[Pw_forum] Many tests failed on new platform

carlo sbraccia sbraccia at sissa.it
Tue Oct 5 11:41:57 CEST 2004

Dear Sergey,

is BFGS optimization the only case in which you find this strange
behaviour ? 
If this is the case, try to change in the header of the file
Modules/basic_algebra_routines.f90 the line:

#undef __NOBLAS


#define __NOBLAS

Let me know if this makes some difference.


On Tue, 2004-10-05 at 11:09, Sergey Lisenkov wrote:
> Dear PWscf authors and users,
> I compiled new PWscf (espresso) package on IBM PPC970 cluster both in 32-bit and 64-bit mode. First, I tested the code on example03. The Aliminium test is passed succesfully, but CO example is not passed, convergence did not reach ( number of scf cycles    =  50, number of bfgs steps    =   1). It is very strange, because on other machines this test is passed.
>   On my inputs, the results are the same. On my HP cluster code finished with results 44 scf sycles, 43 bfgs sycles, but on IBM PPC970 cluster this test did 520 scf sycles, 209 bfgs sycles, and did not finish. 
> So, the question is, what is the problem? I checked all libraries (atlas, lapack, lapack3e, goto-blas, essl for fftw, local fftw) - the results are the same, only the time for calculation was changed. I used different optmization flags - it did not help.
> Any hints?
>  Thanks a lot,
>    Best wishes,
>      Sergey
> _______________________________________________
> Pw_forum mailing list
> Pw_forum at pwscf.org
> http://www.democritos.it/mailman/listinfo/pw_forum

More information about the users mailing list