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
Implement the 3.12+ API using sys.monitoring() to call the PerfStubs API
Implement the <3.12 API using sys.setprofile() to call the PerfStubsAPI
Replace this with PyEval_SetProfile in C for potentially lower overhead?
Optionally filter the Python internal function calls (any timers in files that are or in the python installation)
Optionally delay tracing timers until the _run_module_code timer is encountered. Use a bootstrap start/stop function until that point. Possibly need a custom stop timer event, so that we can shut down at that point. Otherwise, TAU could get an extra stop event for any events on the stack after _run_module_code exits.
Add selective measurement support using the same file format TAU expects
Add and test special support for Spark
Add and test special support for TensorFlow
Add and test special support for PyTorch (if any)
Confirm no support from TAU implementation has been lost (metadata support? error handling? backtrace collection at crash?)
Test with both TAU and APEX
Test with multiple threads - anything else needed? register thread?
Using this documentation: https://docs.python.org/3/library/sys.monitoring.html update the support in TAU for profiling Python code. Additional info: https://blog.jetbrains.com/pycharm/2024/01/new-low-impact-monitoring-api-in-python-3-12/.
Subtasks for this work:
_run_module_code
timer is encountered. Use a bootstrap start/stop function until that point. Possibly need a custom stop timer event, so that we can shut down at that point. Otherwise, TAU could get an extra stop event for any events on the stack after_run_module_code
exits.Completion of this issue will contribute to the completion of issue UO-OACISS/tau2#16 which contributes to project milestone https://github.com/orgs/UO-OACISS/projects/1.
The text was updated successfully, but these errors were encountered: