The peak of memory consumption is too high.

[PeizeLin]

Describe the bug

Two layers of graphene, 1588 C atoms.
PBE, scf.

Between LCAO_gen_fixedH::b_NL_beta_new and OperatorLCAO::init, the memory consumption raises from 40GB to 218GB then reduces to 43GB.

Between Gint_interface::cal_gint_vlocal and OperatorLCAO::folding_fixed, the memory consumption raises from 43GB to 221GB then reduces to 56GB.

In the other parts of program, the memory consumption keeps low.
So these two peak limits the calculation of big molecules.

C_11_k1_mpi1_4.zip

Expected behavior

No response

To Reproduce

No response

Environment

Linux 3.10.0-1160.el7.x86_64, Red Hat 4.8.5-44
icpc 2021.5.0 (gcc 10.2.0)
intelmpi 2021.5 / mpich 4.1
mkl 2021.5
elpa_openmp 2021.11.002
cereal 1.3.2

Additional Context

No response

[dyzheng]

build_Nonlocal_beta_new() will cost too much temporary memory for multi threads calculation, maybe parallel by atoms is not a suitable scheme. Can you look at this problem @Alcanderian ?

[Alcanderian]

Yes, we can consider another multithreading scheme.
Before that, I need to examine the problem size of the inner loop.
And maybe we should switch scheme among different problem size.