Re: [Libmesh-devel] Bug when closing multithreaded libMesh programs?

[Roy S. <roy...@ic...>]

On Fri, 8 Feb 2008, Benjamin Kirk wrote:

> > After configuring with --with-tbb=/usr --with-tbb-lib="-ltbb
>> -ltbbmalloc" (can't the first option be autodetected and the latter
>> option be assumed, by the way?) and recompiling, I get the following
>> when running the example programs with "mpirun -np 4":
>>
>> rm_l_2_26531:  p4_error: interrupt SIGSEGV: 11
>> rm_l_2_26531: (0.651229) net_send: could not write to fd=5, errno = 32
>> rm_l_3_26602:  p4_error: interrupt SIGSEGV: 11
>> rm_l_3_26602: (0.381278) net_send: could not write to fd=5, errno = 32
>> rm_l_1_26460:  p4_error: interrupt SIGSEGV: 11
>> rm_l_1_26460: (0.976042) net_send: could not write to fd=5, errno = 32
>>
>> The segfaults are always after the program's completed, so it's not an
>> extremely urgent problem, but it's probably still something to look
>> into.  Is there something more than just deleting the Threads
>> task_scheduler that needs to be done in libMesh::close()?
>
> Is this in debug, devel, or optimized mode?

Debug, IIRC.

> I have yet to use -ltbbmalloc, by the way.

I didn't know what was needed, and it didn't autodetect when I didn't
specify -with-tbb-lib, so I threw in linker commands for both the tbb
libraries.

> I have had some issues with the perflog and reference counting, but
> I'll get that straightened out.

Worry about that first; it's even possible that this is just some
weird manifestation of that PerfLog error.
---
Roy