https://gcc.gnu.org/bugzilla/show_bug.cgi?id=100497
Bug ID: 100497 Summary: [OpenMP][nvptx] libgomp.c-c++-common/reduction-5.c - fails on some nvptx systems Product: gcc Version: 12.0 Status: UNCONFIRMED Keywords: openmp, wrong-code Severity: normal Priority: P3 Component: target Assignee: unassigned at gcc dot gnu.org Reporter: burnus at gcc dot gnu.org CC: vries at gcc dot gnu.org Target Milestone: --- The two testcases libgomp.c-c++-common/reduction-5.c libgomp.c-c++-common/reduction-6.c fail on some nvptx systems. (Added in r12-614-g33b647956caa977d1ae489f9baed9cef70b4f382 ) The problem are the 'teams' versions with complex arguments; the non-teams ones do work - the problem is that the complex variables are '0 + 0i' instead of '1 + 0i' - hence, those tests which expect 0 also pass. For some reason, it fails on half of the tested systems but not on the other half. The PASS/FAIL are consistent; it does not seem to depend on the CUDA version, it might depend on the used hardware or it might work only by chance due to the way the partitioning is done.