7:12 a.m.
Updated Packages:
GConf2-2.10.0-4
---------------
* Mon May 09 2005 Mark McLoughlin <markmc(a)redhat.com> 2.10.0-4
- Update to upstream evoldap.schema which uses GNOME's OID base
rather than Red Hat's OID.
GFS-6.1-0.pre22.6
-----------------
GFS-kernel-2.6.11.5-20050505.133825.FC4.6
-----------------------------------------
audit-0.7.4-1
-------------
* Sun May 08 2005 Steve Grubb <sgrubb(a)redhat.com> 0.7.4-1
- Make sure ausearch ts & te obey DST.
- Code cleanups to make file system watches work correctly
ccs-0.25-0.17
-------------
* Sun May 08 2005 Florian La Roche <laroche(a)redhat.com>
- mv /etc/init.d -> /etc/rc.d/init.d
- fix Requires
* Fri May 06 2005 Chris Feist <cfeist(a)redhat.com>
- Cleanup .spec file to not include globs.
chkconfig-1.3.20-1
------------------
* Thu May 05 2005 Bill Nottingham <notting(a)redhat.com> 1.3.20-1
- fix deletion of orphaned slave links (#131496, <mitr(a)redhat.com>)
cman-1.0-0.pre33.10
-------------------
* Thu May 05 2005 Chris Feist <cfeist(a)redhat.com>
- Added patch to disable starting up the init scripts.
* Mon Dec 20 2004 Chris Feist <cfeist(a)redhat.com>
- Rebuild with new sources.
* Mon Dec 13 2004 Chris Feist <cfeist(a)redhat.com>
- Rebuild with new sources.
cman-kernel-2.6.11.3-20050425.154843.FC4.14
-------------------------------------------
control-center-1:2.10.1-5
-------------------------
* Thu May 05 2005 Ray Strode <rstrode(a)redhat.com> - 1:2.10.1-5
- Don't pop up accessibility dialogs under currently focused
window
device-mapper-multipath-0.4.4-2.3
---------------------------------
* Fri May 06 2005 Bill Nottingham <notting(a)redhat.com> - 0.4.4-2.3
- Fix last fix.
* Thu May 05 2005 Alasdair Kergon <agk(a)redhat.com> - 0.4.4-2.2
- Fix last fix.
dlm-1.0-0.pre21.8
-----------------
* Sun May 08 2005 Florian La Roche <laroche(a)redhat.com>
- fix -devel requires
dlm-kernel-2.6.11.3-20050425.154843.FC4.16
------------------------------------------
elfutils-0.107-2
----------------
* Sun May 08 2005 Roland McGrath <roland(a)redhat.com> - 0.107-2
- fix strip -f byte-swapping bug
* Sun May 08 2005 Roland McGrath <roland(a)redhat.com> - 0.107-1
- update to 0.107
- readelf: improve DWARF output format
- elflint: -d option to support checking separate debuginfo files
- strip: fix ET_REL debuginfo files (#156341)
evolution-2.2.2-5
-----------------
* Thu May 05 2005 David Malcolm <dmalcolm(a)redhat.com> - 2.2.2-5
- added evolution-2.2.2-fix-new-mail-notify.patch to CVS
* Thu May 05 2005 David Malcolm <dmalcolm(a)redhat.com> - 2.2.2-4
- Removed explicit mozilla_build_version; instead use pkg-config to determine
the path to the NSS/NSPR headers.
- Use a macro to express requirement on pilot-link (was 1:0.11.4, now 0.12;
patches depend on this)
- Re-enabled the new-mail-notify plugin (my patch to handle differing DBus
versions is in the upstream tarball; but configure.in disables the plugin for
dbus versions > 0.23; patched configure.in to allow arbitrary DBus versions,
and run autoconf at the start of the build) (#156328)
* Sat Apr 30 2005 David Malcolm <dmalcolm(a)redhat.com> - 2.2.2-3
- updated mozilla_build_version to 1.7.7
fence-1.27-16
-------------
festival-1.95-3
---------------
* Thu Apr 28 2005 <johnp(a)redhat.com> - 1.95-3
- require info packages so the post does not fail
- remove /usr/bin/VCLocalRule from buildroot since it is
an extranious file that does not need to be installed
* Wed Apr 27 2005 Miloslav Trmac <mitr(a)redhat.com> - 1.95-2
- Fix build with gcc 4 (#156132)
- Require /sbin/install-info for scriptlets (#155698)
- Don't ship /usr/bin/VCLocalRules (#75645)
firefox-0:1.0.3-5
-----------------
* Mon May 09 2005 Christopher Aillon <caillon(a)redhat.com> 0:1.0.3-5
- Correctly position the IM candidate window for most locales
Note: it is still incorrectly positioned for zh_TW after this fix
- Add temporary workaround to not create files in the user's $HOME (#149664)
gcc-4.0.0-4
-----------
* Thu May 05 2005 Jakub Jelinek <jakub(a)redhat.com> 4.0.0-4
- update from CVS
- PRs libgcj/21372, target/21284
- fix fold ICE (James A. Morrison, PR tree-optimization/21085)
- fix weak decl merging (PR c++/20961)
- optimize i386/x86-64 movmem sequences (PR target/21329)
* Wed May 04 2005 Jakub Jelinek <jakub(a)redhat.com> 4.0.0-3
- update from CVS
- PRs middle-end/19985, bootstrap/20633, c++/15875, c++/19542, c/15698,
fortran/16861, java/21022, libfortran/17992, libfortran/19568,
libfortran/19595, libfortran/20005, libfortran/20092,
libfortran/20131, libfortran/20661, libfortran/20744, libgcj/21136,
libstdc++/21209, libstdc++/21286, rtl-optimization/21144,
target/16888, target/21098, treelang/21345
- fix loop unswitching (PR rtl-optimization/21330)
- fix i386 ASHIFT to MULT address canonicalization (PR target/21297)
- propagate tail call info bit through builtin expanders (PR middle-end/21265)
- fix VEC_SELECT in combine (PR rtl-optimization/21239)
- fix i386 V4SFmode vector initialization
- fix Fortran FOREACH (PR fortran/15080)
gimp-2:2.2.7-1
--------------
* Mon May 09 2005 Nils Philippsen <nphilipp(a)redhat.com>
- version 2.2.7, fixes bug in SSE2 assembly for Lighten Only layer mode
(#145771) and various other bugs
- on x86 and x86_64, use -msse and -msse2 to accomodate newer compilers
* Wed Apr 27 2005 Jeremy Katz <katzj(a)redhat.com> - 2:2.2.6-2
- silence %post
* Mon Apr 11 2005 Nils Philippsen <nphilipp(a)redhat.com>
- version 2.2.6
gimp-print-4.2.7-7
------------------
* Thu May 05 2005 Bill Nottingham <notting(a)redhat.com> 4.2.7-7
- rebuild
gnbd-1.0-0.pre14.4
------------------
gnbd-kernel-2.6.11.2-20050420.133124.FC4.19
-------------------------------------------
gnome-desktop-2.10.0-3
----------------------
* Thu May 05 2005 Ray Strode <rstrode(a)redhat.com> - 2.10.0-3
- Revert previous patch--it was wrong(bug 155659).
* Wed May 04 2005 Ray Strode <rstrode(a)redhat.com> - 2.10.0-2
- run gettext initialization routines on startup (bug 155659).
gnupg-1.4.1-3
-------------
* Thu May 05 2005 Nalin Dahyabhai <nalin(a)redhat.com> 1.4.1-3
- fix the execstack problem correctly this time (arjanv)
gstreamer-plugins-0.8.8-6
-------------------------
* Mon May 09 2005 John (J5) Palmieri <johnp(a)redhat.com> - 0.8.8-6
- Added patch to fix segfault (Bug #156622)
gtk2-2.6.7-3
------------
* Sun May 08 2005 Matthias Clasen <mclasen(a)redhat.com>
- remove debug spew
gulm-1.0-0.pre29.2
------------------
* Sun May 08 2005 Florian La Roche <laroche(a)redhat.com>
- mv /etc/init.d -> /etc/rc.d/init.d
- fix requires:
* Fri May 06 2005 Chris Feist <cfeist(a)redhat.com>
- fixed various problems with .spec file
httpd-2.0.54-8
--------------
* Mon May 09 2005 Joe Orton <jorton(a)redhat.com> 2.0.54-8
- drop old "powered by Red Hat" logos
iddev-1.9-21
------------
iiimf-1:12.2-3
--------------
* Mon May 09 2005 Akira TAGOH <tagoh(a)redhat.com> - 1:12.2-3
- Iiimx.Xresource: added a X resource file for iiimx, including
the conversionOnKeys resource to allow shift+space. (#138309)
* Mon May 09 2005 Jens Petersen <petersen(a)redhat.com> - 1:12.2-2
- add gimlet-desktop-dir-149226.patch to install desktop file in the right
place (Mark McLoughlin, 149226)
* Tue May 03 2005 Jens Petersen <petersen(a)redhat.com> - 1:12.2-1
- update to final upstream 12.2 release
- the following patches are now upstream:
iiimgcf-multilib-r2584-155960.patch,
iiimsf-shift-space-for-korean-r2592-156171.patch,
htt_xbe-correct-error-message-r2594-156169.patch,
xiiimp-multilib-r2585-155963.patch
- no longer need to rebootstrap iiimgcf
kernel-2.6.11-1.1288_FC4
------------------------
* Fri May 06 2005 Dave Jones <davej(a)redhat.com>
- Add PCI ID for new sundance driver. (#156859)
libexif-0.6.12-3
----------------
* Fri May 06 2005 Matthias Clasen <mclasen(a)redhat.com>
- Prevent infinite recursion (#156365)
mkinitrd-4.2.12-1
-----------------
* Wed May 04 2005 Peter Jones <pjones(a)redhat.com> - 4.2.12-1
- Don't copy "console=" arguments from /proc/cmdline to init
netpbm-10.27-2
--------------
* Mon May 09 2005 Jindrich Novy <jnovy(a)redhat.com> 10.27-2
- fix invalid strcmp condition in bmptopnm, typo in pnmtojpeg
(David Constanzo, #157106, #157118)
- proper read longs and shorts in libpm.c (David Constanzo, #157110)
- fix segfault in bmptopnm caused by freeing an uninitialized pointer
parted-1.6.22-2
---------------
* Thu May 05 2005 Chris Lumens <clumens(a)redhat.com> 1.6.22-2
- Added upstream patch to display certain Apple_Free partitions (#154479).
perl-PDL-2.4.1-10
-----------------
* Sat Apr 30 2005 Jose Pedro Oliveira <jpo at di.uminho.pt> - 2.4.1-10
- Bring up to date with current Fedora.Extras perl spec template. (#156482)
- disable SMP flags so it actually builds
perl-Text-Kakasi-1.05-13
------------------------
* Sat Apr 30 2005 Jose Pedro Oliveira <jpo at di.uminho.pt> - 1.05-13
- Added missing build requirement: kakasi-dict. (#156479)
- Bring up to date with current Fedora.Extras perl spec template.
perl-XML-NamespaceSupport-1.08-7
--------------------------------
redhat-artwork-0.122-8
----------------------
* Fri May 06 2005 John (J5) Palmieri <johnp(a)redhat.com> 0.122-8
- OpenOffice icon names change to not reflect versions anymore
Change symlinks to reflect this
* Thu May 05 2005 John (J5) Palmieri <johnp(a)redhat.com> 0.122-7
- Add patch that changes the ythickness of the bluecurve-menu-item
style in the bluecurve theme so the menu applet does not expand
the default panel size which causes launchers icons in the panel
to be scaled and become blured (Bug #146980)
sendmail-8.13.4-2
-----------------
* Fri May 06 2005 Thomas Woerner <twoerrner(a)redhat.com> 8.13.4-2
- using new certificates directory /etc/pki/tls/certs
setarch-1.7-3
-------------
* Mon May 09 2005 Bill Nottingham <notting(a)redhat.com> 1.7-3
- don't strip binary, fixes debuginfo generation
struts11-0:1.1-1jpp_6fc
-----------------------
* Mon May 09 2005 Gary Benson <gbenson(a)redhat.com> - 0:1.1-1jpp_6fc
- Work around a classloader issue (#157205).
system-config-nfs-1.3.10-1
--------------------------
* Mon May 09 2005 Nils Philippsen <nphilipp(a)redhat.com> 1.3.10
- pick up updated translations
system-config-services-0.8.25-1
-------------------------------
* Mon May 09 2005 Nils Philippsen <nphilipp(a)redhat.com> - 0.8.25
- pick up updated translations
system-config-users-1.2.38-1
----------------------------
* Mon May 09 2005 Nils Philippsen <nphilipp(a)redhat.com> - 1.2.38
- pick up updated translations
* Fri May 06 2005 Nils Philippsen <nphilipp(a)redhat.com>
- make desktop file rebuild consistently
tcp_wrappers-7.6-39
-------------------
* Fri May 06 2005 Thomas Woerner <twoerner(a)redhat.com> 7.6-39
- fixed sig patch (#141110). Thanks to Nikita Shulga for the patch
tetex-3.0-4
-----------
* Wed May 04 2005 Jindrich Novy <jnovy(a)redhat.com> 3.0-4
- add envlab package (#149212)
- fix usage of uninitialized variable in ttf2afm.c (#149212)
* Mon Apr 25 2005 Jindrich Novy <jnovy(a)redhat.com>
- add Requires: fonts-japanese >= 0.20050222-2 to xdvi
subpackage when Japanese is enabled
- switch from /usr/share/fonts/ja to /usr/share/fonts/japanese (#155880)
* Mon Apr 25 2005 MATSUURA Takanori <t-matsuu(a)estyle.ne.jp>
- use htt as input method for pxdvi
- update pxdvi to 22.84.9 j1.22
- fix pxdvi installation
- update to ptex-src-3.1.8.1, dvips-jpatch-p1.7, mensexk-2.6a
- delete japanese.map, now included in dvipsk-jpatch-p1.7
udev-057-5
----------
* Thu May 05 2005 Bill Nottingham <notting(a)redhat.com> - 057-5
- rebuild
up2date-4.4.18-1
----------------
* Mon May 09 2005 Adrian Likins <alikins(a)redhat.com> 4.4.18
- fix problem with missing import
xscreensaver-1:4.21-3
---------------------
* Mon May 09 2005 Ray Strode <rstrode(a)redhat.com> 1:4.21-3
- Use @libexecdir@/xscreensaver instead of @HACKDIR@ in
default configuration file so that the path gets expanded
fully (bug 156906).
7:19 a.m.
On Tue, May 10, 2005 at 08:12:35AM -0400, Build System wrote:
- on x86 and x86_64, use -msse and -msse2 to accomodate newer
compilers
Not that I want to revive the whole .i386.rpm vs. i686.rpm discussion,
but... how does gcc handle these flags? Does it build alternative
code for processors that do not have these features? Or does the binary
not run on anything below PII or compatible?
7:26 a.m.
On Tue, May 10, 2005 at 02:19:32PM +0200, Ralf Ertzinger wrote:
On Tue, May 10, 2005 at 08:12:35AM -0400, Build System wrote:
> - on x86 and x86_64, use -msse and -msse2 to accomodate newer compilers
Not that I want to revive the whole .i386.rpm vs. i686.rpm discussion,
but... how does gcc handle these flags? Does it build alternative
code for processors that do not have these features? Or does the binary
not run on anything below PII or compatible?
The way it is done in current rawhide's GIMP is definitely wrong.
%ifarch %ix86 x86_64
CFLAGS="%optflags -msse -msse2" CXXFLAGS="%optflags -msse -msse2" \
%endif
%configure \
a) on x86-64 this is unnecessary, all x86-64's have -msse2 by default
b) -msse2 implies -msse
c) if you use -msse2 in CFLAGS for all files, you can't run the latest
GIMP on e.g. Pentium2, or pre-x86_64 AMD chips.
-msse2 should be ONLY used on sources that have SSE/SSE2 stuff in it,
and GIMP should make sure that no routine from those sources will be
ever called on pre-SSE2 chips
Jakub
9:53 a.m.
On Tue, 2005-05-10 at 08:26 -0400, Jakub Jelinek wrote:
On Tue, May 10, 2005 at 02:19:32PM +0200, Ralf Ertzinger wrote:
> On Tue, May 10, 2005 at 08:12:35AM -0400, Build System wrote:
>
> > - on x86 and x86_64, use -msse and -msse2 to accomodate newer compilers
>
> Not that I want to revive the whole .i386.rpm vs. i686.rpm discussion,
> but... how does gcc handle these flags? Does it build alternative
> code for processors that do not have these features? Or does the binary
> not run on anything below PII or compatible?
The way it is done in current rawhide's GIMP is definitely wrong.
%ifarch %ix86 x86_64
CFLAGS="%optflags -msse -msse2" CXXFLAGS="%optflags -msse -msse2" \
%endif
%configure \
a) on x86-64 this is unnecessary, all x86-64's have -msse2 by default
b) -msse2 implies -msse
ACK
c) if you use -msse2 in CFLAGS for all files, you can't run the
latest
GIMP on e.g. Pentium2, or pre-x86_64 AMD chips.
-msse2 should be ONLY used on sources that have SSE/SSE2 stuff in it,
and GIMP should make sure that no routine from those sources will be
ever called on pre-SSE2 chips
According to the gcc man page, you need to use "-mfpmath=sse" if you
want to get SSE/SSE2 instructions generated from FP code. As I
understand the manpage, "-msse" and the like only _enable_ its use in
inline assembler code. GIMP does check whether MMX/SSE etc. are
available when starting and only uses the optimized code paths if
possible.
Nils
--
Nils Philippsen / Red Hat / nphilipp(a)redhat.com
"They that can give up essential liberty to obtain a little temporary
safety deserve neither liberty nor safety." -- B. Franklin, 1759
PGP fingerprint: C4A8 9474 5C4C ADE3 2B8F 656D 47D8 9B65 6951 3011
10:02 a.m.
On Tue, May 10, 2005 at 04:53:55PM +0200, Nils Philippsen wrote:
> c) if you use -msse2 in CFLAGS for all files, you can't run
the latest
> GIMP on e.g. Pentium2, or pre-x86_64 AMD chips.
> -msse2 should be ONLY used on sources that have SSE/SSE2 stuff in it,
> and GIMP should make sure that no routine from those sources will be
> ever called on pre-SSE2 chips
According to the gcc man page, you need to use "-mfpmath=sse" if you
want to get SSE/SSE2 instructions generated from FP code. As I
No. -msse resp. -msse2 says that SSE resp. SSE2 instructions are available
and can be used. So, with -msse2 you can use SSE2 builtins, you can
use those regs in __asm statement clobbers/regs and GCC if it decides
it is worthwhile can use the SSE2 registers/instructions (say with
autovectorization, or with high register preasure to move data around,
etc.).
-mfpmath=sse uses the SSE2 unit for float/double arithmetics.
If the only place that you want to enable -msse2 for are the clobber
lists of __asm statements, then perhaps best would be to conditionalize
them:
__asm ("something" : ... : ... :
#ifdef __SSE2__
"xmm0", "xmm1"
#endif
);
etc., because without -msse2 (resp. -msse (__SSE__ macro)), the registers
are not known to the compiler, so there is no point to tell the compiler
about them.
Jakub
10:08 a.m.
On Tue, 2005-05-10 at 11:02 -0400, Jakub Jelinek wrote:
On Tue, May 10, 2005 at 04:53:55PM +0200, Nils Philippsen wrote:
> > c) if you use -msse2 in CFLAGS for all files, you can't run the latest
> > GIMP on e.g. Pentium2, or pre-x86_64 AMD chips.
> > -msse2 should be ONLY used on sources that have SSE/SSE2 stuff in it,
> > and GIMP should make sure that no routine from those sources will be
> > ever called on pre-SSE2 chips
>
> According to the gcc man page, you need to use "-mfpmath=sse" if you
> want to get SSE/SSE2 instructions generated from FP code. As I
No. -msse resp. -msse2 says that SSE resp. SSE2 instructions are available
and can be used. So, with -msse2 you can use SSE2 builtins, you can
use those regs in __asm statement clobbers/regs and GCC if it decides
it is worthwhile can use the SSE2 registers/instructions (say with
autovectorization, or with high register preasure to move data around,
etc.).
-mfpmath=sse uses the SSE2 unit for float/double arithmetics.
If the only place that you want to enable -msse2 for are the clobber
lists of __asm statements, then perhaps best would be to conditionalize
them:
__asm ("something" : ... : ... :
#ifdef __SSE2__
"xmm0", "xmm1"
#endif
);
etc., because without -msse2 (resp. -msse (__SSE__ macro)), the registers
are not known to the compiler, so there is no point to tell the compiler
about them.
Ahh. If I understand you correctly, given that the various asm optimized
functions are in separate source files, it should be ensured that
-mmmx/sse/sse2/3dnow are used only for the relevant source files?
Nils
--
Nils Philippsen / Red Hat / nphilipp(a)redhat.com
"They that can give up essential liberty to obtain a little temporary
safety deserve neither liberty nor safety." -- B. Franklin, 1759
PGP fingerprint: C4A8 9474 5C4C ADE3 2B8F 656D 47D8 9B65 6951 3011
10:11 a.m.
On Tue, May 10, 2005 at 05:08:34PM +0200, Nils Philippsen wrote:
> If the only place that you want to enable -msse2 for are the
clobber
> lists of __asm statements, then perhaps best would be to conditionalize
> them:
> __asm ("something" : ... : ... :
> #ifdef __SSE2__
> "xmm0", "xmm1"
> #endif
> );
> etc., because without -msse2 (resp. -msse (__SSE__ macro)), the registers
> are not known to the compiler, so there is no point to tell the compiler
> about them.
Ahh. If I understand you correctly, given that the various asm optimized
functions are in separate source files, it should be ensured that
-mmmx/sse/sse2/3dnow are used only for the relevant source files?
Yes (and assuming they don't also contain code that is called
unconditionally).
Or don't pass -mmmx/-msse/-msse2/-m3dnow etc. at all, and hide all the
SSE2 etc. stuff in __asm and conditionally don't expose the clobbers
to GCC.
Jakub
10:17 a.m.
On Tue, 2005-05-10 at 11:11 -0400, Jakub Jelinek wrote:
On Tue, May 10, 2005 at 05:08:34PM +0200, Nils Philippsen wrote:
> > If the only place that you want to enable -msse2 for are the clobber
> > lists of __asm statements, then perhaps best would be to conditionalize
> > them:
> > __asm ("something" : ... : ... :
> > #ifdef __SSE2__
> > "xmm0", "xmm1"
> > #endif
> > );
> > etc., because without -msse2 (resp. -msse (__SSE__ macro)), the registers
> > are not known to the compiler, so there is no point to tell the compiler
> > about them.
>
> Ahh. If I understand you correctly, given that the various asm optimized
> functions are in separate source files, it should be ensured that
> -mmmx/sse/sse2/3dnow are used only for the relevant source files?
Yes (and assuming they don't also contain code that is called
unconditionally).
I expect that to be the case (the code is in files called gimp-
composite-{generic,mmx,altivec,sse,sse2,...}).
Or don't pass -mmmx/-msse/-msse2/-m3dnow etc. at all, and hide
all the
SSE2 etc. stuff in __asm and conditionally don't expose the clobbers
to GCC.
The idea is to use the most optimized code path possible from one
binary. Besides, I don't really think we want to have
i686,pentiumX,athlon variants of GIMP packages ;-).
Nils
--
Nils Philippsen / Red Hat / nphilipp(a)redhat.com
"They that can give up essential liberty to obtain a little temporary
safety deserve neither liberty nor safety." -- B. Franklin, 1759
PGP fingerprint: C4A8 9474 5C4C ADE3 2B8F 656D 47D8 9B65 6951 3011
10:30 a.m.
On Tue, 2005-05-10 at 08:26 -0400, Jakub Jelinek wrote:
a) on x86-64 this is unnecessary, all x86-64's have -msse2 by
default
b) -msse2 implies -msse
c) if you use -msse2 in CFLAGS for all files, you can't run the latest
GIMP on e.g. Pentium2, or pre-x86_64 AMD chips.
-msse2 should be ONLY used on sources that have SSE/SSE2 stuff in it,
and GIMP should make sure that no routine from those sources will be
ever called on pre-SSE2 chips
This sounds like a big pain ... shouldn't there be a way to say
"use sse2/sse only for builtins" ?
Without that, there is no way to use builtins in single functions; you
need to have one file for sse2, one for sse, one for mmx, and do the
detection of the current processor somewhere else entirely.
Owen
5:20 a.m.
On Tue, 2005-05-10 at 11:30 -0400, Owen Taylor wrote:
On Tue, 2005-05-10 at 08:26 -0400, Jakub Jelinek wrote:
> a) on x86-64 this is unnecessary, all x86-64's have -msse2 by default
> b) -msse2 implies -msse
> c) if you use -msse2 in CFLAGS for all files, you can't run the latest
> GIMP on e.g. Pentium2, or pre-x86_64 AMD chips.
> -msse2 should be ONLY used on sources that have SSE/SSE2 stuff in it,
> and GIMP should make sure that no routine from those sources will be
> ever called on pre-SSE2 chips
This sounds like a big pain ... shouldn't there be a way to say
"use sse2/sse only for builtins" ?
Without that, there is no way to use builtins in single functions; you
need to have one file for sse2, one for sse, one for mmx, and do the
detection of the current processor somewhere else entirely.
I concur, with gcc <= 4.0.0-2 it was possible to use MMX/SSE calls in
inline assembly without having to use -mmmx/-msse/-msse2. In order to
build the gimp so that it uses MMX/SSE where available, I had to resort
to serious autofoo munging because it doesn't allow for setting
per-object compiler flags.
Nils
--
Nils Philippsen / Red Hat / nphilipp(a)redhat.com
"They that can give up essential liberty to obtain a little temporary
safety deserve neither liberty nor safety." -- B. Franklin, 1759
PGP fingerprint: C4A8 9474 5C4C ADE3 2B8F 656D 47D8 9B65 6951 3011
9:15 a.m.
Nils Philippsen wrote:
On Tue, 2005-05-10 at 11:30 -0400, Owen Taylor wrote:
>On Tue, 2005-05-10 at 08:26 -0400, Jakub Jelinek wrote:
>
>
>
>>a) on x86-64 this is unnecessary, all x86-64's have -msse2 by default
>>b) -msse2 implies -msse
>>c) if you use -msse2 in CFLAGS for all files, you can't run the latest
>> GIMP on e.g. Pentium2, or pre-x86_64 AMD chips.
>> -msse2 should be ONLY used on sources that have SSE/SSE2 stuff in it,
>> and GIMP should make sure that no routine from those sources will be
>> ever called on pre-SSE2 chips
>>
>>
>This sounds like a big pain ... shouldn't there be a way to say
>"use sse2/sse only for builtins" ?
>
>Without that, there is no way to use builtins in single functions; you
>need to have one file for sse2, one for sse, one for mmx, and do the
>detection of the current processor somewhere else entirely.
>
>
I concur, with gcc <= 4.0.0-2 it was possible to use MMX/SSE calls in
inline assembly without having to use -mmmx/-msse/-msse2. In order to
build the gimp so that it uses MMX/SSE where available, I had to resort
to serious autofoo munging because it doesn't allow for setting
per-object compiler flags.
Nils
why aren't sse/mmx/sse2 enabled by default on x86_64 ? all those chips
supports it.
10:47 a.m.
On Fri, 2005-05-13 at 16:15 +0200, dragoran wrote:
Nils Philippsen wrote:
>On Tue, 2005-05-10 at 11:30 -0400, Owen Taylor wrote:
>
>
>>On Tue, 2005-05-10 at 08:26 -0400, Jakub Jelinek wrote:
>>
>>
>>
>>>a) on x86-64 this is unnecessary, all x86-64's have -msse2 by default
>>>b) -msse2 implies -msse
>>>c) if you use -msse2 in CFLAGS for all files, you can't run the latest
>>> GIMP on e.g. Pentium2, or pre-x86_64 AMD chips.
>>> -msse2 should be ONLY used on sources that have SSE/SSE2 stuff in it,
>>> and GIMP should make sure that no routine from those sources will be
>>> ever called on pre-SSE2 chips
>>>
>>>
>>This sounds like a big pain ... shouldn't there be a way to say
>>"use sse2/sse only for builtins" ?
>>
>>Without that, there is no way to use builtins in single functions; you
>>need to have one file for sse2, one for sse, one for mmx, and do the
>>detection of the current processor somewhere else entirely.
>>
>>
>
>I concur, with gcc <= 4.0.0-2 it was possible to use MMX/SSE calls in
>inline assembly without having to use -mmmx/-msse/-msse2. In order to
>build the gimp so that it uses MMX/SSE where available, I had to resort
>to serious autofoo munging because it doesn't allow for setting
>per-object compiler flags.
>
>Nils
>
>
why aren't sse/mmx/sse2 enabled by default on x86_64 ? all those chips
supports it.
I got some things mixed up in my original assessment of the problem:
- I thought that gcc only allowed MMX/SSE/SSE2-specific inline assembler
when using -mmmx/-msse/-msse2. Actually the real problem was that gcc
only allows MMX/SSE registers in clobber lists when using
-mmmx/-msse/-msse2 and complains about them being present if these
options aren't used. Masking them out with #ifdef __MMX__/__SSE__ ...
#endif is the real solution.
- I thought the problem would show on x86_64 as well while it doesn't
due to the aforementioned reasons.
gimp-2.2.7-4 contains the real fix.
Nils
--
Nils Philippsen / Red Hat / nphilipp(a)redhat.com
"They that can give up essential liberty to obtain a little temporary
safety deserve neither liberty nor safety." -- B. Franklin, 1759
PGP fingerprint: C4A8 9474 5C4C ADE3 2B8F 656D 47D8 9B65 6951 3011
6926
days inactive
6931
days old
11 comments
6 participants
participants (6)
-
Build System -
dragoran -
Jakub Jelinek -
Nils Philippsen -
Owen Taylor -
Ralf Ertzinger