An NVHPC OpenMP target can still pick up CUDA tests unintentionally. I think for the CUDA tests below (line 152) we can add AND NOT OMP_TGT to prevent this for now. A longer-term fix will be explicit test options per backend instead of using compiler identities - we can do this separately.
Originally posted by @ia267 in #228 (comment)
An NVHPC OpenMP target can still pick up CUDA tests unintentionally. I think for the CUDA tests below (line 152) we can add
AND NOT OMP_TGTto prevent this for now. A longer-term fix will be explicit test options per backend instead of using compiler identities - we can do this separately.Originally posted by @ia267 in #228 (comment)