On x86_64 reading a 128-bit value requires multiple instructions.
Under some conditions if the counted pointer counter is read before
the item pointer the fifo can be left in an inconsistent state. This
commit forces the read of the counter to always be read first.
The fifo does not appear to suffer from the same race.
It is possible the compiler can reorder the read of the head item and
the head itself. This could lead to a situation where the item
returned was not really the head item.
Thanks to Nathan for pointing out that I missed snipping one line in
2f9c69f016 (I removed the trailing
comment, but not the trailing pragma -- oops!).
Retain the hetero-nodes flag for those cases where the user *knows* that there are differences and our automated system isn't good enough to see it.
Will obviously require further refinement as we find out which variances it can detect, and which it cannot.
1. Ensure to override CFLAGS properly. Move the setting of CFLAGS outside the AM_CONDITIONAL so that Automake doesn't get confused (because CFLAGS is already set inside an AM_CONDITIONAL -- moving it outside the conditional ensure that this local CFLAGS override trumps all other CFLAGS overrides).
2. Only build libfabric on Linux. Add a little more configury to ensure that we only try to build libfabric on Linux.
3. Remove a dead/unused file
4. Fix typo in condition check
5. Use "false", not "/bin/false"
This commit represents the conversion of the usnic BTL from verbs to
libfabric.
For the moment, libfabric is embedded in Open MPI (currently in the
usnic BTL). This is because the libfabric API is still changing, and
also has not yet been released. Ultimately, this embedded copy of
libfabric will likely disappear and the usnic BTL will rely on an
external installation of libfabric.
New configure options:
* --with-libfabric: will cause configure to fail if libfabric support
cannot be built
* --without-libfabric: will prevent libfabric support from being built
* --with-libfabric=DIR: use an external libfabric installation
* --with-libfabric-libdir=LIBDIR: when paired with --with-libfabric=DIR,
use LIBDIR for the libfabric installation library dir
The --with-libnl3[-libdir] arguments are now gone.
This commit adds a new class: opal_fifo.h. The new class has atomic, non-atomic,
and opal_using_threads() conditoned routines. It should be used when first-in
first-out is required and should perform much better than using locks and an
opal_list_t. Like with opal_lifo_t there are two versions of the atomic
implementation: 128-bit compare-and-swap, and spin-locked. More implementations
can be added later (LL/SC comes to mind).
This commit also adds a unit test for the opal_fifo_t class. This test verifies
the fifo implementation when using multiple threads.
- Rename opal_atomic_lifo_t to opal_lifo_t to reflect both atomic and
non-atomic usage. Added new routines (opal_lifo_*_st) for non-atomic
usage as well as routines conditioned off opal_using_threads(). The
atomic versions are always thread safe and the non-atomic are always
not thread safe.
- Add a new atomic lifo implementation that makes use of 128-bit
compare-and-swap. The new implementation should scale better with
larger numbers of threads.
- Add threading unit test for opal_lifo_t.
There currently is no standard support for 128-bit integer types. Any use
of the __int128 and int128_t types can lead to warnings from the compiler
when using -Wpedantic. Additionally, some compilers may support __int128
and other may support int128_t. This commit addresses both issues by
defining opal_int128_t if there is a supported 128-bit type. In the
case of GCC a pragma has been added to suppress warnings about __int128
not being a standard C type.
A 128-bit compare-and-swap will enable a better atomic lifo implementation
that uses the pointer + counter method to avoid ABA issues. This commit
adds configury to check for the instruction (cmpxchg16b) and adds an
implementation that uses the __int128 type available in C99.
If OPAL_MODEX_RECV() returns OPAL_ERR_NOT_FOUND, the peer didn't
send any Portals4 BTL info. This is not a fatal error. Instead of
disqualifying the Portals4 BTL just ignore that peer.
@jsquyres reported this in #194.
inserted in the ompi_proc_list as soon as it is created and it
is removed only upon the call to the destructor. In ompi_proc_finalize
we loop over all procs in ompi_proc_finalize and release them once.
However, as a proc is not removed from this list right away, we
decrease the ref count for each proc until it reach zero and the
proc is finally removed. Thus, we cannot clean the BML/BTL after
the call the ompi_proc_finalize.
A quick fix is to delay the call to ompi_proc_finalize until all
other frameworks have been finalized, and then the behavior
depicted above will give the expected outcome.
value NULL for the descriptor
The send inline optimization uses the btl_sendi function to achieve
lower latency and higher message rates. The problem is the btl_sendi
function was allowed to return a descriptor to the caller. This is fine
for some paths but not ok for the send inline optimization. To fix
this the btl now must be able to handle descriptor = NULL.
structure
This structure member was originally used to specify the remote segment
for an RDMA operation. Since the new btl interface no longer uses
desriptors for RDMA this member no longer has a purpose. In addition
to removing these members the local segment information has been
renamed to des_segments/des_segment_count.
The old BTL interface provided support for RDMA through the use of
the btl_prepare_src and btl_prepare_dst functions. These functions were
expected to prepare as much of the user buffer as possible for the RDMA
operation and return a descriptor. The descriptor contained segment
information on the prepared region. The btl user could then pass the
RDMA segment information to a remote peer. Once the peer received that
information it then packed it into a similar descriptor on the other
side that could then be passed into a single btl_put or btl_get
operation.
Changes:
- Removed the btl_prepare_dst function. This reflects the fact that
RDMA operations no longer depend on "prepared" descriptors.
- Removed the btl_seg_size member. There is no need to btl's to
subclass the mca_btl_base_segment_t class anymore.
...
Add more
Use a more reliable way to tell if a process is
1) in a Cray PAGG
2) is actually considered an application process on
a compute node (not for example, a process in a PAGG
on a mom node).
using knem
It is valid to modify the remote segment that will be used with the
btl put/get operations as long as the resulting address range falls in
the originally prepared segment. Vader should have been calculating the
offset of the remote address in the registered region. This commit
fixes this issue.
We recognize that this means other users of OPAL will need to "wrap" the opal_process_name_t if they desire to abstract it in some fashion. This is regrettable, and we are looking at possible alternatives that might mitigate that requirement. Meantime, however, we have to put the needs of the OMPI community first, and are taking this step to restore hetero and SPARC support.
There were mistakes in the Makefiles for the ugni btl and
mca/common/ugni that prevented the ugni btl from being
used unless one happened to set the --disable-dlopen option
on the config line.
This commit fixes this problem.
This enables IBM's software iWARP provider. With this driver you can
run iWARP/RDMA over any ethernet NIC. Useful for testing OMPI RDMA
logic without requiring an expensive RDMA adapter/infrastructure.
The Soft iWARP code is at: https://www.gitorious.org/softiwarp
Avoid a problem with double-derefence of a variable macro name (i.e.,
a macro with part of its name from an AC_SUBST, such as
```$(foo@BAR@baz)```.
In what might be a bug in Automake 1.14.1, if you do a pattern like
this:
```makefile
lib_LTLIBRARIES = lib@A_PREFIX@a_lib.la
noinst_LTLIBRARIES = lib@A_PREFIX@a_noinst.la
lib@A_PREFIX@a_lib_la_SOURCES = a.c
lib@A_PREFIX@a_noinst_la_SOURCES = $(lib@A_PREFIX@a_lib_la_SOURCES)
```
Then in the resulting Makefile, the value of
```$(lib@A_PREFIX@a_lib_la_OBJECTS)``` will be *blank* (when it really
should be ```a.o```).
To workaround this potential bug, I've simply avoided doing
double-derefences like this, and effectively set the second
```_SOURCES``` line equal to ```a.c``` (just like the first
```_SOURCES``` line).
Fixes#250.
These two macros set the MCA prefix and MCA cmd line id,
respectively. Specifically, MCA parameters will be named
PREFIX<foo> in the environment, and the cmd line will use
-ID foo bar.
These macros must be called during configure.ac and a value
supplied. In the case of Open MPI, the values given are
PREFIX=OMPI_MCA_ and ID=mca.
Other projects (such as ORCM) will call these macros with
their own unique values. For example, ORCM uses PREFIX=ORCM_MCA_
and ID=omca
This scheme is necessary to allow running Open MPI applications under
systems that use their own versions of ORTE and OPAL. For example,
when running OMPI applications under ORCM, we need the MCA params passed
to the ORCM daemons to be separated from those recognized by the OMPI application.
At some point we added a sanity check to the btl base to ensure that
the btl flags match the available functions (this prevents user's from
specifying get or put when no function exists). This check was
disabling get for the sm btl since at the time of the check there is
no btl_get function. The simplest fix is to set a dummy value to btl_get
that will be overwritten with the proper value on btl initialization.
Closes#239.
If there are no usnic BTL modules, then just avoid sending any modex
message at all (other BTLs do this; it's safe to do).
The change is smaller than it looks: I added a "if 0 ==..." check at
the top to return immediately if there are no BTL modules. Then I
removed some now-unnecessary conditionals and un-indented as
appropriate.
Fixes#248
These two macros set the prefix for the OPAL and ORTE libraries,
respectively. Specifically, the OPAL library will be named
libPREFIXopen-pal.la and the ORTE library will be named
libPREFIXopen-rte.la.
These macros must be called, even if the prefix argument is empty.
The intent is that Open MPI will call these macros with an empty
prefix, but other projects (such as ORCM) will call these macros with
a non-empty prefix. For example, ORCM libraries can be named
liborcm-open-pal.la and liborcm-open-rte.la.
This scheme is necessary to allow running Open MPI applications under
systems that use their own versions of ORTE and OPAL. For example,
when running MPI applications under ORTE, if the ORTE and OPAL
libraries between OMPI and ORCM are not identical (which, because they
are released at different times, are likely to be different), we need
to ensure that the OMPI applications link against their ORTE and OPAL
libraries, but the ORCM executables link against their ORTE and OPAL
libraries.
the OPAL and ORTE libraries. This is required by projects such as ORCM
that have their own ORTE and OPAL libraries in order to avoid library
confusion. By renaming their version of the libraries, the OMPI
applications can correctly dynamically load the correct one for their
build."
This reverts commit 63f619f871.
This commit makes the folowing changes:
- Add support for the knem single-copy mechanism. Initially vader will only
support the synchronous copy mode. Asynchronous copy support may be added
int the future.
- Improve Linux cross memory attach (CMA) when using restrictive ptrace
settings. This will allow Open MPI to use CMA without modifying the system
settings to support ptrace attach (see /etc/sysctl.d/10-ptrace.conf).
- Allow runtime selection of the single copy mechanism. The default behavior
is to use the best available. The priority list of single-copy mehanisms is
as follows: xpmem, cma, and knem.
- Allow disabling support for kernel-assisted single copy.
- Some tuning and bug fixes.
Restore the functionality to error out (and show a helpful message) if
knem support is requested by is either not compiled in or cannot be
activated.
Thanks to Gus Correa for bringing the matter to our attention.
Somehow, this MCA param was accidentally dropped after v1.6.5. Thanks
to Gus Correa for bringing this matter to our attention.
Also moving some MCA params down from level 9 to levels 4/5.
* redefine orte_process_name_t so it can be converted
between host and network format as an opal_identifier_t
aka uint64_t by the OPAL layer.
* correctly send OPAL_DSTORE_ARCH key
Revert "OPAL: drop dead with core on bad flow. rarely happens with helloworld on large scale."
This reverts commit 86f1d5af3e.
Will be reconsidered via RFC as it represents a significant change in behavior
of preregistered buffers
Before this change eager gets we retried on each progress loop. This commit
modifies the protocol to only retry eager gets when another eager get has
completed. This commit also cleans up some callback code that is no longer
needed.
The GNI_RDMAMODE_FENCE bit was a left over from
async progress work that is not needed at this point
in the gni BTL. Removing the bit also allows
for the removal of the GNI_CDM_MODE_BTE_SINGLE_CHANNEL
bit from the GNI_CdmCreate call.
1. It's actually hashing now, whereas the old OPAL hash table was not. Thus, it is a bug fix for and, as such, should be included in the 1.8 series.
2. It is dynamic and can grow and shrink the number of buckets in accordance with job size, whereas the old OPAL hash table had a fixed number of buckets which resulted in poor retrieval performance at large scale.
This scheme has been deployed in the field on very large H.P./Mellanox systems and has been demonstrated to significantly decrease job start-up time (~ 20% improvement) when launching applications directly with srun in SLURM environments. However, neither SLURM nor direct launch are prerequisites to take advantage of this change as any entity that utilizes OPAL hash table objects can benefit (at least partially) from this contribution.
This commit adds initial ugni thread safety support.
With this commit, sun thread tests (excepting MPI-2 RMA)
pass with various process counts and threads/process.
Also osu_latency_mt passes.
With --enable-memchecker builds, use calloc(3) for OBJ_NEW instead of
malloc(3). This cuts down on a lot of valgrind/memory checker false
positive output.
Also make a minor change in the valgrind configure.m4; have it assign
0xf to a char. The prior assignment (of 0xff) was warning about an
overflow. This didn't really matter, but we might as well make the
test not have a gratuitious warning in it.
Properly setup the opal_process_info structure early in the initialization procedure. Define the local hostname right at the beginning of opal_init so all parts of opal can use it. Overlay that during orte_init as the user may choose to remove fqdn and strip prefixes during that time. Setup the job_session_dir and other such info immediately when it becomes available during orte_init.
Update the VERSION file scheme:
* Remove "want_repo_rev".
* Add "tarball_version".
All values are now always included (major, minor, release, greek,
repo_rev). However, configure.ac now runs "opal_get_version.sh
... --tarball", which will return the value of tarball_version (if it
is non-empty) or the "full" version string (i.e.,
"major.minor.releasegreek").
Remove configure.params support: configure.params hasn't been used in
years.
Also remove autogen.subdirs support; those should really be handled by
their respective Makefile.am's.
