All releases that support -march=broadwell cause segfault on following polyhedron test-case: gfortran -Ofast -g -march=broadwell test_fpu2.f90 -o b.out && gdb ./b.out ... Program received signal SIGSEGV, Segmentation fault. (gdb) bt #0 0x0000000000402123 in crout (a=<error reading variable: value requires 32000000 bytes, which is more than max-value-size>, n=2000) at test_fpu2.f90:168 #1 0x000000000040e555 in test_fpu () at test_fpu2.f90:79 #2 0x000000000040118d in main (argc=argc@entry=1, argv=0x7fffffffe428) at test_fpu2.f90:12 #3 0x00007ffff7435feb in __libc_start_main (main=0x401170 <main>, argc=1, argv=0x7fffffffe038, init=<optimized out>, fini=<optimized out>, rtld_fini=<optimized out>, stack_end=0x7fffffffe028) at ../csu/libc-start.c:308 #4 0x00000000004011ea in _start () at ../sysdeps/x86_64/start.S:120 │0x402120 <crout_+272> add %rsi,%rcx >│0x402123 <crout_+275> vmovdqu %ymm0,(%rdx) │0x402127 <crout_+279> add $0x20,%rdx │0x40212b <crout_+283> vpaddd %ymm1,%ymm0,%ymm0 Maybe Richi is interested in that? -O3 works fine; I can reproduce that on Haswell using the march option value
Issue is as simple as missing ulimit -s unlimited.