Porting VASP version 6.4.3 on Fujistu A64FX platforms

[ivr900]

This is a notice for VASP users using Fujitsu A64FX platform.

As found the makefile.include template files for Fujitsu A64FX (makefile.include.fujitsu_a64fx, makefile.include.fujitsu_a64fx_omp) in the vasp-6.4.3 soruce code distribution require update in the line

Code: Select all

-OBJECTS_O2 += fft3dlib.o
+OBJECTS_O2 += fft3dlib.o nonl.o vdw_nl.o

This update reduces compiler optimization level from -Kfast to -O2 when compiling two object files nonl.o and vdw_nl.o

With this update as above, the binaries verification over the whole test suite pass. Otherwise, i.e. with the distributed templates, two test fails.
One is bulk_BN_vdW-DF3-opt1. To pass it nonl.o optimization must be reduced.
Another is TiO2_IBRION=2. It is affected by level of compiler optimization of vdw_nl.o

If your VASP calculations on Fujitsu A64FX machines involve either the non-local projection operators on a plane wave grid or the non-local correlation functional of Dion et al., the above update of makefile.include is strongly advised.

Ivan Rostov
NCI Austrlalia, Canberra

[henrique_miranda]

Thank you for the notice.
I added this to our list of known issues:
https://www.vasp.at/wiki/index.php/Known_issues

When you say "non-local projection operators" you probably mean the ones that are applied to reciprocal space (LREAL=.FALSE.) since you only mention the nonl.F file.

[henrique_miranda]

I have also updated the archetypical files on our wiki:

https://www.vasp.at/wiki/index.php/Make ... itsu_a64fx
https://www.vasp.at/wiki/index.php/Make ... _a64fx_omp

Once again, thank you for letting us know about this!