Change Summary for MPICH2 for week of Aug 06, 2006

Information on all weekly change summaries is available.

List of revisions

• (gropp) Added MTestPrintfMsgs for aid in debugging
• (gropp) Added a dependency on mtest.o
• (thakur) added allreduce test with MPI_INTEGER8, which fails on Windows
• (gropp) Added clean of libmpichf90.a in this directory
• (gropp) Added copyfight and ifdef...included to pmutilconf
• (thakur) added faq on CPPFLAGS
• (toonen) added missing MPI_Request_free()
• (gropp) Added missing copyright and style comments
• (gropp) Added style and copyright to generated include files
• (gropp) Added style and copyright to performance tests
• (gropp) Added the fix for VPATH from req 2706
• (robl) address a race condition when a collective write is followed by an independentwrite: make all collective writers wait until write is finished. While we'reat it, make sure all processes know an error happend if something goes wrong.
• (toonen) applied file additions and removals from trunk
• (toonen) applied file modifications from MPICH2_1_0_4p0 branch
• (gropp) Applied fix for req 2715 (copy user message into ring_msg)
• (toonen) changed the testlist file so that the I/O tests are not run if ROMIO is disabled
• (toonen) changes to allow multiple active process management module and the detection of an appropriate module for the process group associated with MPI_COMM_WORLD. the web services module now compiles but is not functional.
• (gropp) Configure with romio when building Makefile for creating the documents to get the man pages for the file routines
• (gropp) Copy user message into ring msg (fix req 2715)
• (gropp) Created separate routines for eager and rendezvous sends and switched the various mpid_xsend routines to use them instead of the copy/paste versions in the current code
• (mercierg) Fixed send
• (mercierg) Fixed send (again)
• (robl) from HEAD: address a race condition where independent I/O could start beforecollective I/O finishes
• (robl) from HEAD: messed up the pointers in the previous fix. romio tests pass now.
• (robl) from HEAD: replace an incorrect workaround for HDF5 bugs with a correct one
• (robl) from head: goofed up a pointer and a conditional.
• (robl) from head: tweak to PVFS2 noncontig-in-file detection to accomodate a commonHDF5 workload
• (mercierg) GM
• (robl) HDF5 workload
• (gropp) Improved test for VPATH problem and applied to configure files
• (robl) messed up the pointers in the previous fix. romio tests pass now.
• (robl) missed a spot where I should have dereferenced a pointer. also got logicwrong. thanks rajeev
• (gropp) Modified the tests to only print out the first few errors (set by MAX_ERRORS)
• (gropp) More cleanup to use eager and rendezvous routines
• (gropp) Move eager sync and receive support routines to further unify support of point-to-point operations in ch3
• (gropp) Move packet handlers into routines and move all of cancel send into a single file, removing the files that set cancelled to false and did nothing else
• (toonen) moved each of the one-side communication tests into an appropiate rma directory so that the tests would only be run if the implementation supported one-sided communication
• (buntinas) nemesis: fixed configure
• (buntinas) nemesis: fixed error code
• (buntinas) nemesis: more new tcp module commits
• (buntinas) nemesis: more updates to newtcp module
• (buntinas) nemesis: removed net_free_queue field from MPID_nem_mem_region_t
• (buntinas) nemesis: removed network module recv queue
• (robl) same as ad_pvfs2_write: tweak to PVFS2 noncontig-in-file detection: help usavoid expensive workaround in a common HDF5 case.
• (robl) small hdf5-specific tweak to PVFS2 noncontig-in-file detection: help us avoidexpensive workaround in a common HDF5 case.
• (robl) The previous band-aid didn't address the root cause. This band-aid does.

Files for each revision