Subject: publishing
From: Craig Chambers (chambers@cs.washington.edu)
Date: Tue Aug 20 2002 - 17:31:58 PDT
OK, famous last words: "I'm sure what I just did won't break
anything." It did. I'm now publishing some fixes to my last publish
adding threads to Whirlwind/Java. I have successfully compiled and
run most of our Java benchmarks w/ optimization (the rest are in
progress), and I have compiled and run several of them on a Windows
machine, too. (Although the qvt library isn't set up yet to support
multiple architectures simultaneously; fixing that's on my to-do
list.)
*) The code I used for creating the initial thread object tickled a
bug in the DFG construction for prim stmt blocks that have
definitions. I fixed it, and added a similar case to the regression
test suite.
*) There seems to be a problem with the conservative GC library not
properly tracing roots in the original stack when compiled for
threads. I implemented a workaround where the first thing the
Whirlwind application does is fork a thread that starts the main
function; the original stack is unused and contains no GC roots.
(Vortex works this way too.)
*) I fixed a small problem with some thread-related global variables
that wasn't defined in non-Java Whirlwind code.
The following are unrelated to thread problems:
*) I fixed a bug I found with recompilation dependency checking for
global variables, for variables that used to be in files that are no
longer included in the program.
*) I fixed a bug I found with references to declarations that are no
longer reachable. Strangely, this bug only showed up when
verification of graph replacements was turned off. I don't yet know
why turning on verification had an effect on the rest of the system;
it shouldn't.
*) I made some changes to the code for reading text files to make
things work on Windows machines where the Cecil file length and
position primitives don't properly correct for CR/LF conversion of
text files. Now there's a function read_whole_text_file that reads
in a whole file into a string, using code that works correctly on
both unix & windows machines. This is both for Vortex & Whirlwind.
*) I simplified the code for verifying and printing slices
recursively.
-- Craig
_______________________________________________
Cecil mailing list
Cecil@cs.washington.edu
http://majordomo.cs.washington.edu/mailman/listinfo/cecil
This archive was generated by hypermail 2b25 : Tue Aug 20 2002 - 17:32:05 PDT