error on the parallel computing

[caohao001]

Describe the bug

Dear developers,
The error (psi_norm <= 0.0) is occured when I followed the example P009_32H2O_pw(The INPUT file and STRU file are not changed). The 2018_intel_MPI + MKL is used to compile on the Centos system (HPC). The version of EPLA is 2021. The run command is: mpirun -np 12 abacus > log. The log file is presented below.
Thanks.

ITER ETOT(eV) EDIFF(eV) DRHO TIME(s)
m = 3
j = 0 lagrange norm = 4.62493e+21
j = 1 lagrange norm = 7.66347e+20
j = 2 lagrange norm = 5.43346e+16
j = 3 lagrange norm = 8.86966e+38
in DiagoCG, psi norm = -5.36155e+21
If you use GNU compiler, it may due to the zdotc is unavailable.

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
NOTICE
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

psi_norm <= 0.0
CHECK IN FILE : OUT.autotest/warning.log

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
NOTICE
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

Expected behavior

No response

To Reproduce

No response

Environment

No response

Additional Context

No response

[Liu-RX]

I ran the case on Ubuntu 20.04 with ABACUS 3.0.0 compiled with intel-oneAPI 2021.3.0. The elpa version is also 2021. I tested serial run and mpirun with 2 and 4 cores, and did not came into the bug. It might be a bug with different version of compiler or operating system?

[caic99]

I can repeat this bug on 103_PW_15_CF_CS_S1_smallg, using 4 OpenMP threads. ABACUS is compiled with gcc with -O0 flag.

[pxlxingliang]

By using the latest develop codes, I can not repeat the bug with ABACUS applied GNU or intel docker images.