LAPACK/ScaLAPACK Forum.
Or one of the mailing lists, or