dorsal/dolfin 0.9.5: segmentation faults

Looks like a problem in UMFPACK, or the way we call or link to UMFPACK.

Does anyone else get the same problem when running with UMFPACK
instead of PETSc?

You could try enabling PETSc in Dorsal and see if that helps.

--
Anders

On Wed, Dec 09, 2009 at 11:22:06PM -0000, dbeacham wrote:
> New question #93523 on DOLFIN:
> https://answers.launchpad.net/dolfin/+question/93523
>
> Hi all,
>
> I've tried to install the most recent stable builds using dorsal-0.5.1 on karmic, but whenever I try and run a demo I get segmentation faults, for instance the cpp poisson demo gives:
>
> Solving linear variational problem
> Matrix of size 1089 x 1089 has 7361 nonzero entries.
> Sorting sparsity pattern.
> Applying boundary conditions to linear system.
> LU-factorizing linear system of size 1089 x 1089 (UMFPACK).
> [lappytop-dos:20579] *** Process received signal ***
> [lappytop-dos:20579] Signal: Segmentation fault (11)
> [lappytop-dos:20579] Signal code: Address not mapped (1)
> [lappytop-dos:20579] Failing at address: 0x9
> [lappytop-dos:20579] [ 0] [0x6db410]
> [lappytop-dos:20579] [ 1] ./demo(umfpack_dl_qsymbolic+0x5a) [0x808ca2a]
> [lappytop-dos:20579] [ 2] ./demo(umfpack_dl_symbolic+0x64) [0x807a624]
> [lappytop-dos:20579] [ 3] /home/dbeacham/FEM/stable/build/lib/libdolfin.so.0(_ZN6dolfin15UmfpackLUSolver7Umfpack9factorizeEv+0x83) [0x8beb83]
> [lappytop-dos:20579] [ 4] /home/dbeacham/FEM/stable/build/lib/libdolfin.so.0(_ZN6dolfin15UmfpackLUSolver9factorizeERKNS_13GenericMatrixE+0xf7) [0x8beeb7]
> [lappytop-dos:20579] [ 5] /home/dbeacham/FEM/stable/build/lib/libdolfin.so.0(_ZN6dolfin15UmfpackLUSolver5solveERKNS_13GenericMatrixERNS_13GenericVectorERKS4_+0x29) [0x8be009]
> [lappytop-dos:20579] [ 6] /home/dbeacham/FEM/stable/build/lib/libdolfin.so.0(_ZN6dolfin8LUSolver5solveERKNS_13GenericMatrixERNS_13GenericVectorERKS4_+0x15a) [0x84cb5a]
> [lappytop-dos:20579] [ 7] /home/dbeacham/FEM/stable/build/lib/libdolfin.so.0(_ZN6dolfin18VariationalProblem12solve_linearERNS_8FunctionE+0x563) [0x84b303]
> [lappytop-dos:20579] [ 8] /home/dbeacham/FEM/stable/build/lib/libdolfin.so.0(_ZN6dolfin18VariationalProblem5solveERNS_8FunctionE+0x2a) [0x84bf2a]
> [lappytop-dos:20579] [ 9] ./demo(main+0x5e1) [0x80617e1]
> [lappytop-dos:20579] [10] /lib/tls/i686/cmov/libc.so.6(__libc_start_main+0xe6) [0x1548b56]
> [lappytop-dos:20579] [11] ./demo [0x8060b01]
> [lappytop-dos:20579] *** End of error message ***
> Segmentation fault.
>
> I've tried with/without building ginac/swiginac but get the same results. Am I doing something wrong? Previous dorsal installs worked on jaunty and the fenics package (w/ dolfin 0.9.4) works on karmic.
>
> Regards, David.
>

I've compiled a copy of the development release and that works fine with UMFPACK, although a subsequent recompile of the stable source still gave the errors, so I'll just default to the dev stuff. One quick question about that: I assume that the form coefficients generated by ffc - w0, ..., wn - are named in the order that they are first encountered in the ufl file?

On Thu, Dec 10, 2009 at 09:38:59PM -0000, dbeacham wrote:
> Question #93523 on DOLFIN changed:
> https://answers.launchpad.net/dolfin/+question/93523
>
> dbeacham posted a new comment:
> I've compiled a copy of the development release and that works fine with
> UMFPACK, although a subsequent recompile of the stable source still gave
> the errors, so I'll just default to the dev stuff. One quick question
> about that: I assume that the form coefficients generated by ffc - w0,
> ..., wn - are named in the order that they are first encountered in the
> ufl file?

Yes, they should be.

--
Anders