GitLab

Tags were sometimes being tested when the communication call had already
failed, leading to bad asserts later due to uninitialized values. Sometimes,
these results would have ended up with an error anyway, but sometimes not.
Signed-off-by: Huiwei Lu <huiweilu@mcs.anl.gov>

The calls in MPID_Comm_get_all_failed_procs and MPID_Comm_agree  were the
wrong macros for entering and exiting an MPID function. This corrects it.
Signed-off-by: Huiwei Lu <huiweilu@mcs.anl.gov>

Revoke will call MPIDI_CH3U_Clean_recvq to dequeue all requests with
revoked communicators. There is one missing case: when there's hierarchy
communicators that use a different context id. This patch adds a case to
check the hierarchy communicators.
Signed-off-by: Wesley Bland <wbland@anl.gov>

Simplifies the test to only use 2 processes and not make as many MPI calls.

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

This makes sure that macros work the normal way (using a semicolon at
the end, etc.) It also removes a block of unused code from
mx_cancel.c.

Modified by Wesley to split from previous patch.
Signed-off-by: Wesley Bland <wbland@anl.gov>
Signed-off-by: Pavan Balaji <balaji@anl.gov>

We were not setting the function states correctly in a bunch of
functions.

Modifications by Wesley to split up big commit.
Signed-off-by: Wesley Bland <wbland@anl.gov>
Signed-off-by: Pavan Balaji <balaji@anl.gov>

Kenneth Raffenetti authored Oct 16, 2014 and Rob Latham committed Oct 17, 2014

The default linker behavior on BGQ makes interlibrary dependencies
tricky to support correctly. Just disable them to make our lives easier.
Signed-off-by: Rob Latham <robl@mcs.anl.gov>

Pavan Balaji authored Oct 14, 2014 and Rob Latham committed Oct 17, 2014

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

Paul Coffman authored Oct 07, 2014 and Rob Latham committed Oct 17, 2014

There are no pami optimized collectives available for inTERcomms
so when MPIDI_Coll_comm_create is called it checks to see
if the comm is an inTRAcomm and if not just returns,
however it was doing the new malloc for comm->coll_fns
before the if-check which was unsetting the MPICH defaults.
The solution is to move the malloc after the inTRAcomm
if-check.
Signed-off-by: Rob Latham <robl@mcs.anl.gov>

Paul Coffman authored Oct 07, 2014 and Rob Latham committed Oct 17, 2014

The optimized pami code currently within MPIDO_Ibarrier
is incorrect - for now do not run it, instead just
kick back to MPICH if mpir_nbc is set, otherwise
call the blocking MPIR_Barrier().
Signed-off-by: Rob Latham <robl@mcs.anl.gov>

Paul Coffman authored Oct 07, 2014 and Rob Latham committed Oct 17, 2014

At the end of MPIDI_PAMI_context_init the MPIDI_Init_collsel_extension
is called to enable the dynamic optimized collective advisor.  This
is not supported on BGQ so ifdef out the call.
Signed-off-by: Rob Latham <robl@mcs.anl.gov>

Paul Coffman authored Oct 07, 2014 and Rob Latham committed Oct 17, 2014

Since Blue Gene/Q does not support dynamic tasking there was only 1
element in the MPID_VCR_t data structure so a shortcut was taken
to avoid a malloc and free of a new list of pami_task_t in a form
the pami geometry creation was expecting.  However it seems an
array of structures with 1 pami_task_t element in it is not exactly the same
in memory as an array of pami_task_t  themselves so the pami
geometry creation was failing.  The fix is to simply do what
all other platforms do and malloc a separate list of pami_task_t.
Signed-off-by: Rob Latham <robl@mcs.anl.gov>

If a message size is <= PTL_LARGE_THRESHOLD, use a single operation.
Previously, this would generate unnecessary 0-byte operations when
messages were exactly the size of the threshold.
Signed-off-by: Pavan Balaji <balaji@anl.gov>

Previously, the message pointer in an improbe call in the portals4
netmod layer was set to MPI_MESSAGE_NULL if there was no match. This
is incorrect because the ch3 layer eventually returns either a valid
pointer or NULL. As ch3 already starts the value at NULL, we can just
omit the update.
Signed-off-by: Pavan Balaji <balaji@anl.gov>

This CVAR was deprecated in MPICH 3.1 and will now be removed for MPICH 3.2.
Signed-off-by: Junchao Zhang <jczhang@mcs.anl.gov>

They are known to be failing. Mark them as xfail so they will not send
false alarms to other patches.

No reviewer.

Pavan Balaji authored Oct 14, 2014 and Sangmin Seo committed Oct 14, 2014

We need to check both the build and src directories before installing
the man and www pages.  We were only checking the build directory for
man and the src directory for www.  Also, make sure to install both
the man pages and the www pages on install.
Signed-off-by: Sangmin Seo <sseo@anl.gov>

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

Remove incorrect event from send handler assertion. PTL_EVENT_PUT should
only be seen at the target.
Signed-off-by: Pavan Balaji <balaji@anl.gov>

This reverts commit dd62e809.

This doesn't pass with mpich yet.
Signed-off-by: Wesley Bland <wbland@anl.gov>

Igor Ivanov authored Oct 10, 2014 and Pavan Balaji committed Oct 10, 2014

Memory leak can appear in case netmod usage. Comm override functions
are responsible for creating its own request. So they  need to come
before the sreq is created.
Signed-off-by: Igor Ivanov <Igor.Ivanov@itseez.com>
Signed-off-by: Pavan Balaji <balaji@anl.gov>

Igor Ivanov authored Oct 10, 2014 and Pavan Balaji committed Oct 10, 2014

Fixed wrong parameter check condition for MPI_Iallgather and MPI_Iallgatherv
-1 is valid value for sendcount in case MPI_IN_PLACE
MPI spec says:
The in place option for intracommunicators is specified by passing the value
MPI_IN_PLACE to the argument sendbuf at all processes. In such a case, sendcount and
sendtype are ignored, and the input data of each process is assumed to be in the area where
that process would receive its own contribution to the receive buffer.
Signed-off-by: Igor Ivanov <Igor.Ivanov@itseez.com>
Signed-off-by: Pavan Balaji <balaji@anl.gov>

Signed-off-by: Igor Ivanov <Igor.Ivanov@itseez.com>

In some cases, we cannot let hcoll use whatever transport it needs.
For example, ch3:sock assumes that while blocking the next event will
come over the socket channel.  If HCOLL decides to use a different
transport (such as mxm) and the next event comes on that transport,
this can result in a deadlock.  In this patch, we let the channel
specify what transports it can accept.
Signed-off-by: Devendar Bureddy <devendar@mellanox.com>

Tell users that they can disable warnings if needed.
Signed-off-by: Devendar Bureddy <devendar@mellanox.com>

Joe Ratterman authored May 02, 2011 and Pavan Balaji committed Oct 08, 2014

  If the ref-count is 1, we *should* be the only thread holding a
  reference.  It is therefore legal to skip atomics.  If that is
  insufficient justificaion, consider that there are only three things
  a concurrent thread should be doing:

    1) Reading: Not a problem, since the store is atomic.
    2) Decrementing: Illegal; it would go negative using pure atomics.
    3) Incrementing: Illegal; we are about to destroy the object.

Add #ifdef guard for use of 'aggressive counter optimizations'

(ibm) 20a558c3802fb3e138eb6b3c786a434538efffdd
Signed-off-by: Joe Ratterman <jratt@us.ibm.com>
Signed-off-by: Michael Blocksome <blocksom@us.ibm.com>

Updated by Pavan Balaji to use an OPA atomic load instead of a simple
load operation.

Fixes #1835
Signed-off-by: Pavan Balaji <balaji@anl.gov>
Signed-off-by: William Gropp <wgropp@illinois.edu>

Sangmin Seo authored Oct 07, 2014 and Pavan Balaji committed Oct 07, 2014

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

Sangmin Seo authored Oct 07, 2014 and Pavan Balaji committed Oct 07, 2014

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

Sangmin Seo authored Oct 07, 2014 and Pavan Balaji committed Oct 07, 2014

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

Pavan Balaji authored Oct 07, 2014 and Rob Latham committed Oct 07, 2014

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

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

This variable is incremented correctly by the
MPIDI_CH3U_Recvq_FDP_or_AEU routine, but the MPIDI_CH3U_Recvq_DP fails
to do so, although it also goes through the posted recv queue. For
instance, the MXM Netmod calls this routine.
Signed-off-by: Junchao Zhang <jczhang@mcs.anl.gov>

The non-blocking neighborhood collectives were not set up correctly
in pamid.

ibm (D199919)
Signed-off-by: Su Huang <suhuang@us.ibm.com>

We were only exposing a part of the allocated buffer as public
memory, but attempting to use all of it remotely.

Thanks to Charles Archer @ Intel for reporting the error and
contributing the fix.
Signed-off-by: Min Si <msi@il.is.s.u-tokyo.ac.jp>

This reverts commit 32f3a924.

Invalidate the handle of an ME that has been matched and handled in
the portals4 netmod *before* dequeuing the request. The prevents
potential double unlinking failures.
Signed-off-by: Antonio J. Pena <apenya@mcs.anl.gov>

Igor Ivanov authored Sep 05, 2014 and Pavan Balaji committed Oct 03, 2014

Previous code is based on a trick to get true context id using
invalid communicator. There is a call of MPIR_Get_contextid()
for new communicator that is not completelly built (no comm_commit
and comm_hook calls).
netmod/mxm uses comm_hook and can not work with this trick.
This changes allow to avoid call of invalid communicator using
temporary (intermediate) communicator.
Signed-off-by: Igor Ivanov <Igor.Ivanov@itseez.com>
Signed-off-by: Pavan Balaji <balaji@anl.gov>

halim amer authored Sep 05, 2014 and Kenneth Raffenetti committed Oct 03, 2014

This example does not rely on getcpu() or sched_getcpu() which are not
available on old kernels/glibc.
Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Igor Ivanov authored Oct 03, 2014 and Pavan Balaji committed Oct 03, 2014

- Fix issue in anysource match;
- Improve debug output;
Signed-off-by: Igor Ivanov <Igor.Ivanov@itseez.com>
Signed-off-by: Pavan Balaji <balaji@anl.gov>