starpu-max/tools/valgrind/starpu.suppr

{
   don't care about cache hit stats
   Helgrind:Race
   fun:_starpu_msi_cache_hit
   ...
}

{
   don't care about cache miss stats
   Helgrind:Race
   fun:_starpu_msi_cache_miss
   ...
}

{
   known race, but not problematic in practice, see comment in _starpu_tag_clear
   Helgrind:LockOrder
   ...
   fun:_starpu_tag_free
   fun:_starpu_htbl_clear_tags
   ...
   fun:_starpu_tag_clear
   fun:starpu_shutdown
   ...
}


{
   There is actually no race on current_mode, because the mode can not change unexpectedly, until _starpu_notify_data_dependencies() is called further down. Valgrind can not know about such software rwlock.
   Helgrind:Race
   fun:_starpu_release_data_on_node
   fun:_starpu_push_task_output
   ...
}

{
   We do not care about races on profiling statistics
   Helgrind:Race
   fun:_starpu_worker_get_status
   fun:_starpu_worker_reset_profiling_info_with_lock
   ...
}

{
   This is racy, but since we'll always put the same values, this is not a problem.
   Helgrind:Race
   fun:_starpu_codelet_check_deprecated_fields
   ...
}

{
   This is racy, but we don't care, it's only a statistic
   Helgrind:Race
   fun:starpu_task_nsubmitted
   ...
}

{
   This is racy, but we don't care, it's only a statistic
   Helgrind:Race
   fun:starpu_task_nready
   ...
}

{
   This is racy, but we don't care, it's only a statistic
   Helgrind:Race
   fun:_starpu_bus_update_profiling_info
   ...
}

{
   fscanf error
   Memcheck:Cond
   ...
   fun:fscanf
   fun:_starpu_load_bus_performance_files
   ...
}

{
   mc / handle locking order. We always take handle before mc, except when we tidy memory, where we use trylock, but helgrind doesn't handle that :/ https://bugs.kde.org/show_bug.cgi?id=243232
   Helgrind:LockOrder
   fun:mutex_trylock_WRK
   fun:__starpu_spin_trylock
   ...
}

{
   mc / handle locking order1
   Helgrind:LockOrder
   ...
   fun:__starpu_spin_lock
   fun:try_to_free_mem_chunk
   ...
}

{
   mc / handle locking order2
   Helgrind:LockOrder
   ...
   fun:__starpu_spin_lock
   fun:try_to_find_reusable_mem_chunk
   ...
}

{
   mc / handle locking order3
   Helgrind:LockOrder
   ...
   fun:__starpu_spin_lock
   fun:free_potentially_in_use_mc
   ...
}

{
   mc / handle locking order4
   Helgrind:LockOrder
   ...
   fun:__starpu_spin_lock
   fun:free_potentially_in_use_mc
   ...
}

{
   mc / handle locking order5
   Helgrind:LockOrder
   ...
   fun:__starpu_spin_lock
   fun:register_mem_chunk
   ...
}

{
   mc / handle locking order6
   Helgrind:LockOrder
   ...
   fun:__starpu_spin_lock
   fun:_starpu_request_mem_chunk_removal
   ...
}

{
   mc / handle locking order7
   Helgrind:LockOrder
   ...
   fun:__starpu_spin_lock
   fun:_starpu_allocate_interface
   ...
}

{
   mc / handle locking order8
   Helgrind:LockOrder
   ...
   fun:__starpu_spin_lock
   fun:_starpu_memchunk_recently_used
   ...
}

{
   TODO1: This is temporary. It perhaps does not pose problem because only the worker takes this mutex. Fixing this will require changing the scheduler interface, so that the schedulers themselves take the scheduling lock, not the caller. Filter it out for now, so we can see other races more easily.
   Helgrind:LockOrder
   fun:pthread_mutex_lock
   fun:starpu_pthread_mutex_lock
   fun:simple_worker_pull_task
   ...
}

{
   TODO1: This is temporary. It perhaps does not pose problem because only the worker takes this mutex. Fixing this will require changing the scheduler interface, so that the schedulers themselves take the scheduling lock, not the caller. Filter it out for now, so we can see other races more easily.
   Helgrind:LockOrder
   fun:pthread_mutex_lock
   fun:starpu_pthread_mutex_lock
   fun:_starpu_sched_component_worker_lock_scheduling
   fun:simple_worker_pull_task
   ...
}

{
   TODO2: This is temporary. It perhaps does not pose problem because only the worker takes this mutex. Fixing this will require changing the scheduler interface, so that the schedulers themselves take the scheduling lock, not the caller. Filter it out for now, so we can see other races more easily.
   Helgrind:LockOrder
   fun:pthread_mutex_lock
   fun:starpu_pthread_mutex_lock
   fun:_starpu_sched_component_lock_worker
   fun:simple_worker_pull_task
   ...
}

{
   TODO2: This is temporary. It perhaps does not pose problem because only the worker takes this mutex. Fixing this will require changing the scheduler interface, so that the schedulers themselves take the scheduling lock, not the caller. Filter it out for now, so we can see other races more easily.
   Helgrind:LockOrder
   fun:pthread_mutex_lock
   fun:starpu_pthread_mutex_lock
   fun:_starpu_sched_component_lock_worker
   fun:_starpu_sched_component_worker_lock_scheduling
   fun:simple_worker_pull_task
   ...
}