The time taken by GESVD to run

[Erasmo_Coletti]

Hi,

I am runningdgesvd_ in Visual Studio 2008 and comparing the results I get with these from the corresponding function implemented from an algorithm in 'Numerical Recipes in C'. The time taken by the algorithm we implemented from Numerical recepices is roughly constant.In release mode the dgesvd_ for a 60 by 60 matrix sometimes takeszero seconds some other times up to one second.

I do not understand how the time taken by dgesvd_ to run can swing so much.

Does somebody have an idea why?

dgsvd_ is returning non negative orderedeigenvalues. Can somebody advise mea fast to run functionfrom MKL that returns negative eigenvalues?

Thank you.

Erasmo.

[Gennady_F_Intel]

Erasmo,

I do not understand how the time taken by dgesvd_ to run can swing so much.

it depends how you measure the execution time of dgesvd. The simplest way to get the correct execution time -execute NTIMES ( say NTIMES = 11 ) times and calculate the average execution time.

It will take you the first approximation of real execution time of this program.

double start = dsecnd();

for(; NTIMES; ){

dgesvd()

}

double finish=dsecnd();

the execTime = (finish - start)/NTIMES;

--Gennady