You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
When multiple inferences are performed on openvino compiled based on TBB 2020.3, VmHWM and VmRSS will continue to grow.
But the memory scan tool does not explicitly reveal information:
Asan has no memory leak report
Valgrind has only "possibly lost" and "still reachable" leak reports, and there are stack memory requests associated with pthread_create that are not released. I have also adjusted and tested task_scheduler_init::.blocking_terminate(std::nothrow_t{}) of ExecutorManagerImpl, but it does not work.
And for the (TBB_VERSION_MAJOR > 2020) tbb API, there are no memory issues.
After debugging, and ref [destroy](tbb::task::destroy)'s annotation "Usually, calling this method is unnecessary, because a task is implicitly deleted after its execute() method runs. However, sometimes a task needs to be explicitly deallocated, such as when a root task is used as the parent in spawn_and_wait_for_all."
OpenVINO Version
tag 2023.3.0
Operating System
Ubuntu 18.04 (LTS)
Device used for inference
CPU
Framework
None
Model used
No response
Issue description
When multiple inferences are performed on openvino compiled based on TBB 2020.3, VmHWM and VmRSS will continue to grow.
But the memory scan tool does not explicitly reveal information:
pthread_create
that are not released. I have also adjusted and testedtask_scheduler_init::.blocking_terminate(std::nothrow_t{})
ofExecutorManagerImpl
, but it does not work.And for the
(TBB_VERSION_MAJOR > 2020)
tbb API, there are no memory issues.After debugging, and ref
[destroy](tbb::task::destroy)
's annotation "Usually, calling this method is unnecessary, because a task is implicitly deleted after its execute() method runs. However, sometimes a task needs to be explicitly deallocated, such as when a root task is used as the parent in spawn_and_wait_for_all."the problem should be
tbb::task::allocate_root()
is not released.After the following adjustments, the VmHWM and VmRSS are in stable state
Add the
tbb::task::destroy(root);
afterroot.spawn_and_wait_for_all(a);
.Please help confirm this problem.
Step-by-step reproduction
No response
Relevant log output
No response
Issue submission checklist
The text was updated successfully, but these errors were encountered: