GitLab

Wesley Bland authored May 31, 2013 and Pavan Balaji committed Jul 25, 2013

The highest bit of tag space will now be reserved for failure
notification. We are far above the required 32767, so this shouldn't be
a problem.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Jul 24, 2013 and Pavan Balaji committed Jul 24, 2013

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

This parameter gets allocated and broadcasted but near as I can tell
it's never actually used for anything.

Antonio J. Pena authored Jul 22, 2013 and Kenneth Raffenetti committed Jul 22, 2013

Ken Raffenetti reported the default GNU patch for FreeBSD is too old to
recognise the --no-backup-if-mismatch flag. This fix switches to a manual
removal approach.
Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Antonio J. Pena authored Jul 19, 2013 and Pavan Balaji committed Jul 19, 2013

Cleaned up spaces-tabs mismatch in patch file, and fixed a few parameters for
the patch command.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Jul 19, 2013 and Pavan Balaji committed Jul 19, 2013

In MPIDI_CH3I_Win_allocate_shared, if the size of shared memory
segment is 0, we just set window's base to NULL, do not allocate
shared memory segment and mutex memory region.

In MPIDI_CH3I_SHM_Win_free, we detach the shared memory segment
and mutex memory region only when the size of shared memory segment
is not 0.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Jul 19, 2013 and Pavan Balaji committed Jul 19, 2013

MPICH currently has a bug that causes it to abort when all processes
provide zero bytes to the window allocation routine.  While it can be
considered a user error, the MPI-3 standard does not explicitly
disallow this case.

Jeff Hammond authored May 29, 2013 and Pavan Balaji committed Jul 19, 2013

MPIR_ERRTEST_ARGNULL is used in MPI_Get_accumulate and
MPI_Rget_accumulate to verify that the address of a buffer is not
NULL.  However, a NULL buffer is valid when the count is zero or when
a derived datatype of size 0 is used.
Signed-off-by: Jeff Hammond <jhammond@alcf.anl.gov>

strict builds no longer do gcc-specific things.  We test each option
to see if the compiler accepts them before using it.

This program tests to make sure we can do zero-count or zero-size
datatype RMA tests without the MPI library throwing errors.

Xin Zhao authored Jul 18, 2013 and Pavan Balaji committed Jul 18, 2013

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Jul 18, 2013 and Pavan Balaji committed Jul 18, 2013

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Jul 18, 2013 and Pavan Balaji committed Jul 18, 2013

Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Antonio J. Pena authored Jul 16, 2013 and Pavan Balaji committed Jul 18, 2013

Patching libtool.m4 from autogen.sh to add compatibility with the NAG
Fortran compiler plus shared libraries.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

This covers a small part of ticket #1835.

Since different file systems have different default hints, make a start
at being a bit more sophisticated with respect to what we should compare
against.

If no arguments given, file_info will simply test that hints set can be
retrieved.

If given the -b (bluegene) or -u (unix file system) flag, file_info will
check those default values.

todo: add pvfs, lustre, and other file systems of interest.

fixes tt #1810

Su Huang authored Jul 09, 2013 and Pavan Balaji committed Jul 16, 2013

(ibm) D191388 : MPICOM case with mpich2 hit Assertion failed issue
(ibm) AIX71J
Signed-off-by: Charles Archer <archerc@us.ibm.com>

Michael Blocksome authored Jul 15, 2013 and Pavan Balaji committed Jul 16, 2013

Signed-off-by: Haizhu Liu <haizhu@us.ibm.com>

Michael Blocksome authored Jul 15, 2013 and Pavan Balaji committed Jul 16, 2013

Signed-off-by: Haizhu Liu <haizhu@us.ibm.com>

Fixes #1695
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

We were using MPIR_PARAM_NEMESIS_ENABLE_CKPOINT in some cases and
MPIR_PARAM_ENABLE_CKPOINT in some cases.  This patch makes the names
consistent.
Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

It would appear that we can get more than two gigs worth of data down to
the write(2) and read(2) system call, but those syscalls fail to process
all of it.  Perfectly valid behavior for unix to return less than
requested, but until now, ROMIO never dealt with it.

Still to do: update Blue Gene drivers

closes ticket #1742  , since that ticket is only asking about I/O to
less than two gigs.  Lots and Lots more work needed for larger than two
gigs.

This environment is unused, since we never got it to work.  The
original idea was to push the configuration for all compilers into a
single file.  This never worked correctly, so it was disabled, but
never deleted.  We don't fall back to the original config.rpath which
defaults to an acl_cv_ prefix since that's very buggy.  The variables
rely on not having that prefix.

Signed-off-by: Wesley Bland <wbland@mcs.anl.gov>

missed a free of an unused array in one case, and freed some arrays
immediately after calling MPI_Alltoallv in an attempt to chip away at
memory footprint.

Replace "-" with "\-", so the manpage shows a minus instead of a
hyphen.  Thanks to Torquil Macdonald Sørensen <torquil@gmail.com> for
providing the initial patch.
Signed-off-by: Torquil Macdonald Sørensen <torquil@gmail.com>
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

This was added for terrible systems that rely on a slow DNS server to
look up the local hostname.  Most OSs do a much better job setting
local host names in the local /etc/hosts file today.  Even if they
don't, we no longer do a large number of lookups anymore, so this
should not be a problem.

We got a number of complaints with the recent move to prioritize clang
over gcc.  The biggest one was the lack of good OpenMP support in
clang.

http://clang-developers.42468.n3.nabble.com/RFC-OpenMP-Support-in-Clang-td4029194.html

Bill found some misuse of MPIR_ERR_RECOVERABLE and few other error
messages and wrote a test in autogen.sh to discover/report them. This
fixes the errors raised by that test and re-enables it at autogen time.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Addresses #1887.