The error message in MPID_nem_tcp_bind line 580 was incorrectly reporting the
port that was being attempted due to incrementing the variable at the end of
the for loop and not accounting for that in the error message. This patch
subtracts one if the port was not succesfully bound when printing out the
value for the error message.

Thanks to Yauheni Zelenko for reporting this error.

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

Michael Blocksome authored Aug 08, 2013 and Pavan Balaji committed Aug 08, 2013

There are three configure options and two install cases that need to be
handled.

The configure could have be done on a bgq system that has V1R2M1+
installed or on a bgq system that has a pre-V1R2M1 installation. In
V1R2M1 the location and names of the pami libraries changed.

The three configure options are:
  --with-bgq-install-dir ..... forces use of a specific bgq install
  --with-pami[-include|-lib] . forces use of a specific pami install
  PAMILIBNAME ................ forces use of a specific pami library name

Signed-off-by: Bob Cernohous <bobc@us.ibm.com>
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Increase the number of RMA operations issued within the epoch.  The
error seems to happen rarely, so increasing the number of RMA
operations increases the probability of it occurring.

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

We didn't change ABI from 3.0.x, so the 'c' value should still be 10.

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

We were jumping to exit early causing us to try to free an unitialized
string stash list.  We should really use C++ constructor-style
allocation functions for this.  This commit just moves the
initialization early, so we don't try to free garbage when we jump to
exit.

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

Pavan Balaji authored Aug 06, 2013 and Kenneth Raffenetti committed Aug 07, 2013

We need the environment-passed parameters to determine what
thread-level needs to be used.  Fixes ticket #1892.

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

Michael Blocksome authored Jul 18, 2013 and Kenneth Raffenetti committed Aug 07, 2013

Do not overload the 'mpid.next' pointer in the request as this field is
defined by the mpid layer and therefore is specific to the device.  Not
all adi implementations have this field.

Signed-off-by: Haizhu Liu <haizhu@us.ibm.com>
Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Douglas Miller authored Jun 28, 2012 and Kenneth Raffenetti committed Aug 07, 2013

This patch overloads the MPID_Request.mpid.next pointer to use for
referencing the debugger SEND queue structure to facilitate
faster removal during completion. MPID_Request.mpid.next is only
used by receives and free-pools otherwise.

(ibm) Issue 8178
(ibm) e1c4d5cfe88fccc67624f81497a74bc55d774e43

Signed-off-by: Su Huang <suhuang@us.ibm.com>
Signed-off-by: Michael Blocksome <blocksom@us.ibm.com>
Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Keira Zhang authored Dec 18, 2011 and Kenneth Raffenetti committed Aug 07, 2013

(ibm) bd92b21a52afd8f4f6ef018d764d50060966f20b

Signed-off-by: Michael Blocksome <blocksom@us.ibm.com>
Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

Allows the minimal configure ../configure --host=powerpc64-bgq-linux
--with-device=pamid --enable-thread-cs=per-object.  Without this
change the MPID_REQUEST_PREALLOC symbol would be undefined.

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

Michael Blocksome authored Aug 06, 2013 and Pavan Balaji committed Aug 06, 2013

Removed the processing of the (undocumented) environment variable
'PAMID_CORE_ON_ABORT' which was being checked to determine if the user
does *not* want the process to core dump. On Blue Gene/Q the core dump
was accomplished by calling 'abort()' which sends SIGSBRT to all
processes and all processes would then write a core file. This is not
scalable.

Instead, MPID_Abort() will invoke 'exit(1)' which will terminate all
processes in the job. This behavior is identical for both the POE and
the Blue Gene/Q control systems.

On Blue Gene/Q the user may replicate the previous core dump behavior by
using the environment variables 'BG_COREDUMPONERROR=1' or
'BG_COREDUMPONEXIT=1'.

Finally, the 'DYNAMIC_TASKING' #ifdef is moved up so it is checked first.
'MPIDI_NO_ASSERT' and 'DYNAMIC_TASKING' are typically defined for PE. It
appears that the dynamic tasking code was never being invoked.

(ibm) CPS 99YURA

Signed-off-by: Bob Cernohous <bobc@us.ibm.com>

Wesley Bland authored Aug 06, 2013 and Pavan Balaji committed Aug 06, 2013

These tests should not be distributed with the test suite when grabbing a
release. They are still enabled for developer builds.

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

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

Adds an FT test that attempts communication without touching a
failed process. Other changes are to use SIGKILL to simulate failures
and also to flush all output to stdout.

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

This update fixes the following problems:
 * Unexpected connection, mostly appears with number of ranks > 3
 * Several problems related to SCIF DMA usage
 * Intel(R) Symmetric Communication Interface (Intel(R) SCI)
   registered memory used for DMA is not unregistered yet

Current status:
 * IMB-MPI1 w/ CHECK runs well
 * MPICH tests run except spawn tests

Known issues:
 * Spawn tests still fail

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

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

most of this stuff is either broken, for ancient platofrms, or both.
Let's re-commit ourselves to using autoconf as it was intended, testing
for features, not platforms.

Fixes ticket #1862.

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

Added test to help check #1815, as weak symbol support is used to
implement the PMPI and MPI routines in a single object file, and there
are few tests for correct operation of the PMPI routines in the
test suite (a more comprehensive test of PMPI could use CPP macros to
convert all existing tests to try PMPI, but typically either the PMPI
routines are all available, or there is a problem.

This seems to fail with MPICH for >= 4096 back-to-back RMA operations.

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

Xin Zhao authored Jul 30, 2013 and Pavan Balaji committed Aug 01, 2013

shm_base_addrs is allocated only when shared memory region is allocated in nemesis.

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

Xin Zhao authored Jul 31, 2013 and Pavan Balaji committed Aug 01, 2013

Default set 'alloc_shm' info to TRUE. If 'alloc_shm' is set by user and value
is not TRUE, throw an error.

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

Xin Zhao authored Aug 01, 2013 and Pavan Balaji committed Aug 01, 2013

Assign new MPIDI_CH3U_Win_allocate to (*allocate_shared) so that it will be called by MPID_Win_allocate_shared.

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

Xin Zhao authored Jul 30, 2013 and Pavan Balaji committed Aug 01, 2013

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

Xin Zhao authored Aug 01, 2013 and Pavan Balaji committed Aug 01, 2013

In the code, some places use (void**)base_ptr and some places use (void*)base_ptr
which are confusing. We make every win_allocate related function use (void*)base_ptr
as the argument and do conversion inside the function, in order to keep the code clear.

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

Xin Zhao authored Jul 30, 2013 and Pavan Balaji committed Aug 01, 2013

Because we delete the default MPIDI_SHM_Win_free in CH3, which originally deals with
win_allocate_shared in CH3, we need to modify MPIDI_Win_free to handle it.

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

Xin Zhao authored Jul 30, 2013 and Pavan Balaji committed Aug 01, 2013

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

Xin Zhao authored Jul 30, 2013 and Pavan Balaji committed Aug 01, 2013

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

Xin Zhao authored Jul 30, 2013 and Pavan Balaji committed Aug 01, 2013

Also add prototype in header file.

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

Xin Zhao authored Aug 01, 2013 and Pavan Balaji committed Aug 01, 2013

In the code, some places use (void**)base_ptr and some places use (void*)base_ptr
which are confusing. We make every win_allocate related function use (void*)base_ptr
as the argument and do conversion inside the function, in order to keep the code clear.

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

Xin Zhao authored Jul 30, 2013 and Pavan Balaji committed Aug 01, 2013

In nemesis layer, change function name from MPIDI_CH3I_Win_allocate_shared to MPIDI_CH3I_Win_allocate_shm.

MPIDI_CH3I_Win_allocate_shm is assigned to (*allocate_shm).

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

Xin Zhao authored Jul 31, 2013 and Pavan Balaji committed Aug 01, 2013

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

Xin Zhao authored Jul 31, 2013 and Pavan Balaji committed Aug 01, 2013

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

Xin Zhao authored Jul 31, 2013 and Pavan Balaji committed Aug 01, 2013

When judging if origin and target process are on the same node, using vc->node_id flag instead of vc->ch.is_local flag.

Flag 'is_local' is not correct because it is defined in nemesis, not in CH3.
Flag 'node_id' is defined in CH3.

Note that for ch3:sock, even if origin and target are on the same node, they are not within the same SHM region.
Currently ch3:sock is filtered out by checking shm_allocated flag first. In future we need to figure out a way to
check if origin and target are within the same "SHM comm".

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

Xin Zhao authored Jul 31, 2013 and Pavan Balaji committed Aug 01, 2013

Because if MPI_WIN_FLAVOR_SHARED is used in ch3:sock, it will allocate normal memory
instead of shared memory, therefore shm_base_addrs will not be used.

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

Xin Zhao authored Jul 30, 2013 and Pavan Balaji committed Aug 01, 2013

Because CH3 layer needs to know if shared memory region is allocated in lower layer.

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