how could the RAM for ADMM be negative number

Sun Geng

unread,

Jan 28, 2021, 10:01:06 AM1/28/21

to cp2k

Dear CP2K users,

I am running jobs with CP2K/ADMM method, in which the CP2k prints the following lines:

HFX_MEM_INFO| Number of cart. primitive ERI's calculated: 52451602395221

HFX_MEM_INFO| Number of sph. ERI's calculated: 2343013195064

HFX_MEM_INFO| Number of sph. ERI's stored in-core: 2247108266900

HFX_MEM_INFO| Number of sph. ERI's stored on disk: 0

HFX_MEM_INFO| Number of sph. ERI's calculated on the fly: 0

HFX_MEM_INFO| Total memory consumption ERI's RAM [MiB]: -449594

HFX_MEM_INFO| Whereof max-vals [MiB]: 17729

HFX_MEM_INFO| Total compression factor ERI's RAM: -38.13

HFX_MEM_INFO| Total memory consumption ERI's disk [MiB]: 0

HFX_MEM_INFO| Total compression factor ERI's disk: 0.00

HFX_MEM_INFO| Size of density/Fock matrix [MiB]: 408

HFX_MEM_INFO| Size of buffers [MiB]: 25

HFX_MEM_INFO| Number of periodic image cells considered: 33

HFX_MEM_INFO| Est. max. program size after HFX [MiB]: 31121

The total memory consumption ERI's RAM is a negative number.

Why it is a negative number?

How can I know the real RAM consumption in this job?

Thank you very much in advance.

Best

Geng

Tiziano Müller

unread,

Jan 28, 2021, 11:09:03 AM1/28/21

to cp...@googlegroups.com

Hi Geng,

can you please tell us which version of CP2K you are using and possibly
also submit an input file with which you can reproduce the issue?

But since that's quiet a number of ERIs (which seems a bit weird since
you're using ADMM), my first guess is an integer overflow here:

https://github.com/cp2k/cp2k/blob/5b641945b68ab1861fea87544a697b03403b4443/src/hfx_energy_potential.F#L1776-L1777

... while the target variable is an int8, the variables multiplied are
int4, and get multiplied first as such before converted to int8:
https://gcc.godbolt.org/z/x4hYsb
If you have CP2K v8.1 you can try the attached patch and see whether it
fixes the issue (only compile-tested so far).

Best,
Tiziano

> --
> You received this message because you are subscribed to the Google
> Groups "cp2k" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to cp2k+uns...@googlegroups.com
> <mailto:cp2k+uns...@googlegroups.com>.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/cp2k/02d0eebb-898e-4536-913f-548448e11648n%40googlegroups.com
> <https://groups.google.com/d/msgid/cp2k/02d0eebb-898e-4536-913f-548448e11648n%40googlegroups.com?utm_medium=email&utm_source=footer>.

--
Tiziano Müller
University of Zurich
Department of Chemistry
Winterthurerstrasse 190
CH-8057 Zürich

Tel: +41 44 63 54234
www.chem.uzh.ch
tiziano...@chem.uzh.ch

hfx_overflow_fix.patch

Sun Geng

unread,

Jan 28, 2021, 10:05:02 PM1/28/21

to cp2k

Dear Tiziano,

Thanks for your reply.

I applied the patch to the CP2K code (v8.1), and now it can print a reasonable number for the RAM (below).

May I take the advantage to ask two additional questions regarding the RAM in HFX? Thank you very much in advance.

1) The line " HFX_MEM_INFO| Total memory consumption ERI's RAM [MiB]: 540840" prints the total (including all MPI processors ) used memory, am I right?

2) Is there a way to estimate the required memory for the HFX module before running the CP2k program? I thought no because the memory is allocated dynamically according to the occupation matrix, but I would like to have a double-check.

Thanks again.

Best Regards,

Geng

HFX_MEM_INFO| Number of cart. primitive ERI's calculated: 32799680679415

HFX_MEM_INFO| Number of sph. ERI's calculated: 1498990881919

HFX_MEM_INFO| Number of sph. ERI's stored in-core: 683885389270

HFX_MEM_INFO| Number of sph. ERI's stored on disk: 0

HFX_MEM_INFO| Number of sph. ERI's calculated on the fly: 0

HFX_MEM_INFO| Total memory consumption ERI's RAM [MiB]: 540840

HFX_MEM_INFO| Whereof max-vals [MiB]: 10475

HFX_MEM_INFO| Total compression factor ERI's RAM: 9.65

HFX_MEM_INFO| Total memory consumption ERI's disk [MiB]: 0

HFX_MEM_INFO| Total compression factor ERI's disk: 0.00

HFX_MEM_INFO| Size of density/Fock matrix [MiB]: 408

HFX_MEM_INFO| Size of buffers [MiB]: 25

HFX_MEM_INFO| Number of periodic image cells considered: 33

HFX_MEM_INFO| Est. max. program size after HFX [MiB]: 7536

Tiziano Müller

unread,

Jan 29, 2021, 2:49:58 AM1/29/21

to cp...@googlegroups.com

Dear Geng,

for 1) the memory should be the total over MPI ranks (but only the ERI
storage part). Since the HFX code is OpenMP-parallelized I would suggest
to use less MPI ranks per physical node and more threads. Each MPI rank
has to maintain the full density matrix locally, hence you would like to
avoid too many ranks per node. The ERIs on the other hand can be
compressed and even stored to disk if needed.

for 2) much of the memory usage is known a-priori (before any ERI is
actually calculated), see
https://github.com/cp2k/cp2k/blob/d7da0214a5d90c06d8c33ae514e3a7d3c4a4888b/src/hfx_energy_potential.F#L611-L651,
but some part as you already wrote is dynamic. From that one could
probably derive a (pessimistic) upper limit.

Best,
Tiziano

> https://gcc.godbolt.org/z/x4hYsb <https://gcc.godbolt.org/z/x4hYsb>

> <https://groups.google.com/d/msgid/cp2k/02d0eebb-898e-4536-913f-548448e11648n%40googlegroups.com?utm_medium=email&utm_source=footer

> <https://groups.google.com/d/msgid/cp2k/02d0eebb-898e-4536-913f-548448e11648n%40googlegroups.com?utm_medium=email&utm_source=footer>>.
>
>
> --
> Tiziano Müller
> University of Zurich
> Department of Chemistry
> Winterthurerstrasse 190
> CH-8057 Zürich
>
> Tel: +41 44 63 54234

> www.chem.uzh.ch <http://www.chem.uzh.ch>
> tiziano...@chem.uzh.ch

>
> --
> You received this message because you are subscribed to the Google
> Groups "cp2k" group.
> To unsubscribe from this group and stop receiving emails from it, send
> an email to cp2k+uns...@googlegroups.com
> <mailto:cp2k+uns...@googlegroups.com>.
> To view this discussion on the web visit

> https://groups.google.com/d/msgid/cp2k/b579bdce-d28d-4551-b0b9-fb497a5efa3dn%40googlegroups.com
> <https://groups.google.com/d/msgid/cp2k/b579bdce-d28d-4551-b0b9-fb497a5efa3dn%40googlegroups.com?utm_medium=email&utm_source=footer>.

Sun Geng

unread,

Jan 29, 2021, 11:39:18 AM1/29/21

to cp2k

Hi Tiziano,

Thank you very much for your reply, now I understand the first question.

For the second question (about the estimation of the ERI memory consumption), which tag in the CP2K input file allows printing this estimated memory?