Test failure with GCC/Linux on non-x86 architectures at -O2

[musicinmybrain]

Originally found while enabling tests in the python-tvb-dist package in Fedora Linux.

This can also be reproduced on a free-tier Ubuntu 22.04.1 LTS (GNU/Linux 5.15.0-1027-oracle aarch64) VM on Oracle Cloud Infrastructure as follows:

$ sudo apt install git g++ libgtest-dev
$ git clone https://github.com/the-virtual-brain/tvb-gdist.git
$ cd tvb-gdist/tests
$ g++ -O2 -g test_geodesic_utils.cpp -o test_geodesic_utils -lgtest
$ ./test_geodesic_utils 
[==========] Running 1 test from 1 test suite.
[----------] Global test environment set-up.
[----------] 1 test from compute_gdist_impl
[ RUN      ] compute_gdist_impl.flat_traingular_mesh_test

mesh has 121 vertices, 200 faces, 320 edges
40 edges are boundary edges
shortest/longest edges are 0.2/0.282843 = 0.707107
enclosing XYZ box: X[1.5,3.5] Y[-1,1] Z[0,0]
approximate diameter of the mesh is 2.82843
min/max face angles are 45/90 degrees

test_geodesic_utils: ../geodesic_library/geodesic_algorithm_exact.h:1166: void geodesic::GeodesicAlgorithmExact::construct_propagated_intervals(bool, geodesic::edge_pointer, geodesic::face_pointer, geodesic::IntervalWithStop*, unsigned int&, geodesic::interval_pointer): Assertion `p->start() < p->stop()' failed.
Aborted (core dumped)

(I mention this because it’s an easy place to get free interactive access to an aarch64/Linux VM.)

GDB backtrace:

#0  __pthread_kill_implementation (threadid=281474842429984, signo=signo@entry=6, no_tid=no_tid@entry=0)
    at ./nptl/pthread_kill.c:44
#1  0x0000fffff7b8f254 in __pthread_kill_internal (signo=6, threadid=<optimized out>) at ./nptl/pthread_kill.c:78
#2  0x0000fffff7b4a67c in __GI_raise (sig=sig@entry=6) at ../sysdeps/posix/raise.c:26
#3  0x0000fffff7b37130 in __GI_abort () at ./stdlib/abort.c:79
#4  0x0000fffff7b43fd0 in __assert_fail_base (fmt=0xfffff7c5d3f8 "%s%s%s:%u: %s%sAssertion `%s' failed.\n%n", 
    assertion=assertion@entry=0xaaaaaaae9f78 "p->start() < p->stop()", 
    file=file@entry=0xaaaaaaae9680 "../geodesic_library/geodesic_algorithm_exact.h", line=line@entry=1166, 
    function=function@entry=0xaaaaaaae9e28 "void geodesic::GeodesicAlgorithmExact::construct_propagated_intervals(bool, geodesic::edge_pointer, geodesic::face_pointer, geodesic::IntervalWithStop*, unsigned int&, geodesic::interval_pointer)") at ./assert/assert.c:92
#5  0x0000fffff7b44040 in __GI___assert_fail (assertion=0xaaaaaaae9f78 "p->start() < p->stop()", 
    file=0xaaaaaaae9680 "../geodesic_library/geodesic_algorithm_exact.h", line=1166, 
    function=0xaaaaaaae9e28 "void geodesic::GeodesicAlgorithmExact::construct_propagated_intervals(bool, geodesic::edge_pointer, geodesic::face_pointer, geodesic::IntervalWithStop*, unsigned int&, geodesic::interval_pointer)")
    at ./assert/assert.c:101
#6  0x0000aaaaaaaacd24 in geodesic::GeodesicAlgorithmExact::construct_propagated_intervals (
    this=this@entry=0xffffffffecd8, invert=<optimized out>, edge=edge@entry=0xaaaaaab34170, 
    face=face@entry=0xaaaaaab2e220, candidates=candidates@entry=0xffffffffeac8, num_candidates=@0xffffffffeac4: 1, 
    source_interval=source_interval@entry=0xaaaaaab40450) at ../geodesic_library/geodesic_algorithm_exact.h:1166
#7  0x0000aaaaaaab6b9c in geodesic::GeodesicAlgorithmExact::propagate (this=this@entry=0xffffffffecd8, 
    sources=std::vector of length 1, capacity 1 = {...}, 
    max_propagation_distance=max_propagation_distance@entry=1e+100, stop_points=stop_points@entry=0xffffffffec30)
    at ../geodesic_library/geodesic_algorithm_exact.h:648
#8  0x0000aaaaaaaa9bf0 in compute_gdist_impl (points=std::vector of length 363, capacity 363 = {...}, 
    faces=std::vector of length 600, capacity 600 = {...}, sources=std::vector of length 1, capacity 1 = {...}, 
    targets=std::vector of length 1, capacity 1 = {...}, max_distance=max_distance@entry=1e+100, 
    is_one_indexed=is_one_indexed@entry=false, propagate_on_max_distance=propagate_on_max_distance@entry=false)
    at ../geodesic_library/geodesic_utils.h:29
#9  0x0000aaaaaaaaa22c in compute_gdist_impl_flat_traingular_mesh_test_Test::TestBody (this=<optimized out>)
    at /usr/include/c++/11/bits/stl_algobase.h:420
#10 0x0000aaaaaaae8bf0 in void testing::internal::HandleExceptionsInMethodIfSupported<testing::Test, void>(testing::Test*, void (testing::Test::*)(), char const*) ()
#11 0x0000aaaaaaadb5cc in testing::Test::Run() ()
#12 0x0000aaaaaaadb750 in testing::TestInfo::Run() ()
#13 0x0000aaaaaaadbd50 in testing::TestSuite::Run() ()
#14 0x0000aaaaaaadc47c in testing::internal::UnitTestImpl::RunAllTests() ()
#15 0x0000aaaaaaadb89c in testing::UnitTest::Run() ()
#16 0x0000aaaaaaaa8b6c in RUN_ALL_TESTS () at /usr/include/gtest/gtest.h:2490
#17 main (argc=<optimized out>, argv=<optimized out>) at test_geodesic_utils.cpp:32

The assertion failure does not occur with -O1, -Og, or even -O3. It also did not occur at -O2 with in a quick test with clang++ on Apple Silicon.

Please let me know if there is anything else I can do to help diagnose this.

[musicinmybrain]

I’m seeing the same thing on ppc64le and s390x (but not on i686 or x86_64).

[maedoc]

@musicinmybrain thanks for the report, might be numerical error, but would need to step through those lines of code in gdb.