[QE-users] segfault with HSE06
Michal Krompiec
michal.krompiec at gmail.com
Tue Jul 9 13:37:51 CEST 2019
Hi Fabrizio,
Thanks. Do you mean replacing this: (lines 466-475)
IF(DoLoc) then
!$omp parallel do collapse(3) default(shared)
firstprivate(npol,nrxxs,nkqs,ibnd_buff_start,ibnd_buff_end)
private(ir,ibnd,ikq,ipol)
DO ikq=1,SIZE(locbuff,3)
DO ibnd=1, x_nbnd_occ
DO ir=1,nrxxs*npol
locbuff(ir,ibnd,ikq)=0.0_DP
ENDDO
ENDDO
ENDDO
ELSE
with the following:
IF(DoLoc) then
IF(gamma_only) then
!$omp parallel do collapse(3) default(shared)
firstprivate(npol,nrxxs,nkqs,ibnd_buff_start,ibnd_buff_end)
private(ir,ibnd,ikq,ipol)
DO ikq=1,SIZE(locbuff,3)
DO ibnd=1, x_nbnd_occ
DO ir=1,nrxxs*npol
locbuff(ir,ibnd,ikq)=0.0_DP
ENDDO
ENDDO
ENDDO
END IF
ELSE
?
On Tue, 9 Jul 2019 at 11:37, Fabrizio Ferrari <ferrariruffino.fz at gmail.com>
wrote:
> Hello,
> if you put the loop at line 467 of PW/src/exx.f90 inside an
> 'IF(gamma_only)', the segfault should disappear. I'm just checking if that
> is the only fix needed.
>
> Fabrizio
>
> On Tue, Jul 9, 2019 at 12:27 PM Michal Krompiec <michal.krompiec at gmail.com>
> wrote:
>
>> I got a similar segfault using a fresh installation of QE 6.4.1, on a
>> different HPC, this time with Intel 2018 compilers and ELPA, with the same
>> input file and pseudos (SG15) as previously.
>> I noticed that my molecule is a bit too high in the simulation cell (vs.
>> the potential added by assume_isolated), but increasing the size of the
>> cell in the Z direction changed nothing.
>> Switching off assume_isolated also didn't help.
>>
>> from stdout:
>>
>> forrtl: severe (174): SIGSEGV, segmentation fault occurred
>> Image PC Routine Line Source
>> pw.x 00000000056A6D8D Unknown Unknown
>> Unknown
>> libpthread-2.17.s 00002AC70727E5E0 Unknown Unknown
>> Unknown
>> pw.x 00000000005B3618 Unknown Unknown
>> Unknown
>> pw.x 00000000005AE060 Unknown Unknown
>> Unknown
>> pw.x 000000000040ADFD Unknown Unknown
>> Unknown
>> pw.x 000000000059D83D Unknown Unknown
>> Unknown
>> pw.x 00000000004086C9 Unknown Unknown
>> Unknown
>> pw.x 000000000040851E Unknown Unknown
>> Unknown
>> libc-2.17.so 00002AC7077AEC05 __libc_start_main Unknown
>> Unknown
>> pw.x 0000000000408429 Unknown Unknown
>> Unknown
>> forrtl: severe (174): SIGSEGV, segmentation fault occurred
>>
>> Last few lines of the log file:
>>
>> convergence has been achieved in 9 iterations
>>
>> Using localization algorithm with threshold: 0.50D-02
>>
>> Using ACE for calculation of exact exchange
>>
>> EXX grid: 1427071 G-vectors FFT dimensions: ( 90, 90, 375)
>>
>> NBands = 15 nks = 1 nkqs = 9
>> Canonical Orbitals
>>
>> Any suggestions?
>>
>> Thanks,
>> Michal
>>
>>
>> On Mon, 8 Jul 2019 at 13:27, Michal Krompiec <michal.krompiec at gmail.com>
>> wrote:
>>
>>> Hello,
>>> I'm getting a segmentation fault when trying to run a HSE06 SCF
>>> calculation in QE 6.4rc (built with gcc and OpenMPI). I got the same result
>>> regardless of number of OMP threads (1-2) or MPI processes, it is also not
>>> because I'm running out of memory. Increasing OMP_STACK_SIZE didn't help.
>>> I'm using SG15 norm-conserving pseudopotentials.
>>> This is the error message:
>>>
>>> Program received signal SIGSEGV: Segmentation fault - invalid memory
>>> reference.
>>>
>>> Backtrace for this error:
>>>
>>> Backtrace for this error:
>>> #0 0x2ba41caf6607 in ???
>>> #1 0x2ba41caf586d in ???
>>> #2 0x2ba41d9d5fdf in ???
>>> #1 0x2ba41caf586d in ???
>>> #2 0x2ba41d9d5fdf in ???
>>> #3 0x481031 in __exx_MOD_exxinit._omp_fn.41
>>> at
>>> /home/hpcadmin/Cluster-packages/Apps/QE/q-e-qe-6.4-rc/PW/src/exx.f90:471
>>>
>>>
>>> And this is the input file:
>>> &CONTROL
>>> nstep = 150
>>> prefix = 'pz'
>>> calculation = 'scf'
>>> /
>>> &SYSTEM
>>> ecutwfc = 60
>>> ecutrho = 240
>>> occupations = 'smearing'
>>> degauss = 0.03
>>> smearing = 'marzari-vanderbilt'
>>> assume_isolated = '2D'
>>> ntyp = 3
>>> nat = 10
>>> ibrav = 0
>>> vdw_corr='dft-d3'
>>> nosym = .true.
>>> input_dft = 'hse'
>>> localization_thr = 0.005
>>>
>>> /
>>> &ELECTRONS
>>> electron_maxstep = 1000
>>> mixing_mode = 'plain'
>>> mixing_beta = 0.3
>>> mixing_ndim = 10
>>> /
>>> &IONS
>>> ion_dynamics = 'bfgs'
>>> /
>>>
>>> ATOMIC_SPECIES
>>> H 1 H_ONCV_PBE-1.0.upf
>>> C 12 C_ONCV_PBE-1.0.upf
>>> N 14 N_ONCV_PBE-1.0.upf
>>>
>>> K_POINTS automatic
>>> 3 3 1 0 0 0
>>>
>>> CELL_PARAMETERS angstrom
>>> 9.45000000000000 0.00000000000000 0.0
>>> 0.00000000000000 8.90954544295050 0.0
>>> 0.00000000000000 0.00000000000000 40.0
>>>
>>> ATOMIC_POSITIONS angstrom
>>> N 3.220157348 4.070243213 7.161850862
>>> C 2.057591707 4.250681378 7.817575738
>>> C 4.333273229 4.171146830 7.913403134
>>> C 2.009345068 4.525337360 9.193044239
>>> C 4.285082283 4.446195607 9.288766365
>>> N 3.122476937 4.626559339 9.944547282
>>> H 1.139008253 4.174089393 7.227702259
>>> H 5.290517910 4.028527369 7.402554504
>>> H 1.052120461 4.668162403 9.703968061
>>> H 5.203598139 4.523590493 9.878655120
>>>
>>>
>>> I would be grateful for any suggestions. In the meantime, we are
>>> upgrading to 6.4.1 to see if this helps.
>>>
>>> Best regards,
>>>
>>> Michal Krompiec
>>>
>>> Merck KGaA, Darmstadt, Germany & University of Southampton
>>>
>> _______________________________________________
>> Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso)
>> users mailing list users at lists.quantum-espresso.org
>> https://lists.quantum-espresso.org/mailman/listinfo/users
>
> _______________________________________________
> Quantum ESPRESSO is supported by MaX (www.max-centre.eu/quantum-espresso)
> users mailing list users at lists.quantum-espresso.org
> https://lists.quantum-espresso.org/mailman/listinfo/users
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.quantum-espresso.org/pipermail/users/attachments/20190709/37f5fdd1/attachment.html>
More information about the users
mailing list