KPAR deadlock in electron_all.f90 — collective ordering mismatch between KPAR groups

[daniel_marchand]

Hey Everyone,

We got some complains from users of hanging. Might or might not be related to this, also very hard to reproduce so I cannot state anything conclusively. Does the patch above work for all VASP versions on only 6.5.1?

Best,

Daniel

[ahampel]

Hi Daniel,

a part of the fix is actually live in 6.6.0 and 6.6.1 . The original problem of electronic minimization in ALGO=Normal/Fast/VeryFast are all fixed there. However, if you do use ALGO=ALL/Damped there is a 2nd issue. I attach two final patches that fix all electronic minimizers including linear response path.

Do you have any further information about the hang? VASP version at least? Or maybe even which compiler toolchain, or which VASP executable / workload? That would be very helpful to narrow it down. According to our known issues there is also a knew issue that one of our LDA+U routines is not entirely thread safe but other than that there are no known problems that fit. Try to apply the patches to your version but if your users used 6.6.x of VASP with ALGO=Normal/Fast/VeryFast it is unlikely that this here is the cause.

Best,
Alex