gcc news #9

[Mathieu Lacage <Mathieu dot Lacage at sophia dot inria dot fr>]

as usual, http://gccnews.chatta.us/

The full text:
--------------

gcc release

The presentation of the release objectives and tentative schedule for
gcc 3.5 by Mark Mitchel last week brought in quite a bit of discussion
and disagreement. It looks like some developers would like to keep on
working on gcc 3.5 much longer until performance (compilation-time as
well as generated code) has improved in a few areas. Mark and a few
others advocate that the current list of new features and improvements
is good enough to warrant a release relatively soon. Of course, future
releases should keep on improving on this release.

Because there obviously needs a way to make a decision about when the
release is good-enough, Mark sent a lengthy mail requesting all
developers to provide accurate descriptions of things they would like to
see go in before a 3.5 release. Hopefully, this should help foster wide
consensus about the release and will make 3.5 better than ever.


gcc development

Scott Robert Ladd is back with a few new performance benchmark results
which show that, although some areas need improvement, gcc 3.5 often
indeed generates better code than previous gcc releases.

Steven Bosscher reported some issues with the performance of a fortran
benchmark: it looks like the pre pass is overoptimizing the code and
thus using too many registers which has the rather unwelcome side-effect
to make the job of the register allocator pass from very hard to
impossible and thus to generate very bad (from the point of view of
performance) code. Kenneth Zadeck gave a thorough explanation of this
phenomenon and discussed some of its potential solutions.

C language lawyers will love this thread started by Joseph S. Myers 

-- 
Mathieu Lacage <mathieu.lacage@sophia.inria.fr>