Hi all, I've merged the BG/P branch into trunk. Assuming that the nightly tests go through fine, I think we are ready to release 1.1a1. The major feature additions for this release are: 1. Nemesis is the default communication channel and the TCP communication method is replaced by the new and better optimized newtcp module. 2. BG/P support for MPI, MPI-IO and MPE. 3. MPI-2.1 support. Is there any other major addition we should point out? There might be some more minor ones in the change logs which should be pointed out as well. The nightly tests didn't run last night as there's no space on bblogin. Once Narayan clears up the space, and we get one round of the nightly tests through, we should be able to release 1.1a1. -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji
On Aug 3, 2008, at 3:07 PM, Pavan Balaji wrote:
I've merged the BG/P branch into trunk. Assuming that the nightly tests go through fine, I think we are ready to release 1.1a1.
We still have some instability in newtcp, but I haven't been able to track down the bug yet. I can get a failure by running certain tests many iterations, but it's unclear what the root problem is yet. I suspect we won't get a clean run for ch3:nemesis:newtcp until we track it down. We can still release, but we should probably roll out a patch once we track this one down. -Dave
We still have some instability in newtcp, but I haven't been able to track down the bug yet. I can get a failure by running certain tests many iterations, but it's unclear what the root problem is yet. I suspect we won't get a clean run for ch3:nemesis:newtcp until we track it down. We can still release, but we should probably roll out a patch once we track this one down.
Yeah, even the RMA tests are not working, but that should be OK for an alpha release, IMO. Unless someone feels otherwise, we can start preparing for the release once the nightly tests have gone through without any new failures. -- Pavan -- Pavan Balaji http://www.mcs.anl.gov/~balaji
On Aug 3, 2008, at 3:07 PM, Pavan Balaji wrote:
I've merged the BG/P branch into trunk.
It looks like r1249 (the BGP merge) broke the build. It now fails to compile (for nemesis at least) with the following error: ========================================= Beginning make Using variables CC='gcc' CFLAGS='-g3 -g -Wall -O2 -Wstrict-prototypes -Wmissing-prototypes -Wundef -Wpointer-arith -Wbad-function-cast - ansi -DGCC_WALL -D_POSIX_C_SOURCE=199506L -std=c89' LDFLAGS=' ' AR='ar' FC='g77' F90='ifort' FFLAGS=' -g' F90FLAGS=' -g' CXX='c++' make all-local make[1]: Entering directory `/sandbox/mpi/src/mpich2-trunk' make[1]: Leaving directory `/sandbox/mpi/src/mpich2-trunk' make[1]: Entering directory `/sandbox/mpi/src/mpich2-trunk/src' make[2]: Entering directory `/sandbox/mpi/src/mpich2-trunk/src/mpid' make[3]: Entering directory `/sandbox/mpi/src/mpich2-trunk/src/mpid/ common' make[4]: Entering directory `/sandbox/mpi/src/mpich2-trunk/src/mpid/ common/locks' CC mpidu_process_locks.c In file included from ../../../include/mpiimpl.h:100, from /sandbox/mpi/src/mpich2-trunk/src/mpid/ch3/ include/mpidimpl.h:22, from mpidu_process_locks.h:15, from mpidu_process_locks.c:6: /sandbox/mpi/src/mpich2-trunk/src/mpid/ch3/include/mpidpre.h:187: error: expected specifier-qualifier-list before 'MPID_IOV' make[4]: *** [mpidu_process_locks.o] Error 1 make[4]: Leaving directory `/sandbox/mpi/src/mpich2-trunk/src/mpid/ common/locks' make[3]: *** [all-redirect] Error 2 make[3]: Leaving directory `/sandbox/mpi/src/mpich2-trunk/src/mpid/ common' make[2]: *** [all-redirect] Error 2 make[2]: Leaving directory `/sandbox/mpi/src/mpich2-trunk/src/mpid' make[1]: *** [all-redirect] Error 2 make[1]: Leaving directory `/sandbox/mpi/src/mpich2-trunk/src' make: *** [all-redirect] Error 2 ========================================= Why did you reorder the mpidpre.h include above the mpitypedefs.h include in mpiimpl.h? -Dave
Crud! I was waiting on Bill's comments for that, but lost track of it before committing into trunk. The reorder was part of IBM's patches -- this was required since dcmfd defines MPID_CS_ENTER/EXIT macros. Without the reordering, these definitions were duplicated above the ADI interface. This was not caught earlier, since none of our other devices define these. I'll revert that part for the time being, till I figure out the right way to do this. -- Pavan On Sun, 3 Aug 2008, Dave Goodell wrote:
On Aug 3, 2008, at 3:07 PM, Pavan Balaji wrote:
I've merged the BG/P branch into trunk.
It looks like r1249 (the BGP merge) broke the build. It now fails to compile (for nemesis at least) with the following error: ========================================= Beginning make Using variables CC='gcc' CFLAGS='-g3 -g -Wall -O2 -Wstrict-prototypes -Wmissing-prototypes -Wundef -Wpointer-arith -Wbad-function-cast - ansi -DGCC_WALL -D_POSIX_C_SOURCE=199506L -std=c89' LDFLAGS=' ' AR='ar' FC='g77' F90='ifort' FFLAGS=' -g' F90FLAGS=' -g' CXX='c++' make all-local make[1]: Entering directory `/sandbox/mpi/src/mpich2-trunk' make[1]: Leaving directory `/sandbox/mpi/src/mpich2-trunk' make[1]: Entering directory `/sandbox/mpi/src/mpich2-trunk/src' make[2]: Entering directory `/sandbox/mpi/src/mpich2-trunk/src/mpid' make[3]: Entering directory `/sandbox/mpi/src/mpich2-trunk/src/mpid/ common' make[4]: Entering directory `/sandbox/mpi/src/mpich2-trunk/src/mpid/ common/locks' CC mpidu_process_locks.c In file included from ../../../include/mpiimpl.h:100, from /sandbox/mpi/src/mpich2-trunk/src/mpid/ch3/ include/mpidimpl.h:22, from mpidu_process_locks.h:15, from mpidu_process_locks.c:6: /sandbox/mpi/src/mpich2-trunk/src/mpid/ch3/include/mpidpre.h:187: error: expected specifier-qualifier-list before 'MPID_IOV' make[4]: *** [mpidu_process_locks.o] Error 1 make[4]: Leaving directory `/sandbox/mpi/src/mpich2-trunk/src/mpid/ common/locks' make[3]: *** [all-redirect] Error 2 make[3]: Leaving directory `/sandbox/mpi/src/mpich2-trunk/src/mpid/ common' make[2]: *** [all-redirect] Error 2 make[2]: Leaving directory `/sandbox/mpi/src/mpich2-trunk/src/mpid' make[1]: *** [all-redirect] Error 2 make[1]: Leaving directory `/sandbox/mpi/src/mpich2-trunk/src' make: *** [all-redirect] Error 2 =========================================
Why did you reorder the mpidpre.h include above the mpitypedefs.h include in mpiimpl.h?
-Dave
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
Are the tests on bblogin running the trunk now? Rajeev
-----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Pavan Balaji Sent: Sunday, August 03, 2008 3:08 PM To: [email protected] Subject: [mpich2-core] BG/P status
Hi all,
I've merged the BG/P branch into trunk. Assuming that the nightly tests go through fine, I think we are ready to release 1.1a1. The major feature additions for this release are:
1. Nemesis is the default communication channel and the TCP communication method is replaced by the new and better optimized newtcp module.
2. BG/P support for MPI, MPI-IO and MPE.
3. MPI-2.1 support.
Is there any other major addition we should point out? There might be some more minor ones in the change logs which should be pointed out as well.
The nightly tests didn't run last night as there's no space on bblogin. Once Narayan clears up the space, and we get one round of the nightly tests through, we should be able to release 1.1a1.
-- Pavan
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
Yeah, all the tests are on trunk. -- Pavan On Sun, 3 Aug 2008, Rajeev Thakur wrote:
Are the tests on bblogin running the trunk now?
Rajeev
-----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Pavan Balaji Sent: Sunday, August 03, 2008 3:08 PM To: [email protected] Subject: [mpich2-core] BG/P status
Hi all,
I've merged the BG/P branch into trunk. Assuming that the nightly tests go through fine, I think we are ready to release 1.1a1. The major feature additions for this release are:
1. Nemesis is the default communication channel and the TCP communication method is replaced by the new and better optimized newtcp module.
2. BG/P support for MPI, MPI-IO and MPE.
3. MPI-2.1 support.
Is there any other major addition we should point out? There might be some more minor ones in the change logs which should be pointed out as well.
The nightly tests didn't run last night as there's no space on bblogin. Once Narayan clears up the space, and we get one round of the nightly tests through, we should be able to release 1.1a1.
-- Pavan
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
We should highlight MPI 2.1 support first. Rajeev
-----Original Message----- From: [email protected] [mailto:[email protected]] On Behalf Of Pavan Balaji Sent: Sunday, August 03, 2008 3:08 PM To: [email protected] Subject: [mpich2-core] BG/P status
Hi all,
I've merged the BG/P branch into trunk. Assuming that the nightly tests go through fine, I think we are ready to release 1.1a1. The major feature additions for this release are:
1. Nemesis is the default communication channel and the TCP communication method is replaced by the new and better optimized newtcp module.
2. BG/P support for MPI, MPI-IO and MPE.
3. MPI-2.1 support.
Is there any other major addition we should point out? There might be some more minor ones in the change logs which should be pointed out as well.
The nightly tests didn't run last night as there's no space on bblogin. Once Narayan clears up the space, and we get one round of the nightly tests through, we should be able to release 1.1a1.
-- Pavan
-- Pavan Balaji http://www.mcs.anl.gov/~balaji
participants (3)
-
Dave Goodell -
Pavan Balaji -
Rajeev Thakur