[OMPI users] segfault when combining OpenMPI and GotoBLAS2

[Dorian Krause]

Hi,

has any one successfully combined OpenMPI and GotoBLAS2? I'm facing 
segfaults in any program which combines the two libraries (as shared 
libs). The segmentation fault seems to occur in MPI_Init(). The gdb 
backtrace is

Program received signal SIGSEGV, Segmentation fault.
[Switching to Thread 0x424fb940 (LWP 9155)]
0x00002b2956ca9900 in opal_list_t_class () from 
/share/apps/openmpi/1.3.2/gcc-4.1.2/lib/libopen-pal.so.0
(gdb) bt
#0  0x00002b2956ca9900 in opal_list_t_class () from 
/share/apps/openmpi/1.3.2/gcc-4.1.2/lib/libopen-pal.so.0
#1  0x0000003df9c06307 in start_thread () from /lib64/libpthread.so.0
#2  0x0000003df90d1ded in clone () from /lib64/libc.so.6
#3  0x0000000000000000 in ?? ()

(For sake of completeness I have attached the test-program, you need to 
compile it with -DSIZEOF_SCAL_T=8, e.g.
mpicc -DSIZEOF_SCAL_T=8 -o blas blas.c -L$GOTO2_LIB -lgoto2 -lgfortran).

I'm working on a Opteron x86_64 system running CentOs 5.2 . GotoBLAS2 V. 
1.10 was compiled with gcc-4.1.2. This problem applies to OpenMPI 1.3.2 
and OpenMPI 1.4.0 alike (both compiled with gcc-4.1.2).

I could post the `ompi_info --all` if anyone is interested but I think 
both installations are quiet standard (openib, romio).

Thanks,
Dorian

#include <stdio.h>
#include <stdlib.h>
#include <mpi.h>

#if 4 == SIZEOF_SCAL_T
typedef float   Scal_t;
#define GEMM    sgemm_
#endif
#if 8 == SIZEOF_SCAL_T
typedef double  Scal_t;
#define GEMM    dgemm_
#endif

extern void GEMM(char *, char *, int *, int *, int *, Scal_t *, Scal_t *, int *,
                 Scal_t *, int *, Scal_t *, Scal_t *, int *);

int main(int argc, char **argv)
{
        MPI_Init(&argc, &argv);

        Scal_t A[2][3] = { { 5.0, 4.0, 2.0 }, { 1.0, -1.0, 0.0 } };
        Scal_t B[5][2] = { { 1.0, 0.0 }, { 0.0, -1.0 }, { 0.5, 1.5 },
                           { 6.0, 2.0 }, { 8.0, -2.0 } };
        Scal_t C[3][5];

        char   transa = 'N';
        char   transb = 'N';
        int    N      = 3;
        int    K      = 2;
        int    M      = 5;
        int    lda    = N;
        int    ldb    = K;
        int    ldc    = N;
        Scal_t alpha  = 1.0;
        Scal_t beta   = 0.0;

        GEMM(&transa, &transb, &N, &M, &K, &alpha, A, &lda, B, &ldb, &beta, C, 
&ldc);

        return MPI_Finalize();
}