Hi Paolo, Thanks for your prompt reply. I went back to check the code path again. I was only using -D__LINUX_ESSL never -D__ESSL. So I made a mistake (sorry for the confusion) in my previous analysis. Actually, cdiagh_lapack seems causing problems. Now I set #if defined (__ESSL) || defined (__LINUX_ESSL) to take the cdiagh_aix and it works well.
Ciao, Ye =================== Ye Luo, Ph.D. Leadership Computing Facility Argonne National Laboratory 2016-05-09 1:16 GMT-05:00 Paolo Giannozzi <[email protected]>: > Hi > > as far as I now, you should NOT use "-D__ESSL" on BGQ machines, but > "-D__LINUX_ESSL". Also please not that irreps are build using a stochastic > algorithms so their phases are quite unpredictable. Anyway: if it works for > you, it's fine: the performance penalty should be negligible, or even > nonexistent. > > PS: I send an email a few weeks ago and got no reply. I hope someone can >> also answer these questions [...] >> >> > 2) Any ways to reduce the phonon disk I/O to 'none' similar to pw.x? >> Reduce_io seems help very little during the calculation. >> > > it actually does very little because it works for the self-consistent part > only. > Unfortunately, reducing I/O in the phonon code (a problem that must be > addressed sooner or later, sooner better than later) requires a serious > investment in terms of time and effort. > > Paolo > > -- > Paolo Giannozzi, Dip. Scienze Matematiche Informatiche e Fisiche, > Univ. Udine, via delle Scienze 208, 33100 Udine, Italy > Phone +39-0432-558216, fax +39-0432-558222 > > > _______________________________________________ > Pw_forum mailing list > [email protected] > http://pwscf.org/mailman/listinfo/pw_forum >
_______________________________________________ Pw_forum mailing list [email protected] http://pwscf.org/mailman/listinfo/pw_forum
