[meego-commits] GCC 4.5 (was: 5539: Changes to Trunk:Testing/gcc)

[Zhang, Austin]

> No matter where the final fix will be, my point is, we application
> developers need a stable base to work on. Keeping up with newest tools
If you have data/evidence on 4.5.0 is not stable, I'd like to learning. If you are meaning the regression is saying 4.5 is not stable to use, I am not agree on this; We also have some data/benchmarks which indicate some other app will gain in performance. 

> is good, we also want to do that (Mozilla started testing GCC 4.5 early
> even before it's released), but a formal transition will need more time
We don't do it by a sudden thoughts, the 1st evaluation happened at April. 

> and efforts, and the transition is the decision of application
> developers, thus it will not happen at the same time as GCC 4.5 is
> available in distribution.
Yes, we also hope to find those data from your or other applications, in fact, we don't update it only by "it is newer". And we also get data from benchmarks which is indicating some other cases is gaining performance from the new compiler.

> What's more regarding the performance: to do this we need side-by-side
> comparisons between GCC 4.4 vs 4.5 _in the same environment_. That means
> all packages are the same except GCC. So it's necessary for distribution
> to provide GCC 4.4 and 4.5 in the repo at the same time.
4.4.2 can be got from current Trunk and 1.0, 4.5.0 is only existing in Trunk:Testing now.
Trunk and Trunk:Testing have almost all same pkgs. If you want only gcc diff, using Trunk to create image with 4.4, and then install 4.5.
If you'd like to consider some others pkgs possible impact, create image from Trunk and Trunk:Testing respectively. 

> PS: and the 2 versions should be able to install at the same time in a MeeGo system.
Not support so far due to some files conflication. But might be workarounded by rpm installation. 

-----Original Message-----
From: Roger WANG [mailto:roger.wang at linux.intel.com] 
Sent: Saturday, July 10, 2010 6:18 PM
To: Zhang, Austin
Cc: meego-commits at meego.com
Subject: Re: GCC 4.5 (was: [meego-commits] 5539: Changes to Trunk:Testing/gcc)

"Zhang, Austin" <austin.zhang at intel.com> writes:

> Thanks for pointing this out.
>
> For firefox/fennec compiling, according comments in this thread (Quote
> from Taras Glek: "I tried 4.5 -O2 and it's actually faster than 4.3
> -Os.") with 4.5, can we have some changes on their default flags, say
> using -O2 instead of -Os (also check how much the size will increase),
> then see what's the performance regression?

Yeah. Mozilla starts looking after that. 

No matter where the final fix will be, my point is, we application
developers need a stable base to work on. Keeping up with newest tools
is good, we also want to do that (Mozilla started testing GCC 4.5 early
even before it's released), but a formal transition will need more time
and efforts, and the transition is the decision of application
developers, thus it will not happen at the same time as GCC 4.5 is
available in distribution.

What's more regarding the performance: to do this we need side-by-side
comparisons between GCC 4.4 vs 4.5 _in the same environment_. That means
all packages are the same except GCC. So it's necessary for distribution
to provide GCC 4.4 and 4.5 in the repo at the same time.

Thanks.

-- 
Roger WANG                         Intel Open Source Technology Center