[QE-developers] Accuracy issues in QE with FFTW's OPATIENT planning mode

[Raut, S Biplab]

[AMD Official Use Only - Internal Distribution Only]

Dear QE developers,
                                I am trying to use external FFTW library that is modified to use OPATIENT FFTW Planner option/mode instead of default OESTIMATE option.
No change is made to QE application source code to change OESTIMATE to OPATIENT, but instead the external FFTW library is changed to use OPATIENT option.
But, when running QE with AuSURF test case, I am seeing a huge difference in energy values (SCF accuracy is getting distorted drastically).

I would like to know if anybody has tried QE with FFTW's OPATIENT planner option?
The resulting difference between FFTW's OESTIMATE and OPATIENT is generally of the order of floating point precision, then what is causing this accuracy loss in QE with OPATIENT option?
How are the energy values and accuracy calculated in QE - is FFTXLib used as reference at any time during the program execution?

With Regards,
S. Biplab Raut

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/developers/attachments/20200911/c6d358ca/attachment.html>