On Nov 11, 2005, at 7:42 PM, Jonathan Day wrote:
> Still working on the Broadcom MIPS64 port, so no news
> on that front yet except to say that I'm getting
> there. (Which seems to be fairly common with people
> using Broadcom CPUs.)
Cool! Once 1.0 settles down, I plan on cleaning up the MIPS assembly
file to make it more like the other architectures (using some of our
custom perl-based macros instead of the IRIX-specific C preprocessor
code), which should help on that front. Not sure what else would
have trouble porting, but let us know.
> When enabling threads, remember that the configure
> help menu lies. --with-threads=posix should be used,
> not --with-threads=pthreads
Thanks for catching this - I'll fix the configure documentation tonight.
> I've not determined exactly what triggers it, but the
> functions in ompi/ompi/mca/pml/pml_teg_ptl.c can clash
> with functions from ompi/ompi/mca/pml/pml_uniq_ptl.c
> on compile. I'm guessing that I'm using a combination
> of compile options that aren't getting tested that
> often, if I'm the only one noticing this:
> ./configure --prefix=/usr/openmpi --enable-mpi-threads
> --enable-progress-threads --with-threads=posix
> --enable-shared --enable-static --enable-mpi-f77
> Could someone verify that this problem is repeatable
> and not simply a problem at this end?
It's caused by --enable-static --enable-shared. TEG and UNIQ are
both deprecated at this point, and I think that the solution for 1.0
will be to remove the UNIQ PML, which will solve the problem (along
with some other problems you're not even aware exist...). Should be
in the next rc.
Open MPI developer