[gpaw-users] Segmentation fault with scalapack diagonalization

[Carrie Farberow]

Dear gpaw-users,

I found that this was due to linking to the wrong blacs library.  I was incorrectly using libmkl_blacs_lp64.a, rather than libmkl_blacs_openmpi_lp64.a.

Thanks,
Carrie
----- Original Message -----
From: Carrie Farberow <farberow at wisc.edu>
Date: Wednesday, July 7, 2010 12:50 am
Subject: [gpaw-users] Segmentation fault with scalapack diagonalization
To: gpaw users <gpaw-users at listserv.fysik.dtu.dk>


> Dear gpaw-users,
>  
>  When running the scalapack diagonalization test on the GPAW 
> installation webpage:
>  
>  mpirun -np 2 gpaw-python ~/gpaw/test/CH4.py --sl_diagonalize=1,2,2
>  
>  The run crashes and I get the following segmentation fault:
>  
>  [eu095:06249] Signal: Segmentation fault (11)
>  [eu095:06249] Signal code: Address not mapped (1)
>  [eu095:06249] Failing at address: 0xf3
>  [eu095:06249] [ 0] /lib64/libpthread.so.0 [0x3e3420e7c0]
>  [eu095:06249] [ 1] 
> /home/cfarberow/opt/gpaw-0.6.5147/build/bin.linux-x86_64-2.4/gpaw-python(MPI_Comm_size+0x52) 
> [0xc530c2]
>  [eu095:06249] [ 2] 
> /home/cfarberow/opt/gpaw-0.6.5147/build/bin.linux-x86_64-2.4/gpaw-python(blacs_pinfo_+0x9d) 
> [0xc24481]
>  [eu095:06249] [ 3] 
> /home/cfarberow/opt/gpaw-0.6.5147/build/bin.linux-x86_64-2.4/gpaw-python(sl_init_+0x22) 
> [0xa10032]
>  [eu095:06249] [ 4] 
> /home/cfarberow/opt/gpaw-0.6.5147/build/bin.linux-x86_64-2.4/gpaw-python 
> [0x45fb2a]
>  [eu095:06249] [ 5] 
> /home/cfarberow/opt/gpaw-0.6.5147/build/bin.linux-x86_64-2.4/gpaw-python(PyCFunction_Call+0x110) 
> [0x4dd0f0]
>  ...
>  
>  Without the --sl_diagonalize=1,2,2 the CH4.py script produces the 
> correct results.
>  
>  In addition and possibly related, when running the testsuite the 
> following segmentation fault prints:
>  
>  [eu095:30534] Signal: Segmentation fault (11)
>  [eu095:30534] Signal code:  (128)
>  [eu095:30534] Failing at address: (nil)
>  [eu095:30534] [ 0] /lib64/libpthread.so.0 [0x3e3420e7c0]
>  [eu095:30534] [ 1] 
> /home/cfarberow/opt/gpaw-0.6.5147/build/bin.linux-x86_64-2.4/gpaw-python(mkl_lapack_zladiv+0x25) 
> [0x910add]
>  [eu095:30534] [ 2] 
> /home/cfarberow/opt/gpaw-0.6.5147/build/bin.linux-x86_64-2.4/gpaw-python(ZLADIV+0x6) 
> [0xadc85a]
>  [eu095:30534] [ 3] /usr/lib64/atlas/liblapack.so.3(zlahqr_+0x1224) [0x2aaab6c69bf4]
>  [eu095:30534] [ 4] /usr/lib64/atlas/liblapack.so.3(zhseqr_+0x8d3) [0x2aaab6c5b9a3]
>  [eu095:30534] [ 5] /usr/lib64/atlas/liblapack.so.3(zgeev_+0x695) [0x2aaab6bfc0a5]
>  [eu095:30534] [ 6] 
> /usr/lib64/python2.4/site-packages/numpy/linalg/lapack_lite.so [0x2aaab66f36c8]
>  [eu095:30534] [ 7] 
> /home/cfarberow/opt/gpaw-0.6.5147/build/bin.linux-x86_64-2.4/gpaw-python(PyCFunction_Call+0x110) 
> [0x4dd0f0]
>  [eu095:30534] [ 8] 
> /home/cfarberow/opt/gpaw-0.6.5147/build/bin.linux-x86_64-2.4/gpaw-python 
> [0x54c7b0]
>  ...
>  
>  Simple scripts for slab calculations don't throw any errors, but I'd 
> like to do some scaling tests for large clusters and its my 
> understanding that this is best with scalapack enabled.  I've tried 
> multiple releases of GPAW and continue to get the same errors.  I'm 
> currently working with gpaw-0.6.5147.tar.gz.  I'm using Intel_MKL 
> (which I've also tried multiple versions of - currently using 
> 10.0.4.023).  I'm also using Python 2.4.4, numpy-1.2.1 and openmpi-1.4.1.
>  
>  Any recommendations would be greatly appreciated.
>  
>  Best regards,
>  Carrie
>  _______________________________________________
>  gpaw-users mailing list
>  gpaw-users at listserv.fysik.dtu.dk
>  https://listserv.fysik.dtu.dk/mailman/listinfo/gpaw-users
>