5.1 vs 5.2 sigsegv 11 using MPICH

Accelerated Computing HPC Compilers Legacy PGI Compilers
1

Hello,

I’ve recently installed 5.2-4 and mpich 1.2.6 on Linux x86_64. In 5.1-6 the my code executes without a problem. In 5.2-4, the code crashes with sigsegv 11. Incase you are wondering, i’ve installed two seperate versions of mpich, each compiled with their respective compiler. I was able to trace the crash to to MPI_SCATTER(). Here are the lines of code:

!-----------------------------------------------------------------------------------
call MPI_SCATTER(sgrid%sph(1:gnn_even,1), l_gnn, MPI_DOUBLE_PRECISION, &
lgrid%sph(:,1), l_gnn, MPI_DOUBLE_PRECISION, master, MPI_COMM_WORLD, ierr)
!-----------------------------------------------------------------------------------

I’ve also tried the above without a non-derived data type, i.e.

!-----------------------------------------------------------------------------------
allocate(sgrid_sph1(gnn_even),sgrid_sph2(gnn_even))
allocate(lgrid_sph1(l_gnn),lgrid_sph2(l_gnn))

lgrid_sph1 = 0.0
lgrid_sph2 = 0.0
sgrid_sph1 = sgrid%sph(1:gnn_even,1)
sgrid_sph2 = sgrid%sph(1:gnn_even,2)

call MPI_SCATTER(sgrid_sph1, l_gnn, MPI_DOUBLE_PRECISION,&
lgrid_sph1, l_gnn, MPI_DOUBLE_PRECISION, master, MPI_COMM_WORLD, ierr)
!-----------------------------------------------------------------------------------

Am I seeing an incarnation of the bug mentioned in TPR 3439 on the FAQ; its description is so vague!

Thanks,
Rob

2

Hi Rob,

TPR 3439 is probably not releated since it has to do with serial code. For you, I’d first make sure that your not encountering a stack overflow by setting your MPSTKZ environment variable to the largest allowable value for your systems.
Example:

setenv MPSTKZ 8M

If this doesn’t work, then please submit a detailed problem report to trs@pgroup.com since it would most likely be too envolved of a problem for the user forums. Of course, if any other user has encounter a similar problem, please feel free to post your experiences!

Thanks,
Mat