In multithreaded case, it is expensive to release the lock, call the slow match
and retake the lock again just to queue the frag. This patch will eliminate number of
lock taken by queueing the frag right away and return.
Signed-off-by: Thananon Patinyasakdikul <tpatinya@utk.edu>
If Open MPI is configured with CUDA, then user also should be using a CUDA build of
PSM2 and therefore be setting PSM2_CUDA environment variable to 1 while using
CUDA buffers for transfers. If we detect this setting to be missing, force set
it. If user wants to use this build for regular (Host buffer) transfers, we
allow the option of setting PSM2_CUDA=0, but print a warning
message to user that it is not a recommended usage scenario.
Signed-off-by: Aravind Gopalakrishnan <Aravind.Gopalakrishnan@intel.com>
PSM2 enables support for GPU buffers and CUDA managed memory and it can
directly recognize GPU buffers, handle copies between HFIs and GPUs.
Therefore, it is not required for OMPI to handle GPU buffers for pt2pt cases.
In this patch, we allow the PSM2 MTL to specify when
it does not require CUDA convertor support. This allows us to skip CUDA
convertor init phases and lets PSM2 handle the memory transfers.
This translates to improvements in latency.
The patch enables blocking collectives and workloads with GPU contiguous,
GPU non-contiguous memory.
Signed-off-by: Aravind Gopalakrishnan <Aravind.Gopalakrishnan@intel.com>
they are supposed to be unsigned, casting them to a signed
value for all atomic operations is as errorprone as handling
them as signed entities.
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
* Resolves#3705
* Components should link against the project level library to better
support `dlopen` with `RTLD_LOCAL`.
* Extend the `mca_FRAMEWORK_COMPONENT_la_LIBADD` in the `Makefile.am`
with the appropriate project level library:
```
MCA components in ompi/
$(top_builddir)/ompi/lib@OMPI_LIBMPI_NAME@.la
MCA components in orte/
$(top_builddir)/orte/lib@ORTE_LIB_PREFIX@open-rte.la
MCA components in opal/
$(top_builddir)/opal/lib@OPAL_LIB_PREFIX@open-pal.la
MCA components in oshmem/
$(top_builddir)/oshmem/liboshmem.la"
```
Note: The changes in this commit were automated by the script in
the commit that proceeds it with the `libadd_mca_comp_update.py`
script. Some components were not included in this change because
they are statically built only.
Signed-off-by: Joshua Hursey <jhursey@us.ibm.com>
This commit fixes a bug that occurs when the btl callback happens before
the rget returns. In this case the fragment has been returned and is no
longer valid. This commit saves the size before calling rget. This is
valid since the BTL is not allowed to change the read size.
Fixes#3821
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
Passed the below set of symbols into a script that added ompi_ to them all.
Note that if processing a symbol named "foo" the script turns
foo into ompi_foo
but doesn't turn
foobar into ompi_foobar
But beyond that the script is blind to C syntax, so it hits strings and
comments etc as well as vars/functions.
coll_base_comm_get_reqs
comm_allgather_pml
comm_allreduce_pml
comm_bcast_pml
fcoll_base_coll_allgather_array
fcoll_base_coll_allgatherv_array
fcoll_base_coll_bcast_array
fcoll_base_coll_gather_array
fcoll_base_coll_gatherv_array
fcoll_base_coll_scatterv_array
fcoll_base_sort_iovec
mpit_big_lock
mpit_init_count
mpit_lock
mpit_unlock
netpatterns_base_err
netpatterns_base_verbose
netpatterns_cleanup_narray_knomial_tree
netpatterns_cleanup_recursive_doubling_tree_node
netpatterns_cleanup_recursive_knomial_allgather_tree_node
netpatterns_cleanup_recursive_knomial_tree_node
netpatterns_init
netpatterns_register_mca_params
netpatterns_setup_multinomial_tree
netpatterns_setup_narray_knomial_tree
netpatterns_setup_narray_tree
netpatterns_setup_narray_tree_contigous_ranks
netpatterns_setup_recursive_doubling_n_tree_node
netpatterns_setup_recursive_doubling_tree_node
netpatterns_setup_recursive_knomial_allgather_tree_node
netpatterns_setup_recursive_knomial_tree_node
pml_v_output_close
pml_v_output_open
intercept_extra_state_t
odls_base_default_wait_local_proc
_event_debug_mode_on
_evthread_cond_fns
_evthread_id_fn
_evthread_lock_debugging_enabled
_evthread_lock_fns
cmd_line_option_t
cmd_line_param_t
crs_base_self_checkpoint_fn
crs_base_self_continue_fn
crs_base_self_restart_fn
event_enable_debug_output
event_global_current_base_
event_module_include
eventops
sync_wait_mt
trigger_user_inc_callback
var_type_names
var_type_sizes
Signed-off-by: Mark Allen <markalle@us.ibm.com>
Add a monitoring PML, OSC and IO. They track all data exchanges between processes,
with capability to include or exclude collective traffic. The monitoring infrastructure is
driven using MPI_T, and can be tuned of and on any time o any communicators/files/windows.
Documentations and examples have been added, as well as a shared library that can be
used with LD_PRELOAD and that allows the monitoring of any application.
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* add ability to querry pml monitorinting results with MPI Tools interface
using performance variables "pml_monitoring_messages_count" and
"pml_monitoring_messages_size"
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
* Fix a convertion problem and add a comment about the lack of component
retain in the new component infrastructure.
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
* Allow the pvar to be written by invoking the associated callback.
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
* Various fixes for the monitoring.
Allocate all counting arrays in a single allocation
Don't delay the initialization (do it at the first add_proc as we
know the number of processes in MPI_COMM_WORLD)
Add a choice: with or without MPI_T (default).
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
* Cleanup for the monitoring module.
Fixed few bugs, and reshape the operations to prepare for
global or communicator-based monitoring. Start integrating
support for MPI_T as well as MCA monitoring.
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
* Adding documentation about how to use pml_monitoring component.
Document present the use with and without MPI_T.
May not reflect exactly how it works right now, but should reflects
how it should work in the end.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Change rank into MPI_COMM_WORLD and size(MPI_COMM_WORLD) to global variables in pml_monitoring.c.
Change mca_pml_monitoring_flush() signature so we don't need the size and rank parameters.
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
* Improve monitoring support (including integration with MPI_T)
Use mca_pml_monitoring_enable to check status state. Set mca_pml_monitoring_current_filename iif parameter is set
Allow 3 modes for pml_monitoring_enable_output: - 1 : stdout; - 2 : stderr; - 3 : filename
Fix test : 1 for differenciated messages, >1 for not differenciated. Fix output.
Add documentation for pml_monitoring_enable_output parameter. Remove useless parameter in example
Set filename only if using mpi tools
Adding missing parameters for fprintf in monitoring_flush (for output in std's cases)
Fix expected output/results for example header
Fix exemple when using MPI_Tools : a null-pointer can't be passed directly. It needs to be a pointer to a null-pointer
Base whether to output or not on message count, in order to print something if only empty messages are exchanged
Add a new example on how to access performance variables from within the code
Allocate arrays regarding value returned by binding
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add overhead benchmark, with script to use data and create graphs out of the results
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix segfault error at end when not loading pml
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Start create common monitoring module. Factorise version numbering
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix microbenchmarks script
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Improve readability of code
NULL can't be passed as a PVAR parameter value. It must be a pointer to NULL or an empty string.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add osc monitoring component
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add error checking if running out of memory in osc_monitoring
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Resolve brutal segfault when double freeing filename
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Moving to ompi/mca/common the proper parts of the monitoring system
Using common functions instead of pml specific one. Removing pml ones.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add calls to record monitored data from osc. Use common function to translate ranks.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix test_overhead benchmark script distribution
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix linking library with mca/common
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add passive operations in monitoring_test
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix from rank calculation. Add more detailed error messages
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix alignments. Fix common_monitoring_get_world_rank function. Remove useless trailing new lines
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix osc_monitoring mget_message_count function call
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Change common_monitoring function names to respect the naming convention. Move to common_finalize the common parts of finalization. Add some comments.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add monitoring common output system
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add error message when trying to flush to a file, and open fails. Remove erroneous info message when flushing wereas the monitoring is already disabled.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Consistent output file name (with and without MPI_T).
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Always output to a file when flushing at pvar_stop(flush).
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Update the monitoring documentation.
Complete informations from HowTo. Fix a few mistake and typos.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Use the world_rank for printf's.
Fix name generation for output files when using MPI_T. Minor changes in benchmarks starting script
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Clean potential previous runs, but keep the results at the end in order to potentially reprocess the data. Add comments.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add security check for unique initialization for osc monitoring
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Clean the amout of symbols available outside mca/common/monitoring
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Remove use of __sync_* built-ins. Use opal_atomic_* instead.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Allocate the hashtable on common/monitoring component initialization. Define symbols to set the values for error/warning/info verbose output. Use opal_atomic instead of built-in function in osc/monitoring template initialization.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Deleting now useless file : moved to common/monitoring
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add histogram ditribution of message sizes
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add histogram array of 2-based log of message sizes. Use simple call to reset/allocate arrays in common_monitoring.c
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add informations in dumping file. Separate per category (pt2pt/osc/coll (to come)) monitored data
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add coll component for collectives communications monitoring
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix warning messages : use c_name as the magic id is not always defined. Moreover, there was a % missing. Add call to release underlying modules. Add debug info messages. Add warning which may lead to further analysis.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix log10_2 constant initialization. Fix index calculation for histogram array.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add debug info messages to follow more easily initialization steps.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Group all the var/pvar definitions to common_monitoring. Separate initial filename from the current on, to ease its lifetime management. Add verifications to ensure common is initialized once only. Move state variable management to common_monitoring.
monitoring_filter only indicates if filtering is activated.
Fix out of range access in histogram.
List is not used with the struct mca_monitoring_coll_data_t, so heritate only from opal_object_t.
Remove useless dead code.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix invalid memory allocation. Initialize initial_filename to empty string to avoid invalid read in mca_base_var_register.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Don't install the test scripts.
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix missing procs in hashtable. Cache coll monitoring data.
* Add MCA_PML_BASE_FLAG_REQUIRE_WORLD flag to the PML layer.
* Cache monitoring data relative to collectives operations on creation.
* Remove double caching.
* Use same proc name definition for hash table when inserting and
when retrieving.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Use intermediate variable to avoid invalid write while retrieving ranks in hashtable.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add missing release of the last element in flush_all. Add release of the hashtable in finalize.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Use a linked list instead of a hashtable to keep tracks of communicator data. Add release of the structure at finalize time.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Set world_rank from hashtable only if found
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Use predefined symbol from opal system to print int
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Move collective monitoring data to a hashtable. Add pvar to access the monitoring_coll_data. Move functions header to a private file only to be used in ompi/mca/common/monitoring
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix pvar registration. Use OMPI_ERROR isntead of -1 as returned error value. Fix releasing of coll_data_t objects. Affect value only if data is found in the hashtable.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add automated check (with MPI_Tools) of monitoring.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix procs list caching in common_monitoring_coll_data_t
* Fix monitoring_coll_data type definition.
* Use size(COMM_WORLD)-1 to determine max number of digits.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add linking to Fortran applications for LD_PRELOAD usage of monitoring_prof
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add PVAR's handles. Clean up code (visibility, add comments...). Start updating the documentation
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix coll operations monitoring. Update check_monitoring accordingly to the added pvar. Fix monitoring array allocation.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Documentation update.
Update and then move the latex and README documentation to a more logical place
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Aggregate monitoring COLL data to the generated matrix. Update documentation accordingly.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix monitoring_prof (bad variable.vector used, and wrong array in PMPI_Gather).
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add reduce_scatter and reduce_scatter_block monitoring. Reduce memory footprint of monitoring_prof. Unify OSC related outputs.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add the use of a machine file for overhead benchmark
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Check for out-of-bound write in histogram
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Fix common_monitoring_cache object init for MPI_COMM_WORLD
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add RDMA benchmarks to test_overhead
Add error file output. Add MPI_Put and MPI_Get results analysis. Add overhead computation for complete sending (pingpong / 2).
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add computation of average and median of overheads. Add comments and copyrigths to the test_overhead script
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add technical documentation
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Adapt to the new definition of communicators
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Update expected output in test/monitoring/monitoring_test.c
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add dumping histogram in edge case
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Adding a reduce(pml_monitoring_messages_count, MPI_MAX) example
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add consistency in header inclusion.
Include ompi/mpi/fortran/mpif-h/bindings.h only if needed.
Add sanity check before emptying hashtable.
Fix typos in documentation.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* misc monitoring fixes
* test/monitoring: fix test when weak symbols are not available
* monitoring: fix a typo and add a missing file in Makefile.am
and have monitoring_common.h and monitoring_common_coll.h included in the distro
* test/monitoring: cleanup all tests and make distclean a happy panda
* test/monitoring: use gettimeofday() if clock_gettime() is unavailable
* monitoring: silence misc warnings (#3)
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
* Cleanups.
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
* Changing int64_t to size_t.
Keep the size_t used accross all monitoring components.
Adapt the documentation.
Remove useless MPI_Request and MPI_Status from monitoring_test.c.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add parameter for RMA test case
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Clean the maximum bound computation for proc list dump.
Use ptrdiff_t instead of OPAL_PTRDIFF_TYPE to reflect the changes from commit fa5cd0dbe5.
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add communicator-specific monitored collective data reset
Signed-off-by: Clement Foyer <clement.foyer@inria.fr>
* Add monitoring scripts to the 'make dist'
Also install them in the build and the install directories.
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
This commit adds code to allow support for the info assertions added
by mpi-forum/mpi-issues#11. The assertions added are:
mpi_assert_no_any_tag, mpi_assert_no_any_source,
mpi_assert_exact_length, and mpi_assert_allow_overtaking.
This commit also adds support for the mpi_assert_no_any_source and
mpi_assert_allow_overtaking info keys to the ob1 pml.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
This commit adds the `req_start` member to the `ompi_request_t` struct.
The `MPI_START` and `MPI_STARTALL` routines call this callback function
instead of `MCA_PML_CALL(start(...))`. So components that return
persistent request must set this member to their request objects.
`mca_pml_base_module_t::pml_start` is not deleted because
`MCA_PML_CALL(start(...))` is still used elsewhere across OMPI.
Signed-off-by: KAWASHIMA Takahiro <t-kawashima@jp.fujitsu.com>
Yalla has a macro PML_YALLA_INIT_MXM_REQ_DATA that checks if a datatype
is contiguous via opal_datatype_is_contiguous_memory_layout(dt,count)
and if so it selects a size and lb that presumably is what will rdma, as
ompi_datatype_type_size(_dtype, &size); \
ompi_datatype_type_lb(_dtype, &lb); \
This failed when I gave it a datatype constructed as [ ...] with extent 4.
What I mean by that datatype is
lens[0] = 3;
disps[0] = 1;
types[0] = MPI_CHAR;
MPI_Type_struct(1, lens, disps, types, &tmpdt);
MPI_Type_create_resized(tmpdt, 0, 4, &mydt);
So there are 3 chars at offset 1, and the LB is 0 and the UB is 4.
So that macro decides that size=4 and lb=0 and later I suppose size is getting
updated to 3 for the final rdma, and so a send of a buffer
[ 0 1 2 3 ] gets recved as [ 0 1 2 _ ]. I think it should use the true lb
and the true extent.
For "regular" contig datatypes it would be the same, and for the irregular
ones that are still deemed contiguous by that utility function it should
still be the right thing to use.
Signed-off-by: Mark Allen <markalle@us.ibm.com>
This commit fixes a bug that disabled both the RDMA pipeline and RDMA
protocols in ob1. ob1 was internally caching the values of
opal_leave_pinned and opal_leave_pinned_pipeline at init time. This is
no longer valid as opal_leave_pinned may be set by any call to a btl's
add_procs.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
Memory hooks are now set-up on demand. pml/yalla, mtl/mxm and
coll/hcoll need the memory hooks, so make sure those are installed.
Signed-off-by: Yossi Itigin <yosefe@mellanox.com>
According to the MPI-3.1 p.52 and p.53 (cited below), a request
created by `MPI_*_INIT` but not yet started by `MPI_START` or
`MPI_STARTALL` is inactive therefore `MPI_WAIT` or its friends
must return immediately if such a request is passed.
The current implementation hangs in `MPI_WAIT` and its friends
in such case because a persistent request is initialized as
`req_complete = REQUEST_PENDING`. This commit fixes the
initialization.
Also, this commit fixes internal requests used in `MPI_PROBE`
and `MPI_IPROBE` which was marked wrongly as persistent.
MPI-3.1 p.52:
We shall use the following terminology: A null handle is a handle
with value MPI_REQUEST_NULL. A persistent request and the handle
to it are inactive if the request is not associated with any ongoing
communication (see Section 3.9). A handle is active if it is neither
null nor inactive. An empty status is a status which is set to return
tag = MPI_ANY_TAG, source = MPI_ANY_SOURCE, error = MPI_SUCCESS, and
is also internally configured so that calls to MPI_GET_COUNT,
MPI_GET_ELEMENTS, and MPI_GET_ELEMENTS_X return count = 0 and
MPI_TEST_CANCELLED returns false. We set a status variable to empty
when the value returned by it is not significant. Status is set in
this way so as to prevent errors due to accesses of stale information.
MPI-3.1 p.53:
One is allowed to call MPI_WAIT with a null or inactive request
argument. In this case the operation returns immediately with empty
status.
Signed-off-by: KAWASHIMA Takahiro <t-kawashima@jp.fujitsu.com>
`sturct mca_pml_ob1_comm_proc_t`, which is allocated per
connected rank in a communicator, had two paddings after
`expected_sequence` and `send_sequence` by alignments.
By changing the order of the members, the size of
`mca_pml_ob1_comm_proc_t` is reduced by 8 bytes on 64-bit
architectures.
Signed-off-by: KAWASHIMA Takahiro <t-kawashima@jp.fujitsu.com>
this is generally done in mca_pml_ob1_recv_request_free(), but this is not invoked
in via mca_pml_ob1_recv(), so do it manually
Thanks Yvan Fournier for the report
Signed-off-by: Gilles Gouaillardet <gilles@rist.or.jp>
recvreq->req_recv.req_base.req_type should always be set before invoking
MCA_PML_OB1_RECV_REQUEST_INIT(recvreq, ...) otherwise, the previous type
might be set, and you could end up with MPC_PML_REQUEST_IMPROBE when
MCA_PML_REQUEST_RECV is expected.
Thanks Chris Pattison for the report and test case.
Fixesopen-mpi/ompi#2275
- pass field_mask to ucp_init().
- use non-blocking disconnect.
- recv() with pre-allocated request.
- call opal_progress() from iprobe() and improbe().
- use shift pattern in connect/disconnect.
This commit fixes an issue that can occur if a target gets overwhelmed with
requests. This can cause osc/pt2pt to go into deep recursion with a stack
like req_complete_cb -> ompi_osc_pt2pt_callback -> start -> req_complete_cb
-> ... . At small scale this is fine as the recursion depth stays small but
at larger scale we can quickly exhaust the stack processing frag requests.
To fix the issue the request callback now simply puts the request on a
list and returns. The osc/pt2pt progress function then handles the
processing and reposting of the request.
As part of this change osc/pt2pt can now post multiple fragment receive
requests per window. This should help prevent a target from being overwhelmed.
Signed-off-by: Nathan Hjelm <hjelmn@me.com>
This commit updates the btl selection logic for the RDMA and RDMA
pipeline protocols to use a btl iff: 1) the btl is also used for eager
messages (high exclusivity), or 2) no other RDMA btl is available on
an endpoint and the pml_ob1_use_all_rdma MCA variable is true. This
fixes a performance regression with shared memory when an RDMA capable
network is available.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
On start we were not correctly resetting all request fields. This was
leading to a double-completion on persistent receives. This commit
updates the base start code to reset the receive req_bytes_packed and
the send request convertor.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
It was introduced in PR https://github.com/open-mpi/ompi/pull/1228
in particular in commit 041a6a9f53.
Original solution was using "flexible array member" called "mxm_base"
to "fall-through" to the "mxm" send/recv member that located in the
outer structure.
After changing number of elements in "mxm_base" from 0 to 1 we actually
allocating 2 mxm_req_base_t elements which leads to increased overal
size and harms cache performance.
It also brakes "mca_pml_yalla_check_request_state" function.
This commit fixes two bugs in pml/ob1:
- Do not called MCA_PML_OB1_PROGRESS_PENDING from
mca_pml_ob1_send_request_start_copy as this may lead to a recursive
call to mca_pml_ob1_send_request_process_pending.
- In mca_pml_ob1_send_request_start_rdma return the rdma frag object
if a btl fragment can not be allocated. This fixes a leak
identified by @abouteiller and @bosilca.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
* mpi/start: fix bugs in cm and ob1 start functions
There were several problems with the implementation of start in Open
MPI:
- There are no checks whatsoever on the state of the request(s)
provided to MPI_Start/MPI_Start_all. It is erroneous to provide an
active request to either of these calls. Since we are already
looping over the provided requests there is little overhead in
verifying that the request can be started.
- Both ob1 and cm were always throwing away the request on the
initial call to start and start_all with a particular
request. Subsequent calls would see that the request was
pml_complete and reuse it. This introduced a leak as the initial
request was never freed. Since the only pml request that can
be mpi complete but not pml complete is a buffered send the
code to reallocate the request has been moved. To detect that
a request is indeed mpi complete but not pml complete isend_init
in both cm and ob1 now marks the new request as pml complete.
- If a new request was needed the callbacks on the original request
were not copied over to the new request. This can cause osc/pt2pt
to hang as the incoming message callback is never called.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
* osc/pt2pt: add request for gc after starting a new request
Starting a new receive may cause a recursive call into the pt2pt
frag receive function. If this happens and the prior request is
on the garbage collection list it could cause problems. This commit
moves the gc insert until after the new request has been posted.
Signed-off-by: Nathan Hjelm <hjelmn@me.com>
The request code was setting the request as pml_complete before
calling MCA_PML_OB1_SEND_REQUEST_MPI_COMPLETE. This was causing
MCA_PML_OB1_SEND_REQUEST_RETURN to be called twice in some cases. The
code now mirrors the recvreq code and only sets the request as pml
complete if the request has not already been freed.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
Regarding BFO it should be mentionned that this component is currently
unmaintained, and that despite my efforts I could not make it compile
(it would not compile before this patch either).
This fixes a hang caused by the request refactor work. The cm pml was
not updated and was hanging is most cases.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
* Remodel the request.
Added the wait sync primitive and integrate it into the PML and MTL
infrastructure. The multi-threaded requests are now significantly
less heavy and less noisy (only the threads associated with completed
requests are signaled).
* Fix the condition to release the request.
If during the request completion callback we post another request that
completes right away (such a small send or a match for an unexpected
short message) we will try to complete the second request while holding
the lock for the completion of the first. For performance reasons
(mainly to avoid unlocking and locking the request mutex several times)
we have made the request lock recursive.
