> From: Bill Pugh [mailto:pugh@cs.umd.edu]
> Sent: Tuesday, July 10, 2001 7:19 PM
> What we need is the ability to contain synchronization errors, so
> that a synchronization error in one part of the program doesn't
> destroy the correctness of synchronization elsewhere in the program.
You may want to see
http://www.cs.uiuc.edu/~sadve/Publications/isca91.dataraces.ps.
This paper from ISCA'91 describes a sequentially consistent semantics for
all parts of a program that are not "affected" by a data race. The "affects"
relation is in terms of happens-before. The paper gives hardware constraints
to ensure the above as well.
Sarita
-------------------------------
JavaMemoryModel mailing list - http://www.cs.umd.edu/~pugh/java/memoryModel
This archive was generated by hypermail 2b29 : Thu Oct 13 2005 - 07:00:33 EDT