ompi

ompi copied to clipboard

Fixes the following: ** CID 1675024: Memory - illegal accesses (BUFFER_SIZE) ** CID 1675023: Null pointer dereferences (REVERSE_INULL) ** CID 1675022: Error handling issues (CHECKED_RETURN) ** CID 1675021: Memory -...

hjelmn

with clang 21.1.6 as wrapper compiler for cuda-aware openmpi-5.0.9 and compile options -fopenmp and -fopenmp-targets=nvptx64-nvidia-cuda one gets the message "The call to cuMemHostRegister(0x7fb38c000000, 134217728, 0) failed." for every code with OpenMPI

15

Hi [Update: the problem also appears for open-mpi-5.0.9] I am using cuda-aware open-mpi-5.0.9 I want to write applications that use OpenMPI to send data to different nodes with several gpu...

bschulz81

coll/han: set as default except if processes are consecutive across nodes

6

`coll/han` provides better latency than `coll/tuned` in cases where processes are mapped to nodes nonconsecutively, e.g., using `--rank-by node`. In that case `coll/han` reduces the amount of cross-node traffic. Its...

devreal

`srun` with OpenMPI 5.0.3 unexpectedly launches MPI jobs as singletons without ERROR

1

As an OpenMPI user, I noticed unexpected behavior when running MPI programs with Slurm’s `srun`. - **Environment:** `module load OpenMPI/5.0.3` - **What happens:** **Using `mpirun`** (works as expected) ``` Hello...

GayatriManda

Feature: Provide network hardware counter summary at completion time

1

HPE's CrayMPI has a feature where during the call to MPI_Finalize, it can display or output to NIC-specific files a collection of network hardware performance counters and output them at...

sadlochr

[WIP] Add list of environment variables for libfabric cxi provider

1

Thanks for asking me to work on this. I will get started on it and keep this PR's description up to date as I form a plan and make progress....

Copilot

Revoke coll/han subcomms when a comm is revoked (and general framework)

1

Without a way to revoke the subcomms created by the coll modules, processes can get stuck waiting on sub-operations of a collective on a revoked communicator. I chose to put...

Matthew-Whitlock

Tested on a Slingshot 11 cluster with synthetic failures (SIGTERM). I'm not sure how consistent the error code will be across different libfabric backends or types of faults. It may...

Matthew-Whitlock

atomics: add the --disable-cx16-atomics

4

Do not try the -mcx16 flag if --disable-cx16-atomics is used, and prevent the generation of instructions that are not available on all x86 platforms (such as Celeron N4000). Always try...

ggouaillardet

Add Open Cluster Scheduler (fka SGE) support to documentation

2

dgruber