We slow down consecutive ssh connections to the same host to avoid the
ssh server falling over.  This patch allows the user to control how
many ssh connections we allow before throttling.

If the device specified a timer type, use that by default.  If none is
specified, configure will try to detect one.  User-specified types are
always given priority.

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

This will match the #include usage in the `hwloc` package.

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

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

The previous patch was meant to allow devices to automatically pick
the default timer type they like.  However, it forces the device to
pick one of the predefined timer types.

It is now covered under the PARAMS interface and is listed with the
remaining environment variables in README.envvar.  Also improve the
description of the variable in README.envvar.

MPIR_PARAM_ gets higher priority than MPICH_.  So we place the
MPIR_PARAM_ version as the preferred environment variable name, but
provide the MPICH_ version as an alternative.

The ABI string is updated to 11:x:y since we added the C++ binding for
distgraph.  No interfaces were deleted, so 'x' goes to 0, and 'y' goes
to 1.

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

If the user asked for an asynchronous thread, but the device does not
provide MPI_THREAD_MULTIPLE support, throw a warning.

1. Use the params interface to control the default thread level.

2. Don't restrict support for the default thread level only in
THREAD_MULTIPLE installations.

Use the params interface to check the ASYNC_PROGRESS variable.

We were originally initiating the asynchronous thread in SERIALIZED
mode as well.  But this is incorrect, since the remaining threads do
not acquire a lock and will conflict with this thread.

This is taken from the mpich-benchmarks repository, which is not
maintained anymore.

If a higher than single or funneled thread level is required and no
thread package is provided, abort.

1. Remove the option for default, since that can be picked with just
--enable-threads without any additional arguments as well.

2. Don't do any device-specific assignments in the top-level
configure.  This is not the right place to do that.  If the device
wants to use a particular thread-level, it should set that in its
subconfigure.m4.

Sameh Sharkawi authored May 24, 2013 and Pavan Balaji committed Jun 07, 2013

mpi-io cases failed w/ mpich2 on AIX:
AIX compiler pads 3 bytes to chars inside structs.
MPICH2 needed to match that when on AIX. For cross compiles,
two AIX specific cross files are added so the config
is aware of this and match AIX padding.

 (ibm) D190327
 (ibm) 76997eee0762ee9e4ad22ec0ee8b8f8c820bea94

Michael Blocksome authored May 22, 2013 and Pavan Balaji committed Jun 07, 2013

The pamid device does not require a pm and will set the default to 'no'.
This eliminates the requirement to specify the '--with-pm=no' configure
option in order to configure and build successfully.

(includes modifications by Pavan Balaji @ Argonne).

Michael Blocksome authored May 22, 2013 and Pavan Balaji committed Jun 07, 2013

For a bgq configure, this means that the following option is no longer
required to be set:

 --with-cross-file=src/mpid/pamid/cross/bgq8

(with modifications by Pavan Balaji @ Argonne).

Michael Blocksome authored May 22, 2013 and Pavan Balaji committed Jun 07, 2013

When configuring for bgq, the cross compile must be specified which can
then be used as the default pamid implementation.

 --host=powerpc64-bgq-linux
 --target=powerpc64-bgq-linux
 --with-device=pamid

Previously, the device was required to be:

 --with-device=pamid:BGQ

Michael Blocksome authored May 22, 2013 and Pavan Balaji committed Jun 07, 2013

The '--with-bgq-install-dir' will no longer be required for a default
bgq configure as the search logic will now choose an appropriate
install.

Michael Blocksome authored Apr 18, 2013 and Pavan Balaji committed Jun 07, 2013

- add fortran interfaces for new MPIX_* functions
- remove fortran interface for MPIX_Pset_io_node()
- do not use the MPIX_Pset_io_node() function internally

The MPIX_Pset_io_node() function has been deprecated.

(ibm) CPS 92XKPE
(ibm) 6c7f293ccda15d447369ee7327a225d3b7979f15
(ibm) 106eef011bc1e58ada7e208e8c56ee341045611b
(ibm) 23b42d38b9f75ddd93701a021f84418a30a7693a
(ibm) b39954ab0052e0c5676cd9c270f159a4f6a99828

Signed-off-by: Haizhu Liu <haizhu@us.ibm.com>

Michael Blocksome authored Apr 18, 2013 and Pavan Balaji committed Jun 07, 2013

 (ibm) CPS 95PHUZ
 (ibm) 7bfb869616ec9aaffe4e5de9e05840fed4e4065a
 (ibm) d5a3fc9e58ccfcf185f9671b06d8307fd56e653c

Signed-off-by: Haizhu Liu <haizhu@us.ibm.com>

Su Huang authored Jun 03, 2013 and Pavan Balaji committed Jun 07, 2013

 (ibm) D190899
 (ibm) 6693a84bd469820d94e694bbb7e3fbe2b6188e2f

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