Hi,
Sorry for the delay getting back do you on this. I think I managed to
solve some of my issues but haven't quite figured it out exactly. I think
this is at least a different issue than before.
Previously, I had installed MPI with Homebrew which was apparently
installing it in some way libMesh/PETSc couldn't understand. When I
installed it by hand I got both PETSc and libMesh to install without having
to disable MPI. I also ran make check on libMesh and all of the tests have
passed.
However, I still get a runtime error when I initialize libMesh. Here is
the code:
#include "libmesh/libmesh.h"
int main(int argc, char** argv) {
libMesh::LibMeshInit init(argc,argv);
return 0;
}
and I get the (runtime) error:
[Andrews-MacBook-Pro.local:10106] [[INVALID],INVALID] ORTE_ERROR_LOG:
Unknown error: -1 in file runtime/orte_globals.c at line 218
[Andrews-MacBook-Pro:10106] *** Process received signal ***
[Andrews-MacBook-Pro:10106] Signal: Segmentation fault: 11 (11)
[Andrews-MacBook-Pro:10106] Signal code: Address not mapped (1)
[Andrews-MacBook-Pro:10106] Failing at address: 0x0
[Andrews-MacBook-Pro:10106] [ 0] 2 libsystem_c.dylib
0x00007fff972f894a _sigtramp + 26
[Andrews-MacBook-Pro:10106] [ 1] 3 ???
0x0000000000000001 0x0 + 1
[Andrews-MacBook-Pro:10106] *** End of error message ***
Segmentation fault: 11
Do you know what could be causing this? The error message means very
little to me. I have attached the configure log for libMesh.
Thanks again!
Andy
On Thu, Jun 6, 2013 at 4:55 PM, Jed Brown <[email protected]> wrote:
> Andrew Davis <[email protected]> writes:
>
> > This does look like some kind of MPI issue. Just to be safe I tried
> > reinstalling MPI and then reconfigure/compiling PETSc and libMesh.
>
> Exactly which options are you configuring with? This message makes it
> sound like you have an MPI-enabled PETSc and a libMesh configured
> without MPI.
>
> > This resulted in the same runtime error (the stack trace) as before.
>
> Which one? In MPI_Attr_get? Can you attach a debugger and find out
> what out what the value of comm was? MPIUNI's implementation is simple:
>
> int MPI_Attr_get(MPI_Comm comm,int keyval,void *attribute_val,int *flag)
> {
> if (comm-1 < 0 || comm-1 > 3) return 1;
> if (!keyval) Keyval_setup();
> *flag = attr[comm-1][keyval].active;
> *(void**)attribute_val = attr[comm-1][keyval].attribute_val;
> return MPI_SUCCESS;
> }
>
> It appears that libMesh::COMM_WORLD is being set to 0 when not using
> MPI?
>
------------------------------------------------------------------------------
This SF.net email is sponsored by Windows:
Build for Windows Store.
http://p.sf.net/sfu/windows-dev2dev
_______________________________________________
Libmesh-users mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/libmesh-users
