From: Matthew H. <he...@ca...> - 2007-04-25 15:36:34
|
The way that Merlin is implemented in the Jikes RVM, it performs a complete root scan at each accurate point in the trace. This root scan includes both a scan for references in the threads stacks and a scan for all of the references in the JTOC (static region). While there are hooks in place for the implementation to take advantage of a write barrier on the static fields (and therefore omit the scan of statics at each accurate point), it is not relying on these at the moment. In my initial implementation of Merlin (in Jikes 2.0.3 using GCTk), I also created a stack barrier that performed a limited stack analysis at the method call granularity. I have not brought this forward, however, because of the complexity of implementing this code (and sheer number of classes and subsystems it impacts). Thanks! Sincerely, Matthew Hertz Li Gen wrote: > Professor Eliot Moss, > Thank you for your answer. > The academic of Merlin is prefect, and, of course, > has considered references from stack (and registers) and from statics. > But my puzzle is about the implementation of Merlin in RVM. > In BaseBaseGCTrace of RVM 2.9.0, the implementation of static field > write barrier is empty, and the baseline compiler does not implement the > stack writer barrier in emit_astore function, because the astore byte > code is the only instruction which can save object references to the > stack. So I mean, Merlin in RVM 2.9.0 can't know when the object > referenced by stack is updated, Am I right? And is there other reason > for the unimplemenation of the stack write barrier? > Best wishes > > > > > ------------------------------------------------------------------------ > > ------------------------------------------------------------------------- > This SF.net email is sponsored by DB2 Express > Download DB2 Express C - the FREE version of DB2 express and take > control of your XML. No limits. Just data. Click to get it now. > http://sourceforge.net/powerbar/db2/ > > > ------------------------------------------------------------------------ > > _______________________________________________ > Jikesrvm-researchers mailing list > Jik...@li... > https://lists.sourceforge.net/lists/listinfo/jikesrvm-researchers |