Page 1 of 1

BUG: about to overwrite gbound in sphereboundary.F90

Posted: Wed May 04, 2016 2:03 pm
by recohen
I am getting:

--- !BUG
message: |
iloop,igb,mgb= 2 69 68
about to overwrite gbound.
src_file: sphereboundary.F90
src_line: 251
...


and
Image PC Routine Line Source
libifcore.so.5 00002B193ABE0CC9 Unknown Unknown Unknown
abinit 00000000022B63B3 Unknown Unknown Unknown
abinit 0000000001926106 Unknown Unknown Unknown
abinit 00000000006E8788 Unknown Unknown Unknown
abinit 00000000005FE760 Unknown Unknown Unknown
abinit 00000000005A81BF Unknown Unknown Unknown
abinit 000000000059DF1D Unknown Unknown Unknown
abinit 0000000000590E14 Unknown Unknown Unknown
abinit 000000000053EFA9 Unknown Unknown Unknown
abinit 000000000042E1CB Unknown Unknown Unknown
abinit 00000000004151C8 Unknown Unknown Unknown
abinit 000000000040B403 Unknown Unknown Unknown

The input runs fine under abinit 7.2, but not in 7.4 or later. The input is available to someone who can help. I don't want to post it generally. It is a 10 atom relaxation calculation of an insulator using PAWs--nothing unusual.

Re: BUG: about to overwrite gbound in sphereboundary.F90

Posted: Fri May 06, 2016 8:06 pm
by recohen
I don't understand it, but I changed:

paral_kgb 1 npband 1 npfft 1 npkpt 28
to
paral_kgb 1 npband 1 npfft 4 npkpt 14

and now it works (on more processors too). I don't understand this. I have no fft keywords in my input so it should be choosing the correct set. Before this
I tried dialing in nfftg by hand by any numbers I put in gave the same failure. My input without structure info is:

paral_kgb 1 npband 1 npfft 4 npkpt 14
ecutsm 0.5
dilatmx 1.1
znucl 82 22 8
ngkpt 6 6 6
shiftk 0.5 0.5 0.5
occopt 6
#tsmear 0.005
pawecutdg 56
ecut 40
ixc 23
nstep 100
kptopt 1
toldfe 1.0d-8
iscf 17
npulayit 3
pawmixdg 1
iprcel 45
ionmov 2
optcell 2
ntime 50
strtarget 3*-0.002719143138176 3*0.0
restartxf=-2

segfault crash in pawgrnl

Posted: Wed May 11, 2016 2:06 am
by recohen
This problem continues. It seems it is crashing in pawgrnl with a segfault when i run on a different machine. Any suggestions?

Re: BUG: about to overwrite gbound in sphereboundary.F90

Posted: Tue Jun 21, 2016 5:06 pm
by gabriel.antonius
The problem comes from the fft grid being recomputed as the cell parameters change during optimization. The specific values for your unit cell and ecut are pathological for some reason. I don't know how to fix it completely, but there is a way around.

Try increasing the value of ecut. Since ecut dictates the size of the fft grid, by playing with this variables, you might just find a value that avoids the problem.