Another segmentation fault with compiled nvfortran code

[juergen.reuter]

This is another test case from the WHIZARD test suite,which segfaults when compiled with nvfortran, but works with gfortran, Intel, and NAG compilers.
seg_fault_case.f90 (11.8 KB)

(gdb) run
Starting program: <...>/whizard_test 
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
 Mode = 0
 calling importing component

Program received signal SIGSEGV, Segmentation fault.
0x00007ffff5e7d47e in __GI___libc_free (mem=0x547834386a81b1f1) at ./malloc/malloc.c:3368
3368	./malloc/malloc.c: No such file or directory.
(gdb) bt
#0  0x00007ffff5e7d47e in __GI___libc_free (mem=0x547834386a81b1f1) at ./malloc/malloc.c:3368
#1  0x0000000000409c15 in process_configurations::process_configuration_setup_component () at main_ut.f90:363
#2  0x000000000040bb4f in process_configurations_uti::prepare_test_library () at main_ut.f90:398
#3  0x000000000040cb05 in main_ut () at main_ut.f90:410

[juergen.reuter]

This is another isolated single case from our main testsuite, posted here:

Collection of nvfortran problems in our code

The following tarball contains all our unit tests that fail with the nvfortran 23.1 (but do work with ifort, gfortran and nagfor). If you unpack, and do make, then you can check the individual tests with ./whizard_check --check testXX (with XX from 01 to 44) whizard-230404-2235.tar.gz (1.9 MB) {note that originally in this file the reference files for the tests were missing, now they are there, and I even reduced the code a tiny bit) I will try to break this up into individual cases. The ref…

[MatColgrove]

Thanks Juergen, I’ve filed TPR #33503 to report this issue.

Note that I was unable to reproduce the seg fault, but rather see a double free when exiting the “process_configuration_setup_component” subroutine. Though likely it’s the same root cause.

-Mat