[gmx-users] Gromacs 3.3.3 segmentation fault on SLES10.2 x86_64 Xeon architecture - Intel Compilers & Intel MKL

[David van der Spoel]

On 5/28/10 11:08 AM, Juan Alejandro Palomino Benito wrote:
> Hi guys.
>
> I have installed Gromacs 3.3.3 from source code on a machine with
> SLES10.2 x86_64 Xeon architecture - Intel Compilers&  Intel MKL. It
> was configured like this:
>
> export CC=mpiicc
> export  CXX=mpiicpc
> export F77=mpiifort
> export FC=mpiifort
> export MPICC=mpiicc
>
> export CFLAGS="-m64 -O2 -I$MKLROOT/include"
> export LDFLAGS="-L$MKLROOT/lib/em64t"
> export  LIBS="-lm -lmkl_lapack -lmkl_intel_lp64 -lmkl_sequential
> -lmkl_core -lpthread"
>
> export  CPPFLAGS="-m64  -O2"
> export  FFLAGS="-m64  -O2"
> export  CXXFLAGS="-m64  -O2"
>
> ./configure  \
>                --prefix=/app/gromacs/x86_64/3.3.3_s/  \
>                --with-fft=mkl \
>                --disable-software-sqrt \
>                --enable-prefetch-forces  \
>                --enable-mpi  \
>                --with-external-lapack \
>                --with-external-blas
>
> and properly modified configure file in order to fix the MKL problem.
>
> The problem is that I receive a segmentation fault after starting mdrun.
> I attach a backtrace calls for debuging. It seems there is a problem
> with Intel MKL, but I had not this problem with latest Gromacs
> versions (4.0.5 and 4.0.7, single and double precision):
>
>
> Back Off! I just backed up
> /scratch/jpalomino/gromacs/benzocaina/345K/bzca_2_dpps_24_dppc_48_na_24_345K_0001.pdo
> to /scratch/jpalomino/gromacs/benzocaina/345K/#bzca_2_dpps_24_dppc_48_na_24_345K_0001.pdo.21#
> starting mdrun '2 BZCA + 54 DPPS +12 DPPS + 12 Na + 2477 H2O'
> 10000000 steps,  40000.0 ps.
>
>
> Program received signal SIGSEGV, Segmentation fault.
> [Switching to Thread 47952891846560 (LWP 23723)]
> 0x00002b9ce79201e6 in LIncAny_Loop4gas_1 () from
> /opt/intel/mkl/10.2.2.025/lib/em64t/libmkl_mc.so
> (gdb) bt
> #0  0x00002b9ce79201e6 in LIncAny_Loop4gas_1 () from
> /opt/intel/mkl/10.2.2.025/lib/em64t/libmkl_mc.so
> #1  0x00002b9ce51c08cc in _2__STRING.0.0 () from
> /opt/intel/mkl/10.2.2.025/lib/em64t/libmkl_lapack.so
> #2  0x00007fffc66a7ac8 in ?? ()
> #3  0x000000000093f2a0 in ?? ()
> #4  0x00002ba0e891f020 in ?? ()
> #5  0x00007fffc66a7ac8 in ?? ()
> #6  0x00002b9ce47ff183 in mkl_blas_xscopy () from
> /opt/intel/mkl/10.2.2.025/lib/em64t/libmkl_lapack.so
> #7  0x00002b9ce8425cd0 in mkl_dft_xsforward_out () from
> /opt/intel/mkl/10.2.2.025/lib/em64t/libmkl_mc.so
> #8  0x00002b9ce59c88f4 in mkl_dft_xsforward_out () from
> /opt/intel/mkl/10.2.2.025/lib/em64t/libmkl_sequential.so
> #9  0x00002b9ce832514f in mkl_dft_compute_fwd_s_r2c_1d_o ()
>     from /opt/intel/mkl/10.2.2.025/lib/em64t/libmkl_mc.so
> #10 0x00002b9ce558260f in DftiComputeForward_1 () from
> /opt/intel/mkl/10.2.2.025/lib/em64t/libmkl_intel_lp64.so
> #11 0x000000000047f356 in gmx_fft_3d_real ()
> #12 0x0000000000456bd7 in gmxfft3D ()
> #13 0x0000000000450c7a in do_pme ()
> #14 0x000000000042f000 in force ()
> #15 0x0000000000461f46 in do_force ()
> #16 0x00000000004153a7 in do_md ()
> #17 0x0000000000413875 in mdrunner ()
> #18 0x0000000000416d31 in main ()
>
>
> Than you in advance.
>
If you really need gromacs 3.3.3 then the simple workaround is to use 
FFTW3. MKL is only a few % faster since FFT is not the bottleneck in 
gromacs.

-- 
David van der Spoel, Ph.D., Professor of Biology
Dept. of Cell & Molec. Biol., Uppsala University.
Box 596, 75124 Uppsala, Sweden. Phone:	+46184714205. Fax: +4618511755.
spoel at xray.bmc.uu.se	spoel at gromacs.org   http://folding.bmc.uu.se