[OpenMP] Fix a segment fault in __kmp_get_global_thread_id

[tianshilei1992]

Summary

In __kmp_get_global_thread_id, if the gtid mode is 1, after getting the gtid
from TLS, it will store the gtid value to the thread stack maintained in the thread
descriptor. However, __kmp_get_global_thread_id can be called when the library
is destructed, after the corresponding thread info has been release. This will
cause a segment fault. This can happen on an Intel-based Mac.

[tianshilei1992]

improve comment

[jdoerfert]

can we have a test?

[tianshilei1992]

In D159324#4633034, @jdoerfert wrote:

can we have a test?

It can be exposed by existing test on macOS. That's how I found the issue. This *might* cause the issue reported in https://github.com/llvm/llvm-project/issues/63194.

openmp/runtime/src/kmp_runtime.cpp

#endif
   i = __kmp_gtid_get_specific();
 
   /*fprintf( stderr, "=== %d\n", i );  */ /* GROO */
 
   /* if we havn't been assigned a gtid, then return code */
   if (i < 0)
     return i;
 
+  // other_threads[i] can be nullptr at this point because the corresponding
+  // thread could have already been destructed. It can happen when this function
+  // is called in end library routine.
+  if (!TCR_SYNC_PTR(other_threads[i]))
+    return i;
+
   /* dynamically updated stack window for uber threads to avoid get_specific
      call */
   if (!TCR_4(other_threads[i]->th.th_info.ds.ds_stackgrow)) {
     KMP_FATAL(StackOverflow, i);
   }
 
   stack_base = (char *)other_threads[i]->th.th_info.ds.ds_stackbase;
   if (stack_addr > stack_base) {