gpec SIGSEGV at end of multithreaded run (ifx build) without a raised stack

[logan-nc]

Summary

gpec SIGSEGVs at the end of a multithreaded run unless the stack is raised, on an
ifx/MKL build. The computation completes; the fault is in the final output/deallocation
phase.

Environment

• Compiler/libs: Intel oneAPI ifx/icx 2026.0 + MKL 2026.0, built with make FFLAGS="-O3 -qopenmp" OMPFLAG="-qopenmp".
• Case: DIII-D-like reference equilibrium (hamada, ldp, mpsi=256, psihigh=0.995), COIL Biot–Savart forcing, OMP_NUM_THREADS=8.

Symptom

The run reaches 100% ("Computing x and b normal components"), then:

forrtl: severe (174): SIGSEGV, segmentation fault occurred
Image  PC  Routine  Line  Source
libc.so.6  ...  Unknown  Unknown  Unknown
gpec      ...  Unknown  Unknown  Unknown

user time ~845 s vs real ~157 s (so ~8 OMP threads active). No SIGSEGV in the physics
loop; it faults during output/teardown.

Workaround (confirms cause)

Setting ulimit -s unlimited and KMP_STACKSIZE=1G (note: OMP_STACKSIZE is ignored
once KMP_STACKSIZE is set — a warning is printed) gives a clean run:

GPEC STOP => Normal termination.

This points at large automatic arrays / per-thread stack overflow.

Suggested fix

Either compile with -heap-arrays (or move the large local arrays in the output/reconstruction
path to allocatable/heap), or document the ulimit -s unlimited + KMP_STACKSIZE requirement
in the install/run docs so multithreaded runs don't silently segfault on default stacks.