No target is specified in the test, so should OMPVV_TEST_OFFLOADING macro for offloading be removed or left alone?

[andrewkallai]

https://github.com/SOLLVE/sollve_vv/blob/3d38296009168344babf88fdf91a89e6b3a6e129/tests/5.1/scope/test_scope_nowait_construct.c#L61

[tob2]

Before commit a563b9e the following message was printed on Fortran, unless that check was present:
Test ... passed on the device
which was confusing because of "device"; one workaround was to add the "OMPVV_TEST_OFFLOADING;" check, the other is the fix above.

For C/C++, the OMPVV_REPORT message does not mention "on host/device" unless OMPVV_TEST_OFFLOADING or OMPVV_TEST_AND_SET_OFFLOADING was called, avoiding the confusion.

However, I assume the main reason for the proliferation of OMPVV_TEST_OFFLOADING is that the project started by offloading testing and then old files where taken as template for new files. — Therefore, the danger is to use a non-offloading example as template for offloading ...

---

Missing macros — the following files do offloading but have neither of the two offloading macros

• tests/5.2/declare_target/test_declare_target_enter_device_type_any.c
• tests/5.2/declare_target/test_declare_target_enter.c
• tests/5.2/declare_target/test_declare_target_enter_device_type_host.c
• tests/5.0/target/test_target_uses_allocators_pteam.F90
• tests/5.0/target/test_target_uses_allocators_large_cap.F90
• tests/5.0/target/test_target_uses_allocators_thread.F90
• tests/5.0/target/test_target_uses_allocators_const.F90
• tests/5.0/target/test_target_uses_allocators_high_bw.F90
• tests/5.0/target/test_target_uses_allocators_default.F90
• tests/5.0/target/test_target_uses_allocators_cgroup.F90
• tests/5.0/target/test_target_uses_allocators_low_lat.F90
• tests/5.0/loop/test_loop_bind_device.c
• tests/5.0/loop/test_loop_nested_device.c
• tests/5.0/loop/test_loop_bind_device.F90
• tests/5.0/loop/test_loop_nested_device.F90
• tests/5.0/metadirective/test_metadirective_arch_is_nvidia.F90
• tests/5.0/teams/test_team_default_shared.F90
• tests/5.0/taskloop_simd/test_taskloop_simd_in_reduction_device.c
• tests/5.0/declare_target/test_declare_target_device_type_any.c
• tests/5.0/declare_target/test_nested_declare_target.c
• tests/5.0/declare_target/test_declare_target_device_type_any.F90
• tests/5.0/declare_target/test_declare_target_device_type_host.F90
• tests/5.0/declare_target/test_declare_target_device_type_host.c
• tests/5.1/atomic/test_atomic_compare_device.c
• tests/4.5/offloading_success.cpp
• tests/4.5/target_teams_distribute/test_target_teams_distribute_num_teams.F90
• tests/4.5/offloading_success.F90
• tests/4.5/target_simd/test_target_simd_collapse.c
• tests/4.5/target_update/test_target_update_devices.F90
• tests/4.5/target_teams_distribute_parallel_for/test_target_teams_distribute_parallel_for_defaultmap.F90
• tests/4.5/target_teams_distribute_parallel_for/test_target_teams_distribute_parallel_for_firstprivate.F90
• tests/4.5/target_teams_distribute_parallel_for/test_target_teams_distribute_parallel_for_num_teams.F90
• tests/4.5/target_teams_distribute_parallel_for/test_target_teams_distribute_parallel_for_num_threads.F90
• tests/4.5/offloading_success.c

Unnecessary macros — the following files do not do offloading but have either of the two offloading macros

• tests/5.2/metadirective/test_fortran_pure_nothing.F90
• tests/5.2/implementation_defined/test_ompx_free.F90
• tests/5.0/loop/test_loop_order_concurrent.F90
• tests/5.0/requires/test_requires_atomic_default_mem_order_seq_cst.F90
• tests/5.0/requires/test_requires_atomic_default_mem_order_relaxed.F90
• tests/5.0/requires/test_requires_atomic_default_mem_order_acq_rel.F90
• tests/5.0/parallel_master_taskloop/test_parallel_master_taskloop.F90
• tests/5.0/parallel_master_taskloop/test_parallel_master_taskloop.c
• tests/5.0/teams/test_teams_distribute_default_none.F90
• tests/5.0/teams/test_teams.F90
• tests/5.0/task/test_task_depend_iterator.cpp
• tests/5.0/task/test_task_detach.F90
• tests/5.0/parallel_for/test_parallel_for_allocate.F90
• tests/5.0/parallel_for/test_parallel_for_order_concurrent.F90
• tests/5.0/allocate/test_allocate_allocator.F90
• tests/5.1/cpp_attribute_specifier/test_cpp_attribute_specifier.cpp
• tests/5.1/env_var/test_omp_places_env_numa_domains.c
• tests/5.1/env_var/test_omp_num_teams_env_2.c
• tests/5.1/env_var/test_omp_places_env_ll_caches.c
• tests/5.1/env_var/test_omp_teams_thread_limit_env_2.c
• tests/5.1/depend/test_depend_inoutset.c
• tests/5.1/runtime_calls/test_teams_region_routines.c
• tests/5.1/taskloop/test_taskloop_numtask_strict.c
• tests/5.1/metadirective/test_metadirective_target_device_kind_any.c
• tests/5.1/metadirective/test_metadirective_target_device_kind.c
• tests/5.1/default/test_default_firstprivate_taskloop.c
• tests/5.1/default/test_default_firstprivate_parallel.c
• tests/5.1/default/test_task_default_firstprivate.c
• tests/5.1/flush/test_flush_seq_cst.c
• tests/5.1/taskwait/test_task_nowait.c
• tests/5.1/scope/test_scope_nowait_construct.c
• tests/5.1/scope/test_scope_reduction_construct.c
• tests/5.1/scope/test_scope_private_construct.c
• tests/5.1/masked/test_masked.c
• tests/5.1/masked/test_masked_filter.c
• tests/5.1/error/test_error_at_execution.c
• tests/5.1/error/test_error_message_at_execution.c
• tests/5.1/error/test_error_message_at_compilation.c
• tests/5.1/error/test_error_message.c
• tests/5.1/error/test_error_severity_warning.c
• tests/5.1/allocate/test_omp_alloctrait_key.c
• tests/5.1/allocate/test_calloc_host.c
• tests/5.1/allocate/test_omp_aligned_alloc_host.c
• tests/4.5/application_kernels/qmcpack_target_static_lib.c
• tests/4.5/application_kernels/omp_default_device.c

[spophale]

Thank you @tob2! This is very useful for cleaning up. You are right, Fortran codes were using the macro differently than the C codes and once we changed what it means, there were unintended (hopefully benign) consequences.