summaryrefslogtreecommitdiff
path: root/llvm/lib/Bitcode/Reader/MetadataLoader.cpp
diff options
context:
space:
mode:
authoragozillon <Andrew.Gozillon@amd.com>2023-09-26 15:42:29 +0200
committerGitHub <noreply@github.com>2023-09-26 15:42:29 +0200
commit9971aefff8b947896ba8b497880bcb09a625d25f (patch)
tree26c8f51fa893070f886b17cb718edc97f58b6ceb /llvm/lib/Bitcode/Reader/MetadataLoader.cpp
parentf2f61a99f7f754f3e4c819e1e7c65848affe70b3 (diff)
[Flang][OpenMP][Driver][Test] Fix the omp-driver-offload-test commands (#66926)
Since the changes to -### to correctly return error codes in D156363, some additions of the not lit keyword and perhaps some changes to how fopenmp-targets is handled; the test omp-driver-offload.f90 has become a little susceptible to what appears to be sporadic failures. This was because if certain environment variables are set for AMD devices some of the commands succeed happily and the not check causes the test to fail. Whereas, if they were not set the tests would fail and the not test would succeed, allowing the test to pass provided you were not targeting AMD hardware while running the test. When the environment variables for AMD hardware are set the compiler could find the correct offload architecture to compile for, allowing the command to succeed. However, when not set it'd fail as it can't detect the offload architecture. The solution to this is to specify an AMD architecture via --offload-arch so it always passes, the architecture doesn't really matter, I've chosen to select gfx90a.
Diffstat (limited to 'llvm/lib/Bitcode/Reader/MetadataLoader.cpp')
0 files changed, 0 insertions, 0 deletions