GitLab

The check was originally in the ch3 layer, but doesn't seem to use any
ch3 specific information.  This macro will be useful at the upper
layers for optimizations, e.g., in the localcopy routine.
Signed-off-by: Xin Zhao <xinzhao3@illinois.edu>

We already do a check for shared memory before calling the
shared-memory specific functions.  This patch simplifies some of those
redundant checks.
Signed-off-by: Xin Zhao <xinzhao3@illinois.edu>

Xin Zhao authored Sep 26, 2013 and Pavan Balaji committed Sep 26, 2013

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

The memory barrier ensures that all load/store operations issued
directly to shared memory are complete.
Signed-off-by: Xin Zhao <xinzhao3@illinois.edu>

Xin Zhao authored Sep 26, 2013 and Pavan Balaji committed Sep 26, 2013

Because when SHM is allocated, it is possible that orig rank and target rank
are on different nodes, in such situation operations are not done yet and
win_flush cannot exit.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Sep 26, 2013 and Pavan Balaji committed Sep 26, 2013

Check shm_allocated flag in win_flush to determine if do full
memory barrier or not.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Sep 26, 2013 and Pavan Balaji committed Sep 26, 2013

Check both shm_allocated flag and VC's node_id for
req-based operations.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

During a win_flush_all, if a target does not have any operations to
flush out, don't call the win_flush function at all.  This reduces the
number of function calls on large systems where the RMA operations are
sparsely issued.
Signed-off-by: Xin Zhao <xinzhao3@illinois.edu>

We changed the default from FALSE to TRUE for the alloc_shm info flag
for MPI_WIN_ALLOCATE.
Signed-off-by: Xin Zhao <xinzhao3@illinois.edu>

Xin Zhao authored Sep 26, 2013 and Pavan Balaji committed Sep 26, 2013

Flip the default shared memory allocation info argument for
MPI_WIN_ALLOCATE.  We were alredy doing this for
MPI_WIN_ALLOCATE_SHARED, but not for WIN_ALLOCATE.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

No reviewer.

Xin Zhao authored Sep 25, 2013 and Pavan Balaji committed Sep 26, 2013

If SHM is allocated by MPI_Win_allocate and target is on the same node
with origin, origin needs to acquire lock eagerly before it can
perform any SHM RMA operations immediately on target's SHM region.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Sep 25, 2013 and Pavan Balaji committed Sep 26, 2013

If SHM is allocated, perform RMA operations immediately after it is
issued, not queuing them up any more.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Sep 25, 2013 and Pavan Balaji committed Sep 26, 2013

Originally for SHM RMA operations, we create strcutures to queue them up and perform
them lazily when closing the epoch. Because creating queued structure causes siginificant
performance overhead, we decide to not queue them up but perform them immediately. Therefore
MPIDI_DO_SHM_OP macro and some special judgements on SHM operations (to count queued
operations) are not needed anymore.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Sep 25, 2013 and Pavan Balaji committed Sep 26, 2013

If we perform SHM ops immediately, we do not need to do the judgements
when incrementing datatypes, because SHM ops are already performed at
that point and judgements only works for remaining operations.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Sep 25, 2013 and Pavan Balaji committed Sep 26, 2013

Change the condition of full memory barrier when closing an epoch
from *judging create_flavor* to *checking if SHM is allocated*.
Because condition of *SHM is allocated* means either create_flavor
is SHARED or alloc_shm optimization is enabled for MPI_Win_allocate.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Sep 25, 2013 and Pavan Balaji committed Sep 26, 2013

We don't need the full memory barrier when opening an epoch, ordering of
modifications on the same window location can be protected by the full
memory barrier when closing the epoch. User can modify any window location
only within an RMA epoch.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Sep 25, 2013 and Pavan Balaji committed Sep 26, 2013

We don't need the full memory barrier when opening an epoch, ordering of
modifications on the same window location can be protected by the full
memory barrier when closing the epoch. User can modify any window location
only within an RMA epoch.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Sep 25, 2013 and Pavan Balaji committed Sep 26, 2013

Do a memory barrier when winow is allocated by MPI_Win_allocate_shared,
if this fence is (1) not call with MPI_MODE_NO_PROCEDE; (2) not the very
first fence; (3) not following a fence with MPI_MODE_NO_SUCCEED.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Kenneth Raffenetti authored Sep 24, 2013 and Pavan Balaji committed Sep 25, 2013

When a key is not found locally, pass the "get" command upstream.
This allows us to handle special cases like dead processes.
Fixes #1917.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Xin Zhao authored Sep 24, 2013 and Pavan Balaji committed Sep 25, 2013

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

As reported by Michael Raymond @ SGI, we were never checking if the
disp_unit was set correctly.  This patch improves the error checks.
Signed-off-by: Xin Zhao <xinzhao3@illinois.edu>

1. Cleanup check for local IPs.  We check to see if the host IP
matches that of the list of local IP addresses.  If it does, it's a
local host.

2. Do not try to detect remotely accessible nodes.  Our logic for
identifying whether a node is remotely accessible is quite primitive
and hacky.  It doesn't serve any real purpose either.  This patch
deletes that code.

3. Cleanup the is_local lookup function.  If getifaddrs and inet_ntop
are available, try to detect local processes.  If they aren't there,
simply return non-local.  In this case, the launcher will fall back to
using a remote launch even for local processes.  Bad, but workable.

4. Make the hostname lookups more comprehensive, by matching the local
hostname and the IP address the local hostname resolves to as well.
This works well when the /etc/hosts file contains an IP address match
that is not a part of the network addresses.  In this case, the
hostname IP lookup will give that IP address, but it will not match
any of the local network IP addresses.
Signed-off-by: Ken Raffenetti <raffenet@mcs.anl.gov>

In verbose mode, mpit_vars.c prints all MPI_T control variables, performance
variables and categories defined in the MPI implementation.

It is adapted from the code provided Bill Gropp.
It needs more refinement when printing hierarchical categories.

Also added mpit_vars to .gitignore.

Junchao Zhang authored Aug 14, 2013 and Pavan Balaji committed Sep 22, 2013

This commit will increase the ABI version. Fix #1904
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Junchao Zhang authored Aug 16, 2013 and Pavan Balaji committed Sep 21, 2013

Some MPI_T functions return MPI_SUCCESS even when they are not implemented.
Now return MPI_ERR_INTERN instead. Of course it is a temp fix.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

Junchao Zhang authored Aug 13, 2013 and Pavan Balaji committed Sep 21, 2013

Add missing MPI_T error codes in mpi.h. Will increase the ABI version.
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

No reviewer.

Picked up the relevant parts of [2bda6bc7], [49b74176] and [59cbc846].

No reviewer.

No reviewer.

If data sieving is not supported by the underlying file system (e.g. PVFS,
PLFS, BGLOCKLESS) then there is no reason to turn write-only requests
into read-write requests.  Suggestion first reported by David Knaak of Cray.
Reviewed-by: David Knaak <knaak@cray.com>

Kenneth Raffenetti authored Sep 18, 2013 and Pavan Balaji committed Sep 18, 2013

FreeBSD does not support this attribute. Including this runtime
test ensures nemesis will not override the allocate_shm function
pointer unless it is supported. Fixes #1926
Signed-off-by: Pavan Balaji <balaji@mcs.anl.gov>

- large_type now conforms to mpich test guidelines.
- additional datatype cases that do not rely on "large MPI_Aint" (but
  32 bit platforms probably still cannot process these types)

Fixup the f90tof77 failcheck patch to eliminate a stray blank

Add a test for success in the executions of the f77tof90 script in autogen.sh
Signed-off-by: Antonio J. Pena <apenya@mcs.anl.gov>

RobL wears the "bad commit of shame" badge for missing these in the last
commit

An complimentary test to large_count.  Whereas large_count hits an
assertion early on about size being negative, this test progresses
further into the dataloop code... where it asserts about a count of
blocks being negative:

Assertion failed in file ../src/mpid/common/datatype/dataloop/segment_count.c at line 100: *blocks_p > 0

refer to ticket #1893

RobL's large-count/large-type combo branch will address this once it
lands.

Initialized a variable in mpi_t_str.c to a garbage value so that the test will
more easily trigger MPI_T implementation bugs.

Current MPI_T implementation is incorrect, so I also added the test to xfail.