Shutdown library on Windows if possible for better OMPT behavior.

[hbae]

Summary

On Windows, child workers are terminated by the parent during the normal
program exit process (ExitProcess()) and they are not able to finish generating
their OpenMP events. We can force manual library shut down in __kmpc_end() to
fix this at least for the cases where __kmpc_end() is properly inserted.

[AndreyChurbanov]

LGTM

openmp/trunk/runtime/src/kmp_csupport.cpp

void __kmpc_end(ident_t *loc) {
   // returns FALSE and __kmpc_end() will unregister this root (it can cause
   // library shut down).
   if (__kmp_ignore_mppend() == FALSE) {
     KC_TRACE(10, ("__kmpc_end: called\n"));
     KA_TRACE(30, ("__kmpc_end\n"));
 
     __kmp_internal_end_thread(-1);
   }
+#if KMP_OS_WINDOWS && OMPT_SUPPORT
+  // Normal exit process on Windows does not allow worker threads of the final
+  // parallel region to finish reporting their events, so shutting down the
+  // library here fixes the issue at least for the cases where __kmpc_end() is
+  // placed properly.
+  if (ompt_enabled.enabled)
+    __kmp_internal_end_library(__kmp_gtid_get_specific());
+#endif
 }
 
 /*!
 @ingroup THREAD_STATES
 @param loc Source location information.
 @return The global thread index of the active thread.
 
 This function can be called in any context.