Intel MPI Internal Crash on Local Multi-Rank Execution

[fairussiam55]

Issue:
Intel MPI 2021.15 (Build 20250213) crashes with SIGFPE when running any MPI program with more than 1 process locally.

System:

• Single-node setup (node1-X99)

• OS: Ubuntu-based

• MPI initialized via source /opt/intel/oneapi/setvars.sh

Command:

mpiexec -n 4 ./hello_mpi

Minimal code (hello_mpi.c):

#include <mpi.h>
#include <stdio.h>
int main(int argc, char** argv) {
MPI_Init(&argc, &argv);
int rank, size;
MPI_Comm_rank(MPI_COMM_WORLD, &rank);
MPI_Comm_size(MPI_COMM_WORLD, &size);
printf("Hello from rank %d of %d\n", rank, size);
MPI_Finalize();
return 0;
}

Error (strace output):

--- SIGFPE (Floating point exception) ---
+++ killed by SIGFPE (core dumped) +++

Notes:

Works with -n 1

Fails with fork, tcp, shm, and default launchers

/tmp permissions and environment are correct

Likely internal bug in the MPI runtime (e.g., division by zero)

[Deeksha_S_Intel]

Hi Mohd, we would like to inform you that we are routing your query to the engineering team for further assistance.

[taehunkim]

Hi,

The code you shared works well without any error as below.

Would you like to re-install oneAPI?

Thanks.