2007-10-26 Joel Sherrill <joel.sherrill@OARcorp.com>

	* score/Makefile.am, score/include/rtems/score/thread.h,
	score/inline/rtems/score/thread.inl: No longer inline _Thread_Get. It
	resulted in unnessary code explosion, many uncovered paths when
	looking at binary executable coverage, and only optimized getting
	self. Id translations were still getting pushed to a subroutine call
	to _Objects_Get. Later the non-inlined version can be further
	optimized to get Ids in range for the current API, then self, then
	look at other APIs.
	* score/src/threadget.c: New file.

1 files changed, 21 insertions, 0 deletions

diff --git a/cpukit/score/include/rtems/score/thread.h b/cpukit/score/include/rtems/score/thread.h
index 1d8ba50705..53d3f54a4a 100644
--- a/cpukit/score/include/rtems/score/thread.h
+++ b/cpukit/score/include/rtems/score/thread.h
@@ -756,6 +756,27 @@ void rtems_iterate_over_all_threads(
   rtems_per_thread_routine routine
 );
 
+/**
+ *  This function maps thread IDs to thread control
+ *  blocks.  If ID corresponds to a local thread, then it
+ *  returns the_thread control pointer which maps to ID
+ *  and location is set to OBJECTS_LOCAL.  If the thread ID is
+ *  global and resides on a remote node, then location is set
+ *  to OBJECTS_REMOTE, and the_thread is undefined.
+ *  Otherwise, location is set to OBJECTS_ERROR and
+ *  the_thread is undefined.
+ *
+ *  @note  The performance of many RTEMS services depends upon
+ *         the quick execution of the "good object" path in this
+ *         routine.  If there is a possibility of saving a few
+ *         cycles off the execution time, this routine is worth
+ *         further optimization attention.
+ */
+Thread_Control *_Thread_Get (
+  Objects_Id         id,
+  Objects_Locations *location
+);
+
 #ifndef __RTEMS_APPLICATION__
 #include <rtems/score/thread.inl>
 #endif