This is the mail archive of the gcc@gcc.gnu.org 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 Release Status (2004-05-23)


> > 5. Perhaps LNO can also be merged in.
> 
> This (and tree-profiling) may be a good reason to delay the beginning of 
> Stage 2, but all the other things (cleaning up the internals, the RTL 
> passes, etc.) are perfectly applicable to Stage 2.  I'd expect that, 
> like adding new targets, (very) localized changes to make Ada emit 
> function-at-a-time would be good in Stage 3, or even after GCC 3.5 
> branches, since they would be invisible (Ada does not work so far!).

I would actually like to merge the main expansion changes from
tree-profiling.  I believe they are resonably stable now and works as
expected.
It might be problem to merge the early profiling mainly because loop
optimizer needs to be replaced first so it depends on how things will
get synchronized with LNO merge.  However even without that the new
infrastructure is usefull (it allows more cleanups - such as retiring
line number notes and replacing some of RTL expansion complexity by tree
lowering passes and it is slightly faster too).  It might be possible to
merge the tree profiling and branch estimation code (I think even in
stage 2) and simply keep it off by default until old loop is retired (or
updated to keep CFG in shape - possible backup plan I would preffer to
not realize ;)

I can send some early pathes this week, if it seems appropriate, I
however plan to be offline after the summit till July 14th.

Honza
> 
> Paolo


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