Skip site navigation (1)Skip section navigation (2)
Date:      Thu, 20 Nov 2003 15:11:08 -0500
From:      mi+mx@aldan.algebra.com
To:        Joe Marcus Clarke <marcus@marcuscom.com>
Cc:        FreeBSD GNOME Users <gnome@freebsd.org>
Subject:   Re: Mozilla's xptcall
Message-ID:  <200311201511.08686@misha-mx.virtual-estates.net>
In-Reply-To: <1069358484.783.51.camel@gyros>
References:  <200311201454.56022@misha-mx.virtual-estates.net> <1069358484.783.51.camel@gyros>

next in thread | previous in thread | raw e-mail | index | archive | help
On Thu, 2003-11-20 at 14:54, mi+mx@aldan.algebra.com wrote:
=> Hello!
=> 
=> In order to build mozilla (and hence, OpenOffice) I needed the
=> attached patch. It seems, that the xptcall files are very fragile
=> to the compiler optimization, so the patch is needed to replace
=> whatever the configured through the make.conf with the vanilla
=> ``-O''. Curiously, the file being patched already contains similar
=> hacks for other platforms. Yours,

=mozilla and mozilla-devel build fine with -O2.

regxpcom usually crashes for me unless built with -O. But I'm using
-march=pentium4 and -fomit-frame-pointer.

This time I went with a very elaborate:

	-march=pentium4 -O \
	-fomit-frame-pointer -fbuiltin -fstrength-reduce \
	-fthread-jumps -funroll-loops -fcse-follow-jumps \
	-fcse-skip-blocks -frerun-cse-after-loop \
	-fexpensive-optimizations  -fschedule-insns -fschedule-insns2

Most (all?) of the rest of the code is either C or C++ and it is reasonable
to expect the compiler to do the right thing. The directory in question,
however, contains mostly assembly code, which is being mis-compiled with
more agressive optimization options.

	-mi




Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?200311201511.08686>