This is the mail archive of the mailing list for the GCC project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: GCC 3.3.1 and GCC 3.4 Schedule

> Attached patch updates develop.html accordingly.  OK?

Yes, thanks -- I was about to do that myself!
> >  We'll also need to make some progress on
> >compile-time issues before we branch; we've simply got to get that
> >problem under control.  [Nathan, Zack, and I have a couple of good C++
> >tricks up our sleeves that we'll be implementing.]
> >
> Out of curiosity, are other languages going to benefit from your tricks 
> too?  I was under the impression that much of the slowdown is in the 
> middle-end???

No, the tricks we have in mind are C++-specific.  I think people are
seeing different slowdowns from different places depending on the
workloads; for big C++ template programs we see C++-specific problems,
and those are the ones we're going after.

> >GCC 3.5 is expected to contain the tree-ssa merge.  At the GCC Summit
> >it was clear that many people felt that the tree-ssa work is of vital
> >importance to GCC's future.  I expect to work closely with the
> >tree-ssa people as we get closer to GCC 3.5 to set schedules that work
> >well for that group.
> >
> There was some whispering about maybe a partial merge to reduce the work 
> for Diego somewhat.  Is that still something you're considering, or just 
> a bit no-no?

I missed the whispering, so I have no opinion.

In general, we're going to try to keep non-functional code out of the
mainline; that's why we have branches.  So, I'd like to see a reason --
other than merely making things easier for the branch people -- to do a
partial merge.  If that partial merge cleans up the existing code, or
speeds it up, or fixes some bugs that would be a good reason.

Mark Mitchell
CodeSourcery, LLC

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]