Hi,
there's a bug#682 in icedtea bugzilla, that JRE fails to run on i586 machine with guarantee(VM_Version::supports_cmov()) failed: illegal instruction. After reading the code it seems JIT emits cmov* instructions, so the produced native code cannot run on i586 processors.
I think the most clean way how to fix this in openSUSE is build our openjdk (and Sun Java as well?) as i686 only packages as Fedora and RedHat already does. However it seems like pretty challenging in a BuildService, because the only one package with i686 version I'm aware is glibc, but this approach seems not usable for openjdk, where only i686 and not i586 version is needed.
So how can be only i686 rpm packages produced in a buildservice?
[1] http://icedtea.classpath.org/bugzilla/show_bug.cgi?id=682
Regards Michal Vyskocil
El 29/03/11 10:48, Michal Vyskocil escribió:
Hi,
there's a bug#682 in icedtea bugzilla, that JRE fails to run on i586 machine with guarantee(VM_Version::supports_cmov()) failed: illegal instruction. After reading the code it seems JIT emits cmov* instructions, so the produced native code cannot run on i586 processors.
I think the most clean way how to fix this in openSUSE is build our openjdk (and Sun Java as well?) as i686 only packages as Fedora and RedHat already does. However it seems like pretty challenging in a BuildService, because the only one package with i686 version I'm aware is glibc, but this approach seems not usable for openjdk, where only i686 and not i586 version is needed.
So how can be only i686 rpm packages produced in a buildservice?
[1] http://icedtea.classpath.org/bugzilla/show_bug.cgi?id=682
I tried to figure this out too, with no luck. Will be cool if a option is made available in the web interface to create only i686 packages for a whole project and/or specific packages.
Am Dienstag, 29. März 2011, 10:56:36 schrieb Cristian Rodríguez:
El 29/03/11 10:48, Michal Vyskocil escribió:
Hi,
there's a bug#682 in icedtea bugzilla, that JRE fails to run on i586 machine with guarantee(VM_Version::supports_cmov()) failed: illegal instruction. After reading the code it seems JIT emits cmov* instructions, so the produced native code cannot run on i586 processors.
I think the most clean way how to fix this in openSUSE is build our openjdk (and Sun Java as well?) as i686 only packages as Fedora and RedHat already does. However it seems like pretty challenging in a BuildService, because the only one package with i686 version I'm aware is glibc, but this approach seems not usable for openjdk, where only i686 and not i586 version is needed.
So how can be only i686 rpm packages produced in a buildservice?
Just add
BuildArch: i686
to your spec file ?
[1] http://icedtea.classpath.org/bugzilla/show_bug.cgi?id=682
I tried to figure this out too, with no luck. Will be cool if a option is made available in the web interface to create only i686 packages for a whole project and/or specific packages.
El 29/03/11 12:51, Adrian Schröter escribió:
BuildArch: i686
which Of course, doesnt work
%ifarch %ix86 BuildArch: i686 %endif
Helps but %optflags is still wrong
should be according to the build configs
"-march=i686 -mtune=generic -fomit-frame-pointer -fmessage-length=0"
but it is expanding to :
"-fomit-frame-pointer -fmessage-length=0"
Am Dienstag, 29. März 2011 schrieb Cristian Rodríguez:
El 29/03/11 12:51, Adrian Schröter escribió:
BuildArch: i686
which Of course, doesnt work
%ifarch %ix86 BuildArch: i686 %endif
Helps but %optflags is still wrong
should be according to the build configs
"-march=i686 -mtune=generic -fomit-frame-pointer -fmessage-length=0"
but it is expanding to :
"-fomit-frame-pointer -fmessage-length=0"
Interesting. I wonder how long we compile the glibc.i686 with i586 optflags :(
Greetings, Stephan
On Wed, Mar 30, 2011 at 6:15 AM, Stephan Kulow coolo@suse.de wrote:
Am Dienstag, 29. März 2011 schrieb Cristian Rodríguez:
should be according to the build configs
"-march=i686 -mtune=generic -fomit-frame-pointer -fmessage-length=0"
but it is expanding to :
"-fomit-frame-pointer -fmessage-length=0"
Interesting. I wonder how long we compile the glibc.i686 with i586 optflags :(
Greetings, Stephan
Even more interesting, I wonder how many archs will break when i686 optflags are finally used.
(old archs of course)
On 2011-03-30 11:11:46 -0300, Claudio Freire wrote:
Even more interesting, I wonder how many archs will break when i686 optflags are finally used.
(old archs of course)
do the recent via chips support all i686 stuff?
darix
El 30/03/11 12:21, Marcus Rueckert escribió:
On 2011-03-30 11:11:46 -0300, Claudio Freire wrote:
Even more interesting, I wonder how many archs will break when i686 optflags are finally used.
(old archs of course)
do the recent via chips support all i686 stuff?
Yes, there are some old AMD geode micros that will stop working though.
El 29/03/11 10:48, Michal Vyskocil escribió:
So how can be only i686 rpm packages produced in a buildservice?
I also think we should switch the complete distro to i686 as fedora [1] and Ubuntu [2] already did.
Note that Im not claiming performance gains or magic improvements ;)
[1] http://fedoraproject.org/wiki/Features/F12X86Support [2] http://www.thisweekinlinux.com/2010/06/ubuntu-10-10-maverick-meerkat-drops-p...
El 29/03/11 10:48, Michal Vyskocil escribió:
Hi,
there's a bug#682 in icedtea bugzilla, that JRE fails to run on i586 machine with guarantee(VM_Version::supports_cmov()) failed: illegal instruction. After reading the code it seems JIT emits cmov* instructions, so the produced native code cannot run on i586 processors.
I think the most clean way how to fix this in openSUSE is build our openjdk (and Sun Java as well?) as i686 only packages as Fedora and RedHat already does. However it seems like pretty challenging in a BuildService, because the only one package with i686 version I'm aware is glibc, but this approach seems not usable for openjdk, where only i686 and not i586 version is needed.
So how can be only i686 rpm packages produced in a buildservice?
[1] http://icedtea.classpath.org/bugzilla/show_bug.cgi?id=682
SO, as a workaround, I suggest, adding to the spec file
%ifarch %ix86 BuildArch: i686 %endif
and set CFLAGS/CXXFLAGS to
%ifarch %ix86 export CFLAGS="%optflags -march=i686 -mtune=atom" .... %endif
On Tue, Mar 29, 2011 at 03:48:16PM +0200, Michal Vyskocil wrote:
Hi,
there's a bug#682 in icedtea bugzilla, that JRE fails to run on i586 machine with guarantee(VM_Version::supports_cmov()) failed: illegal instruction. After reading the code it seems JIT emits cmov* instructions, so the produced native code cannot run on i586 processors.
I think the most clean way how to fix this in openSUSE is build our openjdk (and Sun Java as well?) as i686 only packages as Fedora and RedHat already does. However it seems like pretty challenging in a BuildService, because the only one package with i686 version I'm aware is glibc, but this approach seems not usable for openjdk, where only i686 and not i586 version is needed.
So how can be only i686 rpm packages produced in a buildservice?
[1] http://icedtea.classpath.org/bugzilla/show_bug.cgi?id=682
So on the end Oracle devs considered this as a regression, so it has been commited into icedtea-1.10 we will distribute, so this is fixed at least for openjdk
http://mail.openjdk.java.net/pipermail/distro-pkg-dev/2011-March/013375.html
Regards Michal Vyskocil
buildservice@lists.opensuse.org
-
Adrian Schröter -
Claudio Freire -
Cristian Rodríguez -
Marcus Rueckert -
Michal Vyskocil -
Stephan Kulow