Jeff Hammond authored Nov 09, 2013 and Rob Latham committed Nov 14, 2013

per http://stackoverflow.com/questions/2132273/what-does-malloc0-return

,
malloc(0) may or may not return NULL.  it is implementation defined in
C99.  therefore, it is incorrect to throw an OoM error on !new when
size=0.

closes #1947

Signed-off-by: Rob Latham <robl@mcs.anl.gov>

Fixes #1781

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

This should give our OSX build slaves enough time to pass.

Signed-off-by: Rob Latham <robl@mcs.anl.gov>

fix testlist

Signed-off-by: Junchao Zhang <jczhang@mcs.anl.gov>

First, this ticket is caused in the user code of ibsend.c where multiple
pthreads are created using the same pthread variable. This leads
to unexpected behaviors where detached threads screw
up with MPI_Comm_free when both of them calls ALLFUNC lock in
unexpected order, thus leads to the assertion failure in initthread.c:213.

Second, BSEND lock in the proposed patch in trac is unnecessary.
Basically the patch is to create a
new lock, BSEND, to lock the function of MPIR_Bsend_isend for thread
safety; however, MPIR_Bsend_isend is already called inside a ALLFUNC
lock both in MPI_Bsend and MPI_Ibsend.

See attached test/mpi/threads/pt2pt/ibsend.c for correct use of
pthread_create and pthread_join

Adds a test case for ibsend/bsend/isend/send in threading
execution.

Antonio J. Pena authored Nov 11, 2013 and Kenneth Raffenetti committed Nov 12, 2013

Fixes #1954

Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Added support to handle --program-suffix and --program-prefix options.

Fixes #1967.

Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Test collectives where all processes do not contribute to the
result. These tests use the MPI_Comm_group_create function to create
a new communicator excluding the failed process in order to do error
collection.

Closes #1901

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

Fixes #1965

Signed-off-by: Michael Blocksome <blocksom@us.ibm.com>

instead of assigning and then checking for overflow, check if the
overflow will happen, and then assign.

Signed-off-by: Antonio J. Pena <apenya@mcs.anl.gov>

Fixes #1971

Signed-off-by: Xin Zhao <xinzhao3@illinois.edu>

No reviewer

Changes singe long line function definitions to multiple lines

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

Adds man pages for nonblocking collectives

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

Also added comments to make it less misleading.

Signed-off-by: Antonio J. Pena <apenya@mcs.anl.gov>

Fixes #884.

Signed-off-by: Junchao Zhang <jczhang@mcs.anl.gov>

The full path of the test is test/mpi/f90/datatype/get_elem_d.f90

Signed-off-by: Junchao Zhang <jczhang@mcs.anl.gov>

Fixed a set of compiler warnings related to MPI_T code when compiled
with -Wall, including "assignment discards 'const' qualifier from pointer
target type", "unused variable", "shadows a global declaration" etc.

Signed-off-by: Antonio J. Pena <apenya@mcs.anl.gov>

Reverted file maint/version.m4 to leave MPICH_RELEASE_DATE with its default
value "unreleased development copy". The proper value is actually generated by
autogen.sh.

Drop payload into a union paired with a double, and payload will end up
double-aligned.  Can't use an anonymous union, though: that's c11.

named-union requires updating any place that used payload.  There aren't too
many, though.

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

Rob Latham authored Nov 01, 2013 and Pavan Balaji committed Nov 03, 2013

MPI_Count in a status object breaks ABI, so we have a scheme to use lo
and hi parts of adjacent struct members.  but it's not foolproof.  Test
case to demonstrate bugs and hopefully catch future regressions.

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

We were checking for (size of datatype < MPIR_COUNT_MAX / count)
earlier.  This is not a good test when the count is very large.  This
patch modifies the assertion to (count * size of datatype <
MPIR_COUNT_MAX) so we don't run into integer division issues.

Signed-off-by: Rob Latham <robl@mcs.anl.gov>

In C, left-shift has well-defined behavior, but right-shift does not
for signed integers.  The implementation is free to pad the shifted
bits with 0's or 1's.  In this patch, we type-cast the values to
unsigned integers to make the behavior well-defined.

Signed-off-by: Rob Latham <robl@mcs.anl.gov>

Antonio J. Pena authored Nov 02, 2013 and Pavan Balaji committed Nov 02, 2013

Fixes ticket #1773

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

Use if (@array) format instead of if (defined(@array)) format, since
the latter is deprecated in perl.

Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Signed-off-by: Huiwei Lu <huiweilu@mcs.anl.gov>

Signed-off-by: Huiwei Lu <huiweilu@mcs.anl.gov>

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

To cover the case when multiple communicators are created by
MPI_Comm_idup.

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

Junchao Zhang authored Oct 30, 2013 and Pavan Balaji committed Oct 31, 2013

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

Rob Latham authored Oct 28, 2013 and Pavan Balaji committed Oct 31, 2013

buildbot reports some crazy error messages:
INTERNAL ERROR: invalid error code 3e (Ring ids do not match) in PMPI_T_cvar_handle_alloc:135

valgrind more helpfully points out places where uninitialized varables
were being passed into the MPI_T interface.  Perhaps MPI_T interface
needs to be robust-ified against garbage inputs?

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

Junchao Zhang authored Oct 30, 2013 and Pavan Balaji committed Oct 31, 2013

This is a conservative setting, but should make MPICH "MPI_T-correct".
Once we know more about cvars, we can incrementally relax their scopes.

Fixes #1959

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

The test file was using MPI-I/O functions that the MPICH test suite
allows us to enable/disable.  This patch moves the test to an iodir
format, so it is not run when configured with --disable-romio.

The correct longer term fix for this is to allow multiple conditions
to disable a given test.  For example, we could have configure set the
@enable_io@ macro to either '' or '#', and different tests that rely
on I/O would add them to the start of the line in the test file.  That
way, a test that uses both F77 and I/O would be disabled when support
for either one is not present.

Signed-off-by: Junchao Zhang <jczhang@mcs.anl.gov>

Signed-off-by: Xin Zhao <xinzhao3@illinois.edu>

MPICH currently blocks in MPI_Test.  This test forces a case where the
program will deadlock if MPI_Test is not nonblocking.

Signed-off-by: Xin Zhao <xinzhao3@illinois.edu>

Apparently my copy/paste was insufficient and I missed this. Sorry about the
multi-commit.