Add omrthread_get_self_thread_time()

[thallium]

This function returns the user and system cpu time of the calling thread.
Related: eclipse-openj9/openj9#20186

[tajila]

@babsingh Please review

[0xdaryl]

Minor nit: please fix the spelling of "thread" in your commit and PR titles. Thanks.

[thallium]

Fixed spelling

[babsingh]

The new function needs to be added similar to the existing functions in thrprof.c: example.

In the new function, the output of omrthread_get_cpu_time and omrthread_get_user_time is aggregated and returned. thrprof.c should have basic functionalities. The aggregation should be done in the upstream project. In other words, the new function shouldn't reside in OMR.

omrthread_get_user_time doesn't have support for Linux. Does anything prevent us to add Linux support in omrthread_get_user_time?

[tajila]

omrthread_get_user_time doesn't have support for Linux. Does anything prevent us to add Linux support in omrthread_get_user_time?

AFAIK, the only way to get user time reliably on linux is with getrusage(RUSAGE_THREAD, ... which can only be called on the current thread. This is why I proposed making a new API since the existing one takes an arbritrary thread as a param.

The goal is to do something like omrthread_get_process_times which returns omrthread_process_time_t

[babsingh]

The new function needs to be added similar to the existing functions in thrprof.c: example.

@thallium port and thread libraries are standalone. a thread library function shouldn't be added to the port library. see an existing function such as thrprof.c:omrthread_get_process_times and follow its implementation. the correct usage of the new function should require #include "omrthread.h" or #include "thread_api.h".

Also, include the justification in the commit on why the new function is needed.

[thallium]

Moved the function to thrprof.c

[babsingh]

@thallium Also, can you add tests for the new function in fvtest/threadextendedtest/processTimeTest.cpp?

[keithc-ca]

Given these existing API functions, what's the real value of adding this function?

If we decide to go ahead with this change, this should use omrthread_get_self_cpu_time() and omrthread_get_self_user_time() instead.

[thallium]

I see what you mean. Yeah, I think the best way is to add Linux implementation to omrthread_get_self_user_time() so we don't need to add a new function. Then we'll be calling omrthread_get_self_cpu_time() and omrthread_get_self_user_time() on the J9 side. @tajila what do you think?

[tajila]

I was hoping to avoid calling getrusage twice, since we need both cpu and user time. What does the support look like on other platforms, are they separate system calls or can those be bundled as well.

[thallium]

Looks like on Windows and AIX they can be bundled as well.

[keithc-ca]

If we're concerned about making two system calls where one would suffice, I suggest this function should be used by omrthread_get_self_cpu_time() and omrthread_get_self_user_time() to reduce code duplication.

[thallium]

This is not a good idea because on some platforms omrthread_get_self_cpu_time get CPU time directly rather than by adding user time and system time.

[thallium]

@babsingh how can I add tests for specific platforms? (since omrthread_get_thread_times will fail on unsupported platforms)

[babsingh]

how can I add tests for specific platforms? (since omrthread_get_thread_times will fail on unsupported platforms)

You can use ifdefs:

#if SUPPORTED_PLATFORMS
   Assert(SUCCESS == rc)
   ...
#else
   Assert(FAILURE == rc)
#endif