this commit syncs ompio related directories in v4.1.x to master. The efforts to bring the lustre performance fixes and support for external32 data representation over were too overwhelming when dealing with every single pr individually.
There are a very few minor modification that had to be done for syncing:
- v4.1.x does not have opal/mutex.h
- v4.1.x does not have opal_atomic_int32_t datatype
- the io module structure has two fewer function pointers (related to info_set/get) compared to the version on master.
Tested so far with the ompio testsuite as well as hdf5-1.10.5 testsuite (testphdf5, t_shapesame, t_bigio) on an XFS file system.
More tests on Lustre and BeeGFS to follow.
Signed-off-by: Edgar Gabriel <egabriel@central.uh.edu>
Add some "const"s that needed to be applied here on the v4.1.x branch,
effectively by cherry-picking part of b65ec273074 from master.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
Exclude HAN, don't include it.
Signed-off-by: Joseph Schuchart <schuchart@icl.utk.edu>
(cherry picked from commit 33105b031bbc821a6c5d816c4801d62072347f9c)
Fixes#8195. This PR doesn't fix all the warnings from #8195, but
fixes many of them (e.g., I didn't get the "string might be truncated"
warnings on my Mac).
This is an adaptation of 14aa5fae3c42f14a1c6a259dede93d5ca7ecb82c from
master; it drops some things that aren't relevant here on the v4.1.x
branch and adds a few more warnings fixes that are relevant here on
v4.1.x that aren't relevant on master.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
(cherry-picked from 14aa5fae3c42f14a1c6a259dede93d5ca7ecb82c)
This has shown to be more effective in achieving overlap
of inter- and intra-node communication and reduces the inital
delay before hitting the network.
Signed-off-by: Joseph Schuchart <schuchart@icl.utk.edu>
(cherry picked from commit 1cdc85564ed6c771f301c63d6bc6d8c1c8cf4a4c)
Also make coll/tuned the default for shared memory communication
as coll/sm has shown performance issues that need investigation.
Signed-off-by: Joseph Schuchart <schuchart@icl.utk.edu>
(cherry picked from commit 971d58c52454a6edecdbb1a44ebd037a86e69a69)
The selectable list is sorted with lowest to highest priority so the
user-defined preferences should be appended to the list.
The preference treatment should also maintain the order provided by the user
(first item has highest priority) so switch the loop order.
Signed-off-by: Joseph Schuchart <schuchart@icl.utk.edu>
(cherry picked from commit dd54af94508dc9ccee3e589276a9ede62fc8e409)
The total size depends on number of ranks so the usual ranges don't work.
Thus, use the average across all ranks to make a decision.
Signed-off-by: Joseph Schuchart <schuchart@icl.utk.edu>
(cherry picked from commit f670364d764bf7409e03860bf539a0a2884ffab3)
MPI_Ialltoallw() and friends take a const MPI_Datatype types[] argument.
In order to be able to call OBJ_RELEASE(types[0]), we used to simply
drop the const modifier. This change make it right by introducing the
OBJ_RELEASE_NO_NULLIFY(object) macro that no more set object = NULL
if the object is freed.
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
(cherry picked from commit c49e5e5c4a6902a7bf462a98c24c9ebd86b4ab63)
This commit removes the unnecessary call to `fi_getinfo()` when
initializing the MTL. `cq_data_size` is a domain attribute that will be
available to the MTL from the initial query itself. FI_DIRECTED_RECV is
a primary capability that has to be requested for a provider to enable
it, so adding that to the initial requirement. The redundant query was
also overwriting the contents of the prov object, which already had the
include/exclude filtering and multi-NIC logic applied to it.
Signed-off-by: Raghu Raja <craghun@amazon.com>
(cherry picked from commit 6233dea68d8495e20746c8e8d8af8d9c03a20206)
These selections seem harmful in my measurements and don't seem to be
motivated by previous measurement data.
Signed-off-by: Joseph Schuchart <schuchart@icl.utk.edu>
(cherry picked from commit a15e5dc7f042f21f8adc08453b13bc7210bf2bac)
Bcast: scatter_allgather and scatter_allgather_ring expect N_elem >= N_procs
Allreduce: rabenseifner expects N_elem >= pow2 nearest to N_procs
In all cases, the implementations will fall back to a linear implementation,
which will most likely yield the worst performance (noted for 4B bcast on 128 ranks)
Signed-off-by: Joseph Schuchart <schuchart@icl.utk.edu>
(cherry picked from commit 04d198fc9f4b592fc512b27428d3158e279b51e1)
The mca parameters coll_tuned_*_algorithm are ignored unless coll_tuned_use_dynamic_rules is true so mention that in the description.
Signed-off-by: Joseph Schuchart <schuchart@icl.utk.edu>
(cherry picked from commit 06f605c1e11dad9804232306838141d5a50a1929)
If PMIX_PACKAGE_RANK is available, uses this value to select between multiple
NIC of equal distance between the current process. If this value is not
available, try to calculate it by getting the locality string from each local
process and assign a package_rank. If everything fails, fall back to using
process_id.rank to select the NIC. This last case is not ideal, but has a small
chance of occuring, and causes an output to be displayed to notify that this is
occuring.
Some of the information in master branch is not available for the multi-NIC
patch, such as myprocinfo.rank. This info is used to select between multiple
NIC of equal distance to the process. This adapts the previous commit to work
with the v4.1.x branch.
Signed-off-by: Nikola Dancejic <dancejic@amazon.com>
(cherry picked from commit 8017f1280137dee4d5fc7dac0a5c627e72e48058)
PGI (20.4) compiler do not define this intrinsic, so only build
AVX512 support if _mm512_mullo_epi64() intrisic is defined.
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
(cherry picked from commit 26e42f9a0c21187247ec072b5bacfb39e7a9f9f7)
Among many other things:
- Fix an imbalance bug in MPI_allgather
- Accept more human readable configuration files. We can now specify
the collective by name instead of a magic number, and the component
we want to use also by name.
- Add the capability to have optional arguments in the collective
communication configuration file. Right now the capability exists
for segment lengths, but is yet to be connected with the algorithms.
- Redo the initialization of all HAN collectives.
Cleanup the fallback collective support.
- In case the module is unable to deliver the expected result, it will fallback
executing the collective operation on another collective component. This change
make the support for this fallback simpler to use.
- Implement a fallback allowing a HAN module to remove itself as
potential active collective module, and instead fallback to the
next module in line.
- Completely disable the HAN modules on error. From the moment an error is
encountered they remove themselves from the communicator, and in case some
other modules calls them simply behave as a pass-through.
Communicator: provide ompi_comm_split_with_info to split and provide info at the same time
Add ompi_comm_coll_preference info key to control collective component selection
COLL HAN: use info keys instead of component-level variable to communicate topology level between abstraction layers
- The info value is a comma-separated list of entries, which are chosen with
decreasing priorities. This overrides the priority of the component,
unless the component has disqualified itself.
An entry prefixed with ^ starts the ignore-list. Any entry following this
character will be ingnored during the collective component selection for the
communicator.
Example: "sm,libnbc,^han,adapt" gives sm the highest preference, followed
by libnbc. The components han and adapt are ignored in the selection process.
- Allocate a temporary buffer for all lower-level leaders (length 2 segments)
- Fix the handling of MPI_IN_PLACE for gather and scatter.
COLL HAN: Fix topology handling
- HAN should not rely on node names to determine the ordering of ranks.
Instead, use the node leaders as identifiers and short-cut if the
node-leaders agree that ranks are consecutive. Also, error out if
the rank distribution is imbalanced for now.
Signed-off-by: Xi Luo <xluo12@vols.utk.edu>
Signed-off-by: Joseph Schuchart <schuchart@icl.utk.edu>
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
Conflicts:
ompi/mca/coll/adapt/coll_adapt_ibcast.c
* first import of Bull specific modifications to HAN
* Cleaning, renaming and compilation fixing Changed all future into han.
* Import BULL specific modifications in coll/tuned and coll/base
* Fixed compilation issues in Han
* Changed han_output to directly point to coll framework output.
* The verbosity MCA parameter was removed as a duplicated of coll verbosity
* Add fallback in han reduce when op cannot commute and ppn are imbalanced
* Added fallback wfor han bcast when nodes do not have the same number of process
* Add fallback in han scatter when ppn are imbalanced
+ fixed missing scatter_fn pointer in the module interface
Signed-off-by: Brelle Emmanuel <emmanuel.brelle@atos.net>
Co-authored-by: a700850 <pierre.lemarinier@atos.net>
Co-authored-by: germainf <florent.germain@atos.net>
a hierarchical, architecture-aware collective communication module.
Add Reduce and remove up_seg_size and low_seg_size in Bcast
Increase HAN's priority
Signed-off-by: Xi Luo <xluo12@vols.utk.edu>
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
The current iprobe/improbe implementations merely checks the return
code on the posted receive operation to tell if there is a match or
not. This commit moves the check to the probe's error callback
instead. Per the semantics defined in libfabric, the peek operation is
asynchronous and the results are to be fetched from the completion
queue. If no message is found matching the tags specified in the peek
request, then a completion queue error entry with err field set to
FI_ENOMSG will be available.
Signed-off-by: Raghu Raja <craghun@amazon.com>
(cherry picked from commit 39f8a86b65e53d4f4c7b9cd64532cf857a4cfa58)
In multi-threaded scenarios, any thread that attempts to read a CQ
when there's a pending error CQ entry gets an -FI_EAVAIL. Without
any serialization here (which is okay, since libfabric will protect
access to critical CQ objects), all threads proceed to read from the
error CQ, but only one thread fetches the entry while others get
-FI_EAGAIN indicating an empty queue, which is not erroneous.
Signed-off-by: Raghu Raja <craghun@amazon.com>
(cherry picked from commit 415dddb9afbed28dc93f65ead867ef5d245d94c4)
This is a meta commit, that encapsulate all the ADAPT commits in the master
into a single PR for 4.1. The master commits included here are:
fe73586, a4be3bb, d712645, c2970a3, e59bde9, ee592f3 and c98e387.
Here is a detailed list of added capabilities:
* coll/adapt: Fix naming conventions and C11 atomic use
* coll/adapt: Remove unused component field in module
* Consistent handling of zero counts in the MPI API.
* Correctly handle non-blocking collectives tags
* As it is possible to have multiple outstanding non-blocking collectives
provided by different collective modules, we need a consistent
mechanism to allow them to select unique tags for each instance of a
collective.
* Add support for fallback to previous coll module on non-commutative operations (#30)
* Replace mutexes by atomic operations.
* Use the correct nbc request type (for both ibcast and ireduce)
* coll/base: document type casts in ompi_coll_base_retain_*
* add module-wide topology cache
* use standard instead of synchronous send and add mca parameter to control mode of initial send in ireduce/ibcast
* reduce number of memory allocations
* call the default request completion.
* Remove the requests from the Fortran lookup conversion tables before completing
and free it.
* piggybacking Bull functionalities
Signed-off-by: Xi Luo <xluo12@vols.utk.edu>
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
Signed-off-by: Marc Sergent <marc.sergent@atos.net>
Co-authored-by: Joseph Schuchart <schuchart@hlrs.de>
Co-authored-by: Lemarinier, Pierre <pierre.lemarinier@atos.net>
Co-authored-by: pierrele <31764860+pierrele@users.noreply.github.com>
The gather and scatter operations did not use the correct message size
(Only did datatype size * com size). This did not correctly reflect the
total message size and prevents fine tuning within a com size. This
patch multiplies the value by the number of elements sent.
Signed-off-by: William Zhang <wilzhang@amazon.com>
(cherry picked from commit 50823fe9a9ef4f93e55ee2087b311303d49f90a8)
Uncovered a problem using the GNI provider with the OFI MTL.
See https://github.com/ofiwg/libfabric/issues/6194.
Related to #8001
Signed-off-by: Howard Pritchard <hppritcha@gmail.com>
(cherry picked from commit d6ac41cbbd8dbfdbf1a87533e2e292f2508a85ff)
Reduce scatter block and reduce scatter algorithms were hitting
correctness issues for non commutative strided tests. We will revert to
the original default algorithms for those two collectives (basic linear
and non overlapping respectively) in the non commutative op case.
See #8010
Signed-off-by: William Zhang <wilzhang@amazon.com>
(cherry picked from commit 57b95bcb45d5ce3ae1a1e00bd17ceeaa206526fe)
the ofi mtl mrecv was not properly setting the message in/out
arg to MPI_MRECV to MPI_MESSAGE_NULL.
Signed-off-by: Howard Pritchard <hppritcha@gmail.com>
(cherry picked from commit e6f81ed6d6f4940c95732451d6ebf4d39cde1591)
Alter the test to validate misaligned data.
Fixes#7954.
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
(cherry picked from commit b6d71aa893db51c393df9c993fde597eff5457e8)
Signed-off-by: Brian Barrett <bbarrett@amazon.com>
The btl/ofi does not currently utilize the common ofi include/exclude
list. Added verification code similar to the mtl/ofi that will check if
the info object is in the include or exclude list. If it isn't in the
include list or is in the exclude list, validate_info will return
OPAL_ERROR. The btl/ofi will no longer pass a provider name as a hint
when calling getinfo, instead filtering the provider during
validate_info.
This patch also moves the is_in_list MTL function into common code and
adds additional debugging output to the BTL to match the MTL standard.
Signed-off-by: William Zhang <wilzhang@amazon.com>
(cherry picked from commit 9b8f463a768206a26e5b1dcea0612a403462b1d0)
The default algorithm selections were out of date and not performing
well. After gathering data from OMPI developers, new default algorithm
decisions were selected for:
allgather
allgatherv
allreduce
alltoall
alltoallv
barrier
bcast
gather
reduce
reduce_scatter_block
reduce_scatter
scatter
These results were gathered using the ompi-collectives-tuning package
and then averaged amongst the results gathered from multiple OMPI
developers on their clusters.
You can access the graphs and averaged data here:
https://drive.google.com/drive/folders/1MV5E9gN-5tootoWoh62aoXmN0jiWiqh3
Signed-off-by: William Zhang <wilzhang@amazon.com>
(cherry picked from commit ce40cfbaa53406be71319041e13e893b0def7ad9)
Add logic to handle different architectural capabilities
Detect the compiler flags necessary to build specialized
versions of the MPI_OP. Once the different flavors (AVX512,
AVX2, AVX) are built, detect at runtime which is the best
match with the current processor capabilities.
Add validation checks for loadu 256 and 512 bits.
Add validation tests for MPI_Op.
Signed-off-by: Jeff Squyres <jsquyres@cisco.com>
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
Signed-off-by: dongzhong <zhongdong0321@hotmail.com>
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
(cherry picked from commit 14b3c706289cfc26e53b13efd3eb85641636e459)
