Bug #2366

HardwareTopologyTest.NumaCacheSelfconsistency fails on aarch64

Added by Christoph Junghans almost 2 years ago. Updated about 1 year ago.

Target version:
Affected version - extra info:
Affected version:


[ RUN      ] HardwareTopologyTest.NumaCacheSelfconsistency
/builddir/build/BUILD/gromacs-2018-beta3/src/gromacs/hardware/tests/hardwaretopology.cpp:192: Failure
Expected: (c.size) > (0), actual: 0 vs 0
/builddir/build/BUILD/gromacs-2018-beta3/src/gromacs/hardware/tests/hardwaretopology.cpp:192: Failure
Expected: (c.size) > (0), actual: 0 vs 0
[  FAILED  ] HardwareTopologyTest.NumaCacheSelfconsistency (7 ms)
[----------] 4 tests from HardwareTopologyTest (26 ms total)
[----------] Global test environment tear-down
[==========] 5 tests from 2 test cases ran. (27 ms total)
[  PASSED  ] 4 tests.
[  FAILED  ] 1 test, listed below:
[  FAILED  ] HardwareTopologyTest.NumaCacheSelfconsistency

found here:, build log attached.
build.log (34.4 MB) build.log Build log on aarch64 Christoph Junghans, 12/25/2017 11:11 PM


#1 Updated by Erik Lindahl almost 2 years ago

  • Status changed from New to Rejected

This seems to be a failure of hwloc on the platform in question, rather than Gromacs.

We've also tested it repeatedly on our own Aarch64 systems, and there it works fine.

#2 Updated by Erik Lindahl almost 2 years ago

NB: We don't use this information yet, so you can go right ahead and patch out this test if it's important for the build to pass all tests for the fedora distro.

#3 Updated by Erik Lindahl almost 2 years ago

If you want to file it with hwloc, here is their link to the hwloc-1.11.8 documentation on caches:

We do accept 0 (=unknown) both for line size and associativity), but I don't think 0 is a reasonable value for the total size of a cache given that their documentation says nothing about it (and these entries are only present for caches that actually exist).

#4 Updated by Christoph Junghans almost 2 years ago

Ok, I just wanted to check that there is nothing inherently wrong with aarch64.

#5 Updated by Christoph Junghans about 1 year ago

Seems to be fixed in Fedora 30.

Also available in: Atom PDF