In this patch, we replace one argument of function
finish_op_on_target, "packet(op) type", with "has_response_data".
Since finish_op_on_target does not care what specific
packet(op) type it is processing on, but only cares
about if the current op has response data (like GET/GACC),
changing the argument in this way can simplify the
code by avoiding acquiring packet(op) type everytime
before calling finish_op_on_target.

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

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

Originally we add "immed_data" and "immed_len" areas to RMA packets,
in order to piggyback small amount of data with packet header to
reduce number of packets (Note that "immed_len" is necessary when
the piggybacked data is not the entire data). However, those areas
potentially increase the packet union size and worsen the two-sided
communication. This patch fixes this issue.

In this patch, we remove "immed_data" and "immed_len" from normal
"MPIDI_CH3_Pkt_XXX_t" operation type (e.g. MPIDI_CH3_Pkt_put_t), and
we introduce new "MPIDI_CH3_Pkt_XXX_immed_t" packt type for each
operation (e.g. MPIDI_CH3_Pkt_put_immed_t).

"MPIDI_CH3_Pkt_XXX_immed_t" is used when (1) both origin and target
are basic datatypes, AND, (2) the data to be sent can be entirely fit
into the header. By doing this, "MPIDI_CH3_Pkt_XXX_immed_t" needs
"immed_data" area but can drop "immed_len" area. Also, since it only
works with basic target datatype, it can drop "dataloop_size" area
as well. All operations that do not satisfy (1) or (2) will use
normal "MPIDI_CH3_Pkt_XXX_t" type.

Originally we always piggyback FOP data into the packet header,
which makes the packet size too large. In this patch we split the
FOP operaton into IMMED packets and normal packets.

Because CAS only work with 2 basic datatype and non-complex
elements, the data amount is relatively small, we always piggyback
the data with packet header and only use "MPIDI_CH3_Pkt_XXX_immed_t"
packet type for CAS.

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

This patch just does code refactoring for RMA operation rountines
to make the code structure clearer. This patch does not change any
functionality.

After code refactoring, in each operation routine, for non-SHM operations
we do the work in the following order:

(1) allocate a new op struct;
(2) fill areas in op struct, except for packet struct in op struct;
(3) initialize packet struct in op struct, fill areas in packet struct;
(4) enqueue op to data structure on window.

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

Originally we added lock_type and origin_rank areas
in RMA packet, in order to piggyback passive lock request
with RMA operations. However, those areas potentially
enlarged the packet union size, and actually they are
not necessary and can be completetly avoided.

"Lock_type" is used to remember what types of lock (shared or
exclusive) the origin wants to acquire on the target. To remove
it from RMA packet, we use flags (already exists in RMA packet)
to remember such information.

"Origin_rank" is used to remember which origin has sent lock
request to the target, so that when the lock is granted to this
origin later, the target can send ack to that origin. Actually
the target does not need to store origin_rank but can only store
origin_vc, which is known from progress engine on target side.
Therefore, we can completely remove origin_rank from RMA packet.

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

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

Some packet wrappers did not include all packet types,
this patch adds missed packet types to those wrappers.

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

This patch re-apply modifications on mpidpkt.h that is
temporarily reverted in bb3f9623

.

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

This reverts commit 389aab16

.

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

We are going to revert the commit 389aab16 because it re-ordered
the attributes in RMA packet structs in mpidpkt.h and messed up
the alignments.

This commit temporarily reverts the following commits, which
only reverts modification on mpidpkt.h after commit 389aab16.

e36203c3, 45afd1fd, 3a05784f, 87acbbbe, b155e7e0

We will re-apply those modifications after we revert 389aab16

.

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

Two libtool patches we were carrying were released in version 2.4.3.
This commit drops our patches and bumps version required to run
autogen.sh.

Our patch for ifort on OSX is still present and updated to work with
the new version.

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

Fixes a potential "arg list too long" error at make time. See
https://lists.gnu.org/archive/html/bug-automake/2014-10/msg00009.html


for more info. Closes #2215

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

Call of MPID_Sched_cb callback function can force list memory
reallocation. As a result entry point proccessed before call
can become invalid. It should be set again after callback call.

Signed-off-by: Devendar Bureddy <devendar@mellanox.com>
Signed-off-by: Igor Ivanov <Igor.Ivanov@itseez.com>
Signed-off-by: Wesley Bland <wbland@anl.gov>

The entire "read-modify-write" should be atomic for CAS, FOP and
GACC operations. This patch adds corresponding tests for them.

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

FOP, CAS and GACC are atomic "read-modify-write" operations,
which means when the target window is defined on a SHM region,
we need inter-process lock to guarantee the atomicity of the
entire "read+OP". The current implementation is correct for
SHM-based RMA operations, but not correct for AM-based RMA
operations: for SHM-based operations, it protects the entire
"read+OP", but for AM-based operations, it only protects the
"OP" part.

This patch fixes this issue by protecting the memory copy to
temporary buffer and computation together for AM-based operations.

Fix ticket 2226

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

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

In commit 7d71278, if node_comm is NULL (only self process is on that
node), we call allocate_no_shm() in CH3 to allocate window. If
node_comm is not NULL (more than one process is on the same node), we
call allocate_shm() in Nemesis to allocate SHM window. However,
the exchanged information amount (in MPI_Allgather) is different
in allocate_no_shm() and allocate_shm(), which leads to wrong execution
when both SHM window and non-SHM window exist. This patch fixes this issue.

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

We allocate / free SHM regions only when node_comm exists,
which means there are more than one processes on the same
node. When node_comm is NULL (only self process is on that
node), we call default allocate / free functions in CH3.
(Please refer to commit f02eed5b

)

Here we delete unnecessary code dealing with node_comm being
NULL in SHM allocate / free functions.

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

code of MPIU_Strncpy. Added test program.
Closes #2225

Signed-off-by: William Gropp <wgropp@illinois.edu>

Some of the MPIX functions did not have weak symbols set up correctly
which causes problems on some compilers (Pathscale). This patch adds the
correct attribute for all of them that were missing.

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

Merges the existing send callbacks into a single function. Uses the
completion counter to track remaining operations and complete the
request once finished.

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

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

Rajeev Thakur authored Jan 27, 2015 and Rob Latham committed Feb 03, 2015

set by type_create_resized are not sticky.

Changes darray and subarray types to use type_create_resized instead
of type_struct with explicit lb/ub, because explicit MPI_LB/MPI_UB
have been removed from MPI in MPI-3 and they also cause other problems
because they were defined to be sticky in MPI-1.

Fixes type_create_struct, which was incorrectly setting lb and ub to
true_lb and true_ub in the non-sticky case.

Closes #2218
Closes #2220
Closes #2224

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

The errflag in the request object is used in common code and should not
have been put in device specific code. This moves it up to the
MPI_Request object.

Signed-off-by: Sameh Sharkawi <sssharka@us.ibm.com>

This include was present but commented out. Normally, it wasn't needed,
but to pick up the definition of MPIX_ERR_PROC_FAILED correctly, it
needs to be there.

No reviewer

The previous commits didn't take into account empty requests when
extracting the status. It also introduced a dumb bug that didn't get
tested first about a null pointer check.

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

The error code set in the status was being ignored for NBC and one-sided
requests (which wasn't right anyway so it didn't matter). This grabs the
error code from the status now.

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

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

The scheduler functions now use MPIC_* functions to handle their
communication instead of directly calling the MPID_* functions. This
helps to simplify code related to error handling and allows the
collectives to complete even if a failure is detected because the error
will be tracked via the errflag inside the request object.

Fixes #2222

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

Non-blocking communication requests need a way to track whether an error
has occurred in a previous part of the NBC schedule. This adds an
errflag to the request object itself so the tracking is possible.

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

Having mpir_errflag_t defined in mpiimpl.h causes a problem if it needs
to be used in some other headers. This moves the definition to
mpitypedefs.h so it can be used elsewhere.

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

Part of converting the NBC code to use the MPIC_* functions requires
an MPIC_Issend function to exist. This adds it.

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

The MPIC helper functions have been using MPI_Comm and MPI_Request
objects instead of their MPID_* counterparts. This leads to a bunch of
unnecessary conversions back and forth between the two types of objects
and makes the work incompatible with other parts of the codebase
(non-blocking collectives for instance).

This patch converts all of the MPIC_* functions to use MPID_Comm and
MPID_Request and changes all of the collective calls to use them now
too.

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

The collective helper functions generally have an errflag that is used
when a failure is detected to allow the collective to continue while
also communicating that a failure occurred. That flag is now included
as a parameter for MPIC_Wait.

The rest of this commit is the refactoring necessary in the rest of the
helper functions to support the change.

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

Signed-off-by: Charles J Archer <charles.j.archer@intel.com>

The tag for send was ignored and recvtag incorrectly used in its place.

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

Rob Latham authored Jan 20, 2015 and Kenneth Raffenetti committed Jan 23, 2015

HDF5 folks reported a bug with ROMIO and one of their slightly-strange (but
100% legal) datatypes.  git-bisect points to the "promote size of length"
change.  Seems that MPICH does not like struct datatypes with zero-count
elements?  Further investigation requred.  This change (construct a simpler
datatype in more cases) is sufficient to help HDF5 move forward.

See #2221

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

Rob Latham authored Jan 19, 2015 and Kenneth Raffenetti committed Jan 23, 2015

many many places where a 64 bit value is stored in a 32 bit value

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

Rob Latham authored Jan 19, 2015 and Kenneth Raffenetti committed Jan 23, 2015

- bump up subtypes from 3 to 6. The limit is arbitrary.  I am trying to
  figure out a type with 4 sub-types.
- split up indexed/hindexed lists onto separate lines.  MPICH debug
  output format adds its own newlines, but we have to clean out MPICH's
  extra debug output anyway: joining a few lines isn't that much more
  work.
- output a name of the digraph that graphviz can actually parse.

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

When process fails, fault tolerance scheme takes a different path to
deal with MPI object reference counts than the existing one. Some
reference counts were not properly set in FT path so when configured
with --enable-g=all, some ft tests will show leaked context id, dirty
COMM, GROUP and REQUEST objects and so on when exit.

This patch fixes ft/shrink and ft/agree with "--enable-g=all". Stack
allocated objects of requests, communicators and groups will be freed by
FT.

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