Subject: publishing
From: Craig Chambers (chambers@cs.washington.edu)
Date: Mon Jul 16 2001 - 12:59:11 PDT
I'm publishing a bunch of changes I made to Vortex and the surrounding
infrastructure, in order to do the benchmarking for Sorin's POPL
submission. I'm installing a new vortex executable and (shortly) a
new vortex.nCCP profile.
*) I made some small changes to the class-analysis pass so that
inlining of static calls can be done through a separate composed pass
(if modular_inlining is on; not by default).
*) I did a lot of work to break apart the monolithic CSE pass into a
half-dozen separate passes (if modular_cse is on; not by default).
*) If combined_pass is off, then the various passes that would be run
as a single composed pass are instead run in sequence. I added an
option, iterate_main_analyses, which if true (the default) iterates
the non-composed analyses until none of them makes changes.
*) I augmented make-profile to exercise these other configurations.
*) I changed the meaning of the composed framework so that if one pass
snoops on another in the body of the commit closure, it's snooping on
the *new* info, possibly updated if the other pass has already been
committed. It used to snoop on the old info, which might be more
elegant, but doesn't allow any updating of info of the snooped
analysis during the commit closure. This change allowed me to factor
apart value and pointer analysis, while still allowing them to
interact closely to process escaped values. (These passes are rather
intricately interwoven; the earlier monolithic nature of the old CSE
was not without reason.)
*) I fixed a subtle bug in the code generated for field accessor
methods that only shows up in the face of iterated, non-composed copy
propagation.
*) I fixed a problem in the typechecker, where if you (accidentally)
declared a nested method, the typechecker would crash. It now reports
an error message.
*) I updated the benchmark source code (in
~/vortex/Cecil/src/benchmarks), which had gotten out of date with the
runtime system changes over the last few years. The benchmarks can
now all be compiled and run again.
*) I added two new small benchmarks to the .../benchmarks/tiny
directory, merge sort and sumTo. merge sort is adapted from the
.../Cecil/src/tests code. I also did some modification to the msort
benchmark driver to make its output more standard.
*) I extended the ~/vortex/bin/shell/configs.perl file with entries
for several tiny Cecil benchmarks, as well as configurations tested as
part of Sorin's POPL submission.
*) I fixed a bug in the Vortex Java implementation of floating-point % reported
by Tony Hosking's group.
-- Craig
_______________________________________________
Cecil mailing list
Cecil@cs.washington.edu
http://majordomo.cs.washington.edu/mailman/listinfo/cecil
This archive was generated by hypermail 2b25 : Mon Jul 16 2001 - 13:00:05 PDT