Subject: publishing
From: Craig Chambers (chambers@cs.washington.edu)
Date: Thu Sep 06 2001 - 18:13:23 PDT
I'm publishing some small extensions to Whirlwind, to fix up some
problems with initial and once declarations before launching into
bigger projects: instanceof testing and GF dispatching. The test
suite program compiles and runs, and even the Java towers benchmark
now compiles & links successfully, without optimization (it doesn't
run successfully because it uses GFs & dispatching, and it doesn't
compile successfully w/ optimization because of (at least) problems
building a proper DFG in all cases and with generating C code with phi
nodes).
*) For initial declarations at top-level, all the initial declarations
of all the files need to be done before any of the regular
declarations of any files are done. But the old way of compiling
initial declarations just did initial declarations before regular
declarations within each file. I changed things to generate two
separate top-level functions in each file, one just for initial
declarations and one just for regular declarations. The application's
main function now calls all the initial top-level functions of all
files before calling the regular top-level functions. In the process,
I created some new WindIRs to hold only CFGs and later slices, and got
that approach to work out.
*) Once declarations used to be compiled just like regular initially
declarations; in particular, duplicate once declarations led to
duplicate declaration link errors. I fixed this, adding some code to
identify the representative declaration out of a set of declarations
of the same name, and treating only the representative as the one
getting the declaration; all others act like extern declarations.
*) I fixed some dependency problems revealed by the above changes.
*) I extended the test suite with some code to test initial and once
declarations.
*) I changed how the Java native methods refer to C stdlib functions
to not use the var decl "source" feature of Wil but instead to use C
prim stmt blocks. I also wrote implementations of a couple of runtime
functions that were being referenced from our Java native code, but
not yet implemented.
*) I fixed some problems relating to building the CFG & DFG for C prim
stmt blocks.
*) Amazingly, there was a problem with the implementation of basic
string printing, where it called fwrite to print the string to stdout,
and fwrite didn't always print out the whole string (in particular, it
stopped after newlines often).
-- Craig
_______________________________________________
Cecil mailing list
Cecil@cs.washington.edu
http://majordomo.cs.washington.edu/mailman/listinfo/cecil
This archive was generated by hypermail 2b25 : Thu Sep 06 2001 - 18:14:05 PDT