oneapi-src / oneapi-src/ishmem
sporadic segfault when not using MPI_THREAD_MULTIPLE
Nobody has claimed this yet.
- Dominant language
- C++
- Stars
- 33
- Forks
- 9
- PR merge metrics
- No merged PRs in 30d
Description
Reproducer
https://github.com/ye-luo/gpu_shmem/tree/main/ishmem
In app.cpp, if the MPI initialization is changed from MPI_THREAD_MULTIPLE to
MPI_Init_thread(MPI_THREAD_FUNNELED)
MPI_Init()
I got segfault.
On Aurora/Sunspot with oneapi SDK 2025.3.1
$ mkdir build && cd build
$ cmake -DCMAKE_CXX_COMPILER=icpx -DCMAKE_CXX_FLAGS="-fiopenmp -fopenmp-targets=spir64" ..
$ make -j32
$ mpiexec -np 2 gpu_tile_compact.sh ./my_ishmem_app
x1921c1s0b0n0-hsn0.hsn.cm.sunspot.alcf.anl.gov: rank 9 died from signal 11
x1921c1s0b0n0-hsn0.hsn.cm.sunspot.alcf.anl.gov: rank 5 died from signal 15
Most applications actually don't use MPI_THREAD_MULTIPLE and requiring it significantly restrict the adoption of intel shmem.
Contributor guide
First steps
- Read the whole issue, then the project's contributing guide.
- Comment on the issue to say you are picking it up — it saves two people doing the same work.
- Fork the repository and make your change on a branch.
- Open a pull request that references the issue number.
Research direction
Start with the ishmem reproducer linked in the issue, especially app.cpp, and build it with the provided icpx, CMake, and mpiexec commands on the stated Aurora/Sunspot environment. Compare the MPI initialization using MPI_THREAD_MULTIPLE with the reported MPI_THREAD_FUNNELED change; done means the reproducer no longer segfaults without requiring MPI_THREAD_MULTIPLE.
Written by the indexing model from the issue text.
Assessment
- Tech stack
- cpp
- Domain
- distributed-systems
- Issue type
- Bug
- Difficulty
- 4/5
- Estimated time
- 3-5 days
- Activity status
- Stale
- Clarity
- Mostly clear
- Newbie friendliness
- 35/100