Correct. The per-object code was experimental and for research purposes (we only implemented it in ch3:sock and never fully ported it to ch3:nemesis). Later IBM picked it up for their pamid device, where it is well supported. — Pavan On Jun 24, 2014, at 11:31 PM, Lu, Huiwei <[email protected]> wrote:

I can reproduce your bug on my Linux desktop with GCC, too. I guess the ‘per-object’ option is not well supported. We may need to create a new ticket for this issue.

— Huiwei

On Jun 24, 2014, at 11:21 PM, Halim <[email protected]> wrote:

...

Hi,

I am facing an error when building MPICH 3.1.1 release with per-object critical sections enabled.

I am getting the following errors related to multiple definition of "MPIU_Thread_CS_enter_lockname_impl_" :

..... GEN lib/libmpi.la src/mpi/attr/.libs/lib_libmpi_la-comm_delete_attr.o: In function `MPIU_Thread_CS_enter_lockname_impl_': src/mpi/attr/comm_delete_attr.c:(.text+0x530): multiple definition of `MPIU_Thread_CS_enter_lockname_impl_' src/mpi/attr/.libs/lib_libmpi_la-comm_create_keyval.o:src/mpi/attr/comm_create_keyval.c:(.text+0x290): first defined here src/mpi/attr/.libs/lib_libmpi_la-comm_delete_attr.o: In function `MPIU_Thread_CS_exit_lockname_impl_': src/mpi/attr/comm_delete_attr.c:(.text+0x570): multiple definition of `MPIU_Thread_CS_exit_lockname_impl_' src/mpi/attr/.libs/lib_libmpi_la-comm_create_keyval.o:src/mpi/attr/comm_create_keyval.c:(.text+0x2d0): first defined here .....

This is the configure line: ./configure --disable-fortran --disable-cxx --enable-thread-cs=per-object

I am testing on x86_64 linux desktop with Intel compilers and device/channel: ch3/nemesis.

Is there a quick fix to this issue?

Regards, Halim

_______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss

_______________________________________________ discuss mailing list [email protected] To manage subscription options or unsubscribe: https://lists.mpich.org/mailman/listinfo/discuss