Paul Coffman authored May 11, 2014 and Rob Latham committed Jul 14, 2014

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

Paul Coffman authored Apr 04, 2014 and Rob Latham committed Jul 14, 2014

The MP_IOAGENT_CNT is another environment variable to control I/O
aggregation selection. Teach ROMIO about this to ease PE->ROMIO transistion.

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

Paul Coffman authored Mar 31, 2014 and Rob Latham committed Jul 14, 2014

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

Paul Coffman authored Mar 30, 2014 and Rob Latham committed Jul 14, 2014

We extend gpfs filesystem to PE.  Right now, the main accomodation is
for ROMIO aggregation selection using MP_IOTASKLIST to mimic current PE
functionality

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

On FreeBSD and Darwin, the read and write system calls think 2147483648 (2^31:
larger than a signed 32 bit integer can hold) is an invalid amount of data.
Maybe, despite taking a size_t parameter, something internally treats the count
as an int.  Fine, we'll take measures to never write more than 2^31-1 bytes.

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

Mohamad Chaarawi authored Jul 11, 2014 and Rob Latham committed Jul 11, 2014

Commit [e527fa1d

] missed an update to an assertion in the read path: with
bufcount promoted to MPI_Count, making sure it does not overflow an
'unsigned' is incorrect.

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

Testcase for "RMA fails with derived type containing struct of struct"

See #2115

declare "handle_error" with a const string to make compiler happy

this optimization in ROMIO's noncontigous type handling detects if a
user-defined type will fit into a single write request (e.g. a call to
MPI_Type_create_subarray that was actually the entire array).  Code was
passing in a count of bytes, which can overflow an int. since we know
the type is contiguous, simply pass a count of that type (not bytes) to
write_contig

Signed-off-by: Mohamad Chaarawi <chaarawi@hdfgroup.org>

Xin Zhao authored Jul 13, 2014 and Pavan Balaji committed Jul 13, 2014

See "Notes for memory barriers in RMA synchronizations" in
src/mpid/ch3/src/ch3u_rma_sync.c.

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

Xin Zhao authored Jul 13, 2014 and Pavan Balaji committed Jul 13, 2014

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

The f2c/c2f handle conversion functions are only defined in C in the standard.
We provide non-standard extensions in Fortran. All functions, except MPI_File_f2c/c2f,
are trival to implement. For MPI_File_f2c/c2f, we bind the Fortran functions to C.
We do not provide PMPI version for them since it is not required by the standard.

No review since F08 binding is experimental now.

We were creating the rank list for the window start group twice
earlier, once for synchronization and once for the actual issuing of
the operations.  This patch combines them into a single creation of
the array.

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

We were using a stack variable earlier, so when FLUSH from a different
thread tried to send that data out it couldn't access that address.

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

MPI_Win_shared_query() currently returns the base address of the passed in rank.
All tasks other than task 0 assume that the first set of data belongs to task 0,
second set belongs to task 1 and so on. The assumption caused the data integrity
problem found by win_shared.c.

To fix the problem, instead of returning the base address of a rank,
MPI_Win_shared_query() should return the base address of shared segment
allocated for the window.

(ibm) D198663

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

So that f08 tests can be tested separately.

No review since F08 binding is experimental now.

When the debugging option about memory usage tracing is turned on,
additional memory overhead for tracing may be significant especially
when a large number of small memory allocations are requested, as
reported in #2068. However, since the tracing data is necessary to
keep track of memory usage, it cannot be removed. Here, we provide
a warning message when the memory overhead for tracing is over the
threshold value.

See #2068

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

A previous commit [33d82829

] increased the feature level in files
using pread/pwrite functions to strictly adhere to the C99 standard.
The fix inadvertantly caused failed builds when C99 was not enabled
on Solaris. Now we conditionally raise the feature level depending
on the strictness of the build.

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

Ensures that a VC is activated during a recv operation. Before this,
a VC may only be activated on the sender side. MPI_Finalize would then
hang when waiting for all VCs to terminate.

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

Since this is tied to the amount of cache we can maintain for the PMI
keyvals, a higher number can improve performance.

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

When we were flushing the PMI keyval cache, the cache was flushed in
the form of a single very long PMI command.  The server would
interpret this as a single PMI command with many arguments.  However,
if the cache is larger than the allowed number of PMI arguments, we
might run out of the allocated space.  This patch connects these two
values, so the cache can never overrun the number of PMI arguments.

Thanks to Norio Yamaguchi @ Riken for reporting this issue.

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

Array size on fence_shm was set to 0, which was not caught by most
platforms, but made solaris go all crazy.

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

We need to add a memory barrier at the end of the Win_complete
function, so that shared memory operations issued during the
start/complete epoch are visible to other processes on the node.

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

When a window uses direct shared-memory operations that are
immediately issued internally, we cannot avoid synchronization during
the start operation.  This patch synchronizes processes that reside on
the same node during start and the processes that do not reside on the
same node during complete.

Fixes #2041.

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

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

When SHM is allocated for RMA window, operations are completed
eagerly (as soon as they are posted by the user), therefore we
need barrier semantics in the FENCE that opens an epoch to prevent
SHM ops happening on target process before that target process
starts an epoch.

Note that we need memory barrier before and after synchronization
calls in both FENCEs that starts and ends an epoch to guarantee the
ordering of load/store operations with synchronizations.

See #2041.

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

If there is only one process on the node in the particular
communicator, we don't need to allocate any shared memory for it.  In
this case, we simply call the non-shared-memory functions for
Win_allocate and free.

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

This seems to be a leftover change from earlier that got pulled into
[b61fc702

].

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

Antonio J. Pena authored Jul 03, 2014 and Kenneth Raffenetti committed Jul 07, 2014

After fixing #2002, MPICH_ATTR_WEAK_ALIAS is not needed. This commit
cleans up its leftovers.

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

Antonio J. Pena authored Jul 03, 2014 and Kenneth Raffenetti committed Jul 07, 2014

Moved the weak,alias attribute declarations from header files to the
implementation. Complies with the requirement that alias targets are
defined in the same compilation unit.

Fixes #2002

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

Antonio J. Pena authored Mar 17, 2014 and Kenneth Raffenetti committed Jul 07, 2014

Since GCC4, it is required that when declaring a function alias,
the target function is defined in the same compilation unit.

See #2002

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

(ibm) D198243

Signed-off-by: Michael Blocksome <blocksom@us.ibm.com>
Signed-off-by: Wesley Bland <wbland@anl.gov>

If the size of the input buffers is 0, don't bother to check if they are
aliased.

Fixes #2124

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

Pavan Balaji authored May 19, 2013 and Pavan Balaji committed Jul 07, 2014

Picked up the relevant parts of [a460c1ee], [dab3166d], and [1eb07c6e

].

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

No reviewer.

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

Pavan Balaji authored Jul 05, 2014 and Kenneth Raffenetti committed Jul 07, 2014

While moving symbols to the correct libraries in [9c337914

], we forgot
to move MPI_Status_c2f.

Fixes #2122.

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

Norio Yamaguchi authored May 28, 2014 and Pavan Balaji committed Jul 04, 2014

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

Norio Yamaguchi authored May 26, 2014 and Pavan Balaji committed Jul 04, 2014

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