REPOST: Announcing the basic-improvements and faster-compiler branches

REPOST: Announcing the basic-improvements and faster-compiler branches

[Zack Weinberg]

I'm reposting this since it got no reaction the first time round,
which was toward the end of the week, so perhaps it dropped off
people's radar.

-----

We have a certain amount of momentum on basic improvements and major
performance work right now; it would be a shame to lose that due to
the code freeze.  I've therefore created two branches in the CVS
repository:

gcc-3_4-basic-improvements-branch: This branch is for basic,
straightforward improvements, which will be easy to merge back into
the mainline at the beginning of development stage 1 for gcc 3.4.  I
do not intend this branch to continue past that point, hence the name.

Examples of work appropriate for this branch:

 - Continued conversion of target macros to target hooks.
 - Other cleanup work on the back-end interface.
 - Continued conversion of front-end interfaces to language hooks.
 - Other cleanup work on the front-end interface.
 - Continued elimination of CPP_PREDEFINES.
 - Continued removal of target dependencies from cpplib.

Checkins to this branch require approval as if they were being
considered for checkin to the mainline during Stage 2 development.

faster-compiler-branch: This one is for major invasive work to make
the compiler go faster.  Anything goes as long as it comes with a
measurable performance improvement, but would not be appropriate to
include in the mainline.  However, PCH is specifically excluded from
this branch, as we already have a branch for that.  This branch is
likely to be longer lived, although I hope that most or all of the
changes will be suitable for inclusion in 3.4.

Examples of work appropriate for this branch:

 - Reducing the size of data structures.
 - Eliminating pointers between tree and RTL.
 - Exploring different memory allocation algorithms.
 - Exploring a "compact" RTL representation.
 - Squelching quadratic or worse algorithmic bottlenecks, such as
   fixup_var_refs.

I'm not sure what checkin policy is appropriate for this branch.
Suggestions?

I will do regular (hopefully weekly) merges from mainline to both
branches.  I will shortly add notes about both to the projects
webpage.

zw

Re: REPOST: Announcing the basic-improvements and faster-compiler branches

[Nathan Sidwell]

Zack Weinberg wrote:

> gcc-3_4-basic-improvements-branch: This branch is for basic,
oh goody. I'll have some of those ...

nathan
-- 
Dr Nathan Sidwell   ::   http://www.codesourcery.com   ::   CodeSourcery LLC
         'But that's a lie.' - 'Yes it is. What's your point?'
nathan@codesourcery.com : http://www.cs.bris.ac.uk/~nathan/ : nathan@acm.org

Re: REPOST: Announcing the basic-improvements and faster-compiler branches

[Kaveh R. Ghazi]

 > From: Zack Weinberg <zack at codesourcery dot com> 
 > 
 > I'm reposting this since it got no reaction the first time round,
 > which was toward the end of the week, so perhaps it dropped off
 > people's radar.

I'm in favor of these.  I've already posted two patches for the
gcc-3_4-basic-improvements-branch:

http://gcc.gnu.org/ml/gcc-patches/2002-08/msg01443.html

--
Kaveh R. Ghazi			Director of Systems Architecture
ghazi@caip.rutgers.edu		Qwest Solutions

Re: REPOST: Announcing the basic-improvements and faster-compilerbranches

[Mark Mitchell]

--On Monday, August 26, 2002 12:40:48 AM -0700 Zack Weinberg 
<zack@codesourcery.com> wrote:

I'm reposting this since it got no reaction the first time round,
which was toward the end of the week, so perhaps it dropped off
people's radar.

Re: REPOST: Announcing the basic-improvements and faster-compiler branches

[Daniel Berlin]

On Monday, August 26, 2002, at 12:57  PM, Mark Mitchell wrote:

Re: REPOST: Announcing the basic-improvements and faster-compiler branches

[Richard Henderson]

On Tue, Aug 27, 2002 at 12:07:29AM -0400, Daniel Berlin wrote:
> Assuming things on the faster-compiler-branch actually make the 
> compiler significantly faster, it should be a review limited to code 
> only, rather than design (IE don't ask the people to make massive 
> changes to patches because the reviewer thinks it would make it 
> "better").

I won't agree to this unconditionally.  What if the patch
is really really gross and seen to be unmaintainable in
the long term?


r~

Re: REPOST: Announcing the basic-improvements and faster-compiler branches

[Daniel Berlin]

On Thursday, August 29, 2002, at 06:41  PM, Richard Henderson wrote:

Re: REPOST: Announcing the basic-improvements and faster-compiler branches

[Mike Stump]

On Monday, August 26, 2002, at 12:40 AM, Zack Weinberg wrote:

Re: REPOST: Announcing the basic-improvements and faster-compiler branches

[Jan Hubicka]

On Mon, Aug 26, 2002 at 11:49:20AM -0700, Mike Stump wrote:
> On Monday, August 26, 2002, at 12:40 AM, Zack Weinberg wrote:
> >gcc-3_4-basic-improvements-branch:
> >faster-compiler-branch:
> 
> Sounds good.  We talked a little bit about the later one around here.

Hi,
the original mail has missed my inbox somehow, but I will check archives as
soon as line will be again up (the flood is not good for connection).
Concerning the second branch, I've recently experimented with bootstrap using
profile feedback.  The results were about 7% time savings on compiling binutils
and more on C++ that is more CPU bound (I will measure again).  THis were
measured on Athlon, the RISCy architectures generally benefits more.

Is that considered usefull?  If so, I can make patch, I guess some distribution
builders may want to use it and we will get some testing for the feature.

Honza