Subject: publishing
From: Craig Chambers (chambers@cs.washington.edu)
Date: Sun Aug 18 2002 - 15:18:48 PDT
I'm publishing some additions to Whirlwind.
*) I implemented lightweight threads for Java in Whirlwind, adapted
from Jonathan's Vortex Java implementation of the same, and reusing
the same qvt library. This involved some minor changes to the Java
front-end, some additions to thread- and monitor-related native
methods in the Java native stdlib code, and some small extensions to
the Wil runtime system makefiles. There's now a single definition in
vortex/runtime/wil/makefiles/MakefileRules that controls whether or
not multiple threads should be used. I've only tested this by
compiling the three multithreaded Java benchmarks w/o optimization
and running them successfully (and verifying that they can still be
compiled with threads turned off), but I can't imagine that any of my
changes will break other benchmarks or optimization. :) I'm running
some overnight tests to confirm this.
Note that one limitation of the current implementation is that I only
poll for thread-switching at function entries. If there's a
long-running loop that contains no function calls (after optimization
& inlining), then it won't check for thread-switching. I should add
a pass somewhere that finds loops that contain no function calls
(after inlining), and inserts thread-switch checks at all back-edges.
*) I fixed a bug, where if you change the representations of some
field of a class, then certain derived representation information
wasn't properly recomputed.
-- Craig
_______________________________________________
Cecil mailing list
Cecil@cs.washington.edu
http://majordomo.cs.washington.edu/mailman/listinfo/cecil
This archive was generated by hypermail 2b25 : Sun Aug 18 2002 - 15:14:48 PDT