Subject: publishing
From: Craig Chambers (chambers@cs.washington.edu)
Date: Thu Aug 01 2002 - 15:41:52 PDT
I'm publishing some tweaks to Vortex and Whirlwind.
*) I expanded the constant-folding code to work on virtually all of
the integer and boolean operations, and to correctly handle overflow
for the different sizes of operations. I also added folding of the
float comparison operators. (I don't yet support folding of
operators that return floats, or conversion functions.)
*) I fixed a sort-of problem with Cecil's >>_logical (logical right
shift) operator. There's a fundamental problem with this operation,
which is: what's the word size that's assumed? It makes a difference
for negative numbers being shifted, since this defines the transition
between the 1 sign bit and the higher-order 0's. The old version
assumed 31-bit words, which is the number of significant bits
represented in a Cecil tagged integer. But this is different than
the underlying Vortex RTL >>_logical_int operator (and Whirlwind's
analogous >> operator) which operate on normal 32-bit word sizes. In
the constant-folding code for V's >>_logical_int and W's >>, it was
assumed (incorrectly) that Cecil's >>_logical could be used to compute the
result of >>_logical_int/>>. I changed the meaning of Cecil's
>>_logical to match the underlying RTL/Wil operators, and added a
comment to the >>_logical operator.
*) I did some other small clean-ups of the optimization passes.
*) I fixed a problem with lowering representations revealed when
running make-whirlwind-profile. (I'm not sure why this problem
didn't manifest itself in other tests.)
*) I fixed a problem with Vortex asm code compilation revealed by
rerunning the Vortex make-profile regression test script.
*) I fixed an ambiguous method problem revealed not by
implementation-side typechecking (which we don't run) but by
cone-testing optimizations that warn when they detect weird methods.
*) I tweaked the make-whirlwind-profile script.
*) I (re)generated vortex.nCCP and whirlwind.nCCP profiles, and
installed them in the standard place.
*) In building a .prof (gprof-ready) version of Whirlwind, I
discovered that the references to moncontrol weren't found. For now,
I commented them out, until I find out how to reference this function
again.
*) I added support for the solaris8 architecture to the profile
script.
*) I discovered how to access the machine name on solaris8 machines,
allowing lock-run and zap to have a nicer user interface.
*) I have noticed that the process_size Cecil primitive no longer
gives good data on Solaris8. So I disabled it. I've been trying to
find an alternative OS call, but I haven't found something that works
right yet. (getrusage doesn't seem to do the trick, for instance.)
*) I fixed a type error introduced into Vortex by my last publish of
Whirlwind.
*) I removed all the (**debug**) pragmas from Whirlwind.
-- Craig
_______________________________________________
Cecil mailing list
Cecil@cs.washington.edu
http://majordomo.cs.washington.edu/mailman/listinfo/cecil
This archive was generated by hypermail 2b25 : Thu Aug 01 2002 - 15:42:00 PDT