all_gatherv_test segfaults on s390x

[junghans]

66/68 Test #66: all_gatherv_test .................***Exception: SegFault 36.10 sec
Running 2 test cases...
�[1;32;49m*** No errors detected
�[0;39;49m
      Start 67: sendrecv_test
67/68 Test #67: sendrecv_test ....................   Passed    1.43 sec
      Start 68: PdbParser_test
68/68 Test #68: PdbParser_test ...................   Passed    0.00 sec
99% tests passed, 1 tests failed out of 68
Total Test time (real) = 131.09 sec
The following tests FAILED:
	 66 - all_gatherv_test (SEGFAULT)
Errors while running CTest

build_s390x.txt

[jngrad]

Segfault not reproducible with 4.1.2 + 48f0423 and 4.1.3rc, using boost 1.73.0 from the fedora:33 docker image as well as boost compiled from sources with boostorg/mpi#119, in a docker image emulated as s390x with QEMU. The unit tests themselves pass (the logfile says no errors were detected), and the test doesn't segfault with MPICH, so maybe the issue comes from OpenMPI 4.0.4?

[aminiussi]

Works fine with gcc 7.5.0 and openmpi-4.0.0

[aminiussi]

Works fine with gcc 7.5.0 and openmpi-4.0.4

[aminiussi]

And Intel's 2020 compiler and MPI

[jngrad]

@junghans We will soon release 4.1.3. Unless there is a need for a 4.1.4, this will be the last release where we test s390x. The s390x builds are disabled on Koji for espresso-4.1.2-4, espresso-4.1.2-5. Is this a temporary measure?

[junghans]

Yes, https://src.fedoraproject.org/rpms/espresso/blob/master/f/espresso.spec#_9

[RudolfWeeber]

We are only testing x64. I don't think, we can do something w/r to s3900.