Skip site navigation (1)Skip section navigation (2)
Date:      Wed, 27 Jun 2007 14:47:34 +0000
From:      Andrey Chernov <ache@freebsd.org>
To:        Xin LI <delphij@delphij.net>
Cc:        Kostik Belousov <kostikbel@gmail.com>, cvs-src@FreeBSD.org, src-committers@FreeBSD.org, Harti Brandt <harti@FreeBSD.org>, cvs-all@FreeBSD.org
Subject:   Re: cvs commit: src/share/mk sys.mk
Message-ID:  <20070627144734.GA49174@freebsd.org>
In-Reply-To: <46826A8C.8070908@delphij.net>
References:  <200706261910.l5QJALb8093717@repoman.freebsd.org> <20070627093610.GU2268@deviant.kiev.zoral.com.ua> <20070627113859.N64822@knop-beagle.kn.op.dlr.de> <20070627101148.GB44554@nagual.pp.ru> <46826A8C.8070908@delphij.net>

next in thread | previous in thread | raw e-mail | index | archive | help
On Wed, Jun 27, 2007 at 09:47:56PM +0800, Xin LI wrote:
> Nitpicking: I think -O1 implies no strict-aliasing.  So -O1 -pipe might
> be just Ok.

It is for easy change-back.

> Well, I'd say that all these changes looks scary to me.
> 
> Is there any code in our base system to trigger tree-vrp bug?  Do we
> still have some time to have gcc fixed and tested rather than using
> band-aid like this?  IMHO fixing gcc sounds better than "fix"ing sys.mk
> if time permits us to fix and test a vendor solution.

It is hard to find such cases because of the silent nature of the bug:
basically any array usage inside loop, with array size smaller than 
the loop iterations can trigger that (premature exit from the loop).
I think this case is general enough to worry at the system level.

There is no vendor solution right now and will not be until gcc 4.2.1
I doubt it will be available as separate patch because of complexity of tree 
vrp optimization.

This change will be backed out right after fixing gcc in any way.

-- 
http://ache.pp.ru/



Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20070627144734.GA49174>