Page 1 of 1

Sigsegv with Intel Compiler

Posted: Mon Apr 11, 2016 9:17 am
by falsafi
Hi,

I configured and made abinit with mkl and cuda, on intel cpu and nvida GPU, but the program seemed to be problematic.

With gdb, I found the place where the program emitted sigsegv:

Code: Select all

Program received signal SIGSEGV, Segmentation fault.
0x00007fffd4af8d0d in mkl_blas_avx2_xzcopy () from /opt/intel/[b]composer_xe_2015.3.187[/b]/mkl/lib/intel64/libmkl_avx2.so
Missing separate debuginfos, use: debuginfo-install xorg-x11-drv-nvidia-libs-352.79-1.el7.x86_64
(gdb) r


I would appreciate your help!

Re: Sigsegv with Intel Compiler  [SOLVED]

Posted: Mon Apr 11, 2016 9:46 am
by gmatteo
with_linalg_libs="-L/home/xxx/lib/magma/lib -lmagma -L/opt/intel/mkl/lib -lmkl_scalapack_ilp64 -lmkl_intel_ilp64 -lmkl_core -lmkl_intel_thread lmkl_blacs_intelmpi_ilp64 -lpthread -lm -ldl -liomp5 -lmkl_lapack95_ilp64"


Use the lp64 version of the mkl library instead of ilp64.
The ilp64 version assumes the 64-bit integer type whereas the default integer in Fortran is usually 4bytes.
One should compile Abinit with `ifort -i8` to be able to link against ilp64 but we don't test this option
on the test farm so, for the time being, I strongly suggest the lp64 interface.

See also https://software.intel.com/en-us/node/528682 for further details

BTW - Could you post the full backtrace given by gdb?