I've run this hg tree with the intel, IBM, and IMB suites on a Romley server with 20 cores for NP values of 2,3,4,...,19,20. The only failures were the IBM spawn tests (which are expected, because I ran with "--mca btl sm,self").
So -- +1 on this from me (disclaimer: I haven't looked at the code).
On Aug 5, 2012, at 6:38 AM, Gutierrez, Samuel K wrote:
> The majority of the code required to get away from explicit RML usage during
> shared memory initialization is complete. With this update, the sm BTL and sm
> mpool modex around the information required for setup.
> I've completed some initial testing and everything seems to be working
> properly, but additional testing and feedback is greatly appreciated -- and
> almost certainly required.
> Things that haven't been tested:
> o The sm BTL with progress threads enabled (OMPI_ENABLE_PROGRESS_THREADS).
> o The sm BTL with any kernel assistance enabled.
> WHY: In preparation for the eventual BTL move.
> M opal/mca/shmem/posix/shmem_posix_module.c
> M opal/mca/shmem/sysv/shmem_sysv_module.c
> M opal/mca/shmem/mmap/shmem_mmap_module.c
> M opal/mca/shmem/windows/shmem_windows_module.c
> M ompi/mca/btl/sm/btl_sm.c
> M ompi/mca/btl/sm/btl_sm_component.c
> M ompi/mca/btl/sm/help-mpi-btl-sm.txt
> M ompi/mca/btl/sm/btl_sm.h
> M ompi/mca/btl/sm/btl_sm_fifo.h
> M ompi/mca/mpool/sm/mpool_sm_component.c
> M ompi/mca/mpool/sm/mpool_sm.h
> M ompi/mca/common/sm/common_sm.c
> M ompi/mca/common/sm/common_sm.h
> The majority of changes that most will care about are in btl/sm/* and mpool/sm/*
> The branch can be found:
> TIMEOUT: If everything looks okay, Aug 10, 2012. If not, some time after issues
> have been resolved.
> devel mailing list
For corporate legal information go to: http://www.cisco.com/web/about/doing_business/legal/cri/