[CP2K-user] [CP2K:13765] CP2K 7.1 memory leak

[Ole Schütt]

Hi Paul,

that looks like https://github.com/cp2k/cp2k/issues/695.

You can either try to patch the arch file manually or simply rebuild the 
toolchain using the latest version from git.

-Ole


On 2020-08-12 10:43, Paul Schwarz wrote:
> Hi Ole,
> 
> I tried building the pdbg binary, but I came across "undefined
> reference" errors during compilation for libxsmm
> 
> /opt/chem/CP2K/cp2k-7.1.0/tools/build_utils/fypp -n
> --line-marker-format=gfortran5
> /opt/chem/CP2K/cp2k-7.1.0/src/eri_mme/eri_mme_lattice_summation.F
> eri_mme_lattice_summation.F90
> /opt/chem/CP2K/cp2k-7.1.0/tools/toolchain/install/libxsmm-1.14/lib/libxsmm.a(libxsmm_gemm.o):
> In function `__real_dgemm_':
> libxsmm_gemm.c:(.text.__real_dgemm_+0x1): undefined reference to
> `dgemm_'
> /opt/chem/CP2K/cp2k-7.1.0/tools/toolchain/install/libxsmm-1.14/lib/libxsmm.a(libxsmm_gemm.o):
> In function `__real_sgemm_':
> libxsmm_gemm.c:(.text.__real_sgemm_+0x1): undefined reference to
> `sgemm_'
> /opt/chem/CP2K/cp2k-7.1.0/tools/toolchain/install/libxsmm-1.14/lib/libxsmm.a(libxsmm_gemm.o):
> In function `__real_dgemv_':
> libxsmm_gemm.c:(.text.__real_dgemv_+0x1): undefined reference to
> `dgemv_'
> /opt/chem/CP2K/cp2k-7.1.0/tools/toolchain/install/libxsmm-1.14/lib/libxsmm.a(libxsmm_gemm.o):
> In function `__real_sgemv_':
> libxsmm_gemm.c:(.text.__real_sgemv_+0x1): undefined reference to
> `sgemv_'
> collect2: error: ld returned 1 exit status
> /opt/chem/CP2K/cp2k-7.1.0/obj/local/pdbg/all.dep:108: recipe for
> target '/opt/chem/CP2K/cp2k-7.1.0/exe/local/graph.pdbg' failed
> make[3]: *** [/opt/chem/CP2K/cp2k-7.1.0/exe/local/graph.pdbg] Error 1
> 
> I used the standard local.pdbg that was created by toolchain.
> Is this some sort of linking error? Any idea how to resolve that?
> 
> Best wishes,
> Paul
> 
> Paul Schwarz schrieb am Dienstag, 11. August 2020 um 15:06:06 UTC+2:
> 
>> Hi Ole,
>> 
>> thanks for your reply.
>> I will try out the pdbg binary, and compare output to the known
>> leaks.
>> 
>> Paul
>> 
>> Ole Schütt schrieb am Dienstag, 11. August 2020 um 14:16:53 UTC+2:
>> 
>>> Hi Paul,
>>> 
>>> you can run the pdbg binary which has LeakSanitizer enabled. When
>>> CP2K
>>> exits it will print a detailed report of code locations that
>>> leaked
>>> memory.
>>> 
>>> Note that there are already a few known leaks which we usually
>>> ignore:
>>> 
>>> 
>> 
> https://github.com/cp2k/cp2k/blob/master/tools/toolchain/scripts/install_gcc.sh#L165
>>> 
>>> 
>>> Generally, those leaks are considered benign, but if nothing else
>>> show
>>> up then you might want to check them as well.
>>> 
>>> Cheers,
>>> Ole
>>> 
>>> On 2020-08-11 13:54, Paul Schwarz wrote:
>>>> Dear CP2K developers,
>>>> 
>>>> I downloaded the latest 7.1 version and used toolchain to
>>> install the
>>>> libraries libint, libxc, and libxsmm with the following command:
>>> 
>>>> 
>>>> $ ./install_cp2k_toolchain.sh --enable-omp --with-libxc=install
>>>> --with-libint=install --with-fftw=system --with-libxsmm=install
>>>> --with-mkl=system --with-elpa=no --with-cmake=system
>>>> --with-openblas=no --with-hdf5=system
>>>> 
>>>> After the successfull installations, I used the resulting
>>> local.popt
>>>> to make an executable (without any further modifications).
>>>> 
>>>> The used compilers were gcc, g++, and mpif90, which in my case
>>> is
>>>> Intel version 2019.5.281, and not OpenMPI.
>>>> 
>>>> Running relaxation simulations of molecules on slab surfaces, I
>>>> experience an increasing memory usage.
>>>> 
>>>> Sometimes the simulation relaxes fast enough, so that I don’t
>>> run
>>>> out of memory.
>>>> 
>>>> Sometimes I run out of memory before the relaxation finishes.
>>>> 
>>>> I call cp2k_shell.popt, since I run my relaxations via an ase
>>>> interface (but I guess it doesn’t matter, since there is no
>>> shell
>>>> binary, but just a link of the shell to cp2k.popt).
>>>> 
>>>> Does anyone else experience memory issues? I have the same issue
>>> on a
>>>> HPC center, where a system admin installed the 7.1 version for
>>> me, and
>>>> here my jobs also crash due to running out of memory.
>>>> 
>>>> Does someone have a suggestion as to what the problem is or how
>>> to
>>>> avoid it?
>>>> 
>>>> Is there maybe a compiler flag/option I should avoid or put in?
>>>> 
>>>> Best regards,
>>>> 
>>>> Paul Schwarz
>>>> 
>>>> --
>>>> 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 cp... at googlegroups.com.
>>>> To view this discussion on the web visit
>>>> 
>>> 
>> 
> https://groups.google.com/d/msgid/cp2k/d9628406-6b9d-4e00-8b6b-edfe1b765d54n%40googlegroups.com
>>> 
>>>> [1].
>>>> 
>>>> 
>>>> Links:
>>>> ------
>>>> [1]
>>>> 
>>> 
>> 
> https://groups.google.com/d/msgid/cp2k/d9628406-6b9d-4e00-8b6b-edfe1b765d54n%40googlegroups.com?utm_medium=email&utm_source=footer
> 
>  --
> 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 cp... at googlegroups.com.
> To view this discussion on the web visit
> https://groups.google.com/d/msgid/cp2k/fd14f21f-7b10-4549-bdc2-1a6b32b5b4abn%40googlegroups.com
> [1].
> 
> 
> Links:
> ------
> [1]
> https://groups.google.com/d/msgid/cp2k/fd14f21f-7b10-4549-bdc2-1a6b32b5b4abn%40googlegroups.com?utm_medium=email&utm_source=footer