The rdma_frag attached to the send request was not correctly released
upon request completion, leaking until MPI_Finalize. A quick solution
would have been to add RDMA_FRAG_RETURN at different locations on the
send request completion, but it would have unnecessarily made the
sendreq completion path more complex. Instead, I added the length to
the RDMA fragment so that it can be completed during the remote ack.
Be more explicit on the comment.
The rdma_frag can only be freed once when the peer forced a protocol
change (from RDMA GET to send/recv). Otherwise the fragment will be
returned once all data pertaining to it has been trasnferred.
Signed-off-by: George Bosilca <bosilca@icl.utk.edu>
In case of using a btl_put in ob1, the handle of the locally registered
memory is sent with a PUT control message. In the current master code
the sent handle is necessary the handle in the frag but if the handle
has been successfully registered in the request, the frag structure does
not have any valid handle and all fragments use the request one.
I suggest to check if the handle in the fragment is valid and if not to
send the handle from the request.
Signed-off-by: Brelle Emmanuel <emmanuel.brelle@atos.net>
In the case the btl_get fails Ob1 tries to fallback on btl_put first but
the return code was ignored. So the code fell back on both btl_put and
btl_send.
Signed-off-by: Brelle Emmanuel <emmanuel.brelle@atos.net>
This commit updates the new custom matching code in pml/ob1 so it can
not be enabled with a configure option. This commit also renames the
fuzzy-matching headers to avoid potential name conflicts and removes
the use of C reserved identifiers.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
This code is the implementation of Software-base Performance Counters as described in the paper 'Using Software-Base Performance Counters to Expose Low-Level Open MPI Performance Information' in EuroMPI/USA '17 (http://icl.cs.utk.edu/news_pub/submissions/software-performance-counters.pdf). More practical usage information can be found here: https://github.com/davideberius/ompi/wiki/How-to-Use-Software-Based-Performance-Counters-(SPCs)-in-Open-MPI.
All software events functions are put in macros that become no-ops when SOFTWARE_EVENTS_ENABLE is not defined. The internal timer units have been changed to cycles to avoid division operations which was a large source of overhead as discussed in the paper. Added a --with-spc configure option to enable SPCs in the Open MPI build. This defines SOFTWARE_EVENTS_ENABLE. Added an MCA parameter, mpi_spc_enable, for turning on specific counters. Added an MCA parameter, mpi_spc_dump_enabled, for turning on and off dumping SPC counters in MPI_Finalize. Added an SPC test and example.
Signed-off-by: David Eberius <deberius@vols.utk.edu>
This commit renames the arithmetic atomic operations in opal to
indicate that they return the new value not the old value. This naming
differentiates these routines from new functions that return the old
value.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
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>
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>
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>
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>
* 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.
Added mca parameter to turn progress thread on/off
Add a flag to check if we have btl progress thread.
Added macro for ob1 matching lock.
Update the AUTHORS file.
Fix CID 1315271: Constant expression result
The intent of this conditional is to not produce a peruse event for
probe or mprobe requests. Coverity is correct that the expression is
always true. Changed the || to && to fix. Also moved the conditional
within an OMPI_WANT_PERUSE to ensure the conditional is not evaluated
if peruse is disabled.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
This commit contains the following changes:
- pml/ob1: use the bml accessor function when requesting a bml
endpoint. this will ensure that bml endpoints are only created when
needed. for example, a bml endpoint is not requested and not
allocated when receiving an eager message from a peer.
- pml/ob1: change the pml_procs array in the ob1 communicator to a
proc pointer array. at the cost of a single level of extra
redirection this will allow us to allocate pml procs on demand.
- pml/ob1: add an accessor function to access the pml proc structure
for a given peer. this function will allocate the proc if it
doesn't already exist.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
This commit should resolve an issue seen with CUDA-aware support. The
problem came in with BTL 3.0. Before 3.0 the size of the copy was
stored in the incoming segment's des_remote_count field. This field
does not exist in BTL 3.0 so I stored the value in the
des_segment_count field. This caused problems with the cuda support
code. To fix the issue the endpoint pointer is now stored in the in
fragment's endpoint pointer which free's up the segment's des_cbdata
pointer for storing the transfer size.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
A little background. Historically ob1 always registered the entire memory
region when the RGET protocol was in use. This changed when Mellanox
added support to fragment RGET using the btl_prepare_dst function. Now
that the BTL layer has changed to split out the limits of get/put there
is explicit fragmentation code in ob1. Before this commit the registration
was still done per RGET fragment.
This commit will attempt to register the entire region before creating
RGET fragments. If the registration is successfull then all RGET
fragments will use this registration otherwise they will each attempt
to register their own segment of the receive buffer. If that fails
enough times each fragment will give up and fall back on send/recv.
Signed-off-by: Nathan Hjelm <hjelmn@lanl.gov>
attempting to break the get into multiple rdma fragments
A little background. Historically ob1 always registered the entire memory
region when the RGET protocol was in use. This changed when Mellanox
added support to fragment RGET using the btl_prepare_dst function. Now
that the BTL layer has changed to split out the limits of get/put there
is explicit fragmentation code in ob1. Before this commit the registration
was still done per RGET fragment.
This commit will attempt to register the entire region before creating
RGET fragments. If the registration is successfull then all RGET
fragments will use this registration otherwise they will each attempt
to register their own segment of the receive buffer. If that fails
enough times each fragment will give up and fall back on send/recv.
opal_mutex_t must be OBJ_DESTRUCTed in order to avoid
a memory leak (pthread_mutex_init allocates memory under
Cygwin, so pthread_mutex_destroy is mandatory)
Thanks to Marco Atzeri for reporting this issue
WHAT: Open our low-level communication infrastructure by moving all necessary components (btl/rcache/allocator/mpool) down in OPAL
All the components required for inter-process communications are currently deeply integrated in the OMPI layer. Several groups/institutions have express interest in having a more generic communication infrastructure, without all the OMPI layer dependencies. This communication layer should be made available at a different software level, available to all layers in the Open MPI software stack. As an example, our ORTE layer could replace the current OOB and instead use the BTL directly, gaining access to more reactive network interfaces than TCP. Similarly, external software libraries could take advantage of our highly optimized AM (active message) communication layer for their own purpose. UTK with support from Sandia, developped a version of Open MPI where the entire communication infrastucture has been moved down to OPAL (btl/rcache/allocator/mpool). Most of the moved components have been updated to match the new schema, with few exceptions (mainly BTLs where I have no way of compiling/testing them). Thus, the completion of this RFC is tied to being able to completing this move for all BTLs. For this we need help from the rest of the Open MPI community, especially those supporting some of the BTLs. A non-exhaustive list of BTLs that qualify here is: mx, portals4, scif, udapl, ugni, usnic.
This commit was SVN r32317.
